采集与预览
快速确认摄像头链路可用
推荐组合：Camera + DisplayUDP
核心函数：snapshot, display, is_valid
1) 快速开始
1. 放置库文件
将 visiong.py 与 _visiong.so 放入当前 Python 环境可导入路径（如 site-packages）。
2. 检查导入
运行 python3 -c "import visiong; print(visiong.__version__)"，确认无报错。
3. 先跑最小样例
先验证 Camera + DisplayUDP，确认图像流正常，再叠加 IVE/NPU/GUI 模块。
4. 识别模块开关
某些类可能不存在（如 IVE/NPU/GUI），通常是构建时对应模块未启用。
2) 学习路径
- 阶段 A：Camera + ImageBuffer（采集、格式转换、基础绘制）
- 阶段 B：显示与传输（DisplayFB/UDP/RTSP/HTTP）
- 阶段 C：IVE 算子（阈值、边缘、形态学、模板相关）
- 阶段 D：NPU 任务（检测、OCR、跟踪、底层张量 IO）
- 阶段 E：系统能力（PinMux、GPIO、ADC、NPU 时钟）
- 阶段 F：GUI 交互（触控窗口、控件、画布绘制）
2.1 学习者排错顺序
- 先保证采集链路：Camera.snapshot() 得到 valid 图像，再进入算法/推流步骤。
- 每次只叠加一个能力模块：先显示，再加 IVE，再加 NPU，最后接 GUI 或系统控制。
- 遇到结果异常时优先检查输入格式（rgb/bgr/gray/yuv）和宽高是否符合模型/算子要求。
- 实时循环里先做最小逻辑，保证帧率，再逐步加绘制、日志、异常处理。
- 资源类对象统一在 finally 里释放（stop/release/close），避免句柄泄漏。
- 改参数遵循“一次只改一个维度”（阈值、分辨率、码率、NMS）便于定位效果变化。
2.2 实时循环通用骨架
import visiong
cam = visiong.Camera(640, 360, 'bgr')
out = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
# 1) 可选预处理
# 2) 可选算法推理
# 3) 可选结果绘制
out.display(frame)
finally:
out.release()
cam.release()2.3 性能与稳定性清单
- 先降低分辨率再提 FPS：一般先从 320x240 或 640x360 起步。
- 算法输入格式和模型格式保持一致（rgb/bgr/gray），避免隐式转换损耗。
- 推流与推理分离思考：先测纯推流帧率，再叠加推理。
- NPU 推理先记录耗时与置信度分布，再调整 box/nms 阈值。
- 长期运行时定期观察内存与句柄，确保 close/release/stop 都被调用。
- 线上脚本保留最低限度日志，并为关键模块加 try/finally 保护。
3) 任务快速参考
按“要做什么”而不是“类名”组织。先选任务，再进入对应 API 和示例。
图像预处理
在推理前做阈值/边缘/颜色空间处理
推荐组合：IVE + ImageBuffer
核心函数：threshold, canny, sobel, csc
目标检测
做目标框与类别可视化
推荐组合：NPU + ImageBuffer
核心函数：infer, draw_rectangle, draw_string
OCR 识别
识别图中文字并回绘
推荐组合：PPOCR + ImageBuffer
核心函数：infer, rect, text
目标跟踪
初始框后持续更新轨迹
推荐组合：NanoTrack + Camera
核心函数：init, track
网络推流
浏览器/播放器实时查看画面
推荐组合：DisplayHTTP/RTSP/FLV
核心函数：display, stop
录像保存
编码并落盘
推荐组合：VencRecorder + ImageBuffer
核心函数：write, close, save_venc_h264/h265
系统 IO
GPIO/ADC/Pin 复用控制
推荐组合：PinMux + GpioLineConfig
核心函数：gpio_request_line, gpio_set_value, read_adc
GUI 交互
构建菜单、按钮和画布叠加
推荐组合：GUI + Touch + Canvas
核心函数：begin_frame, begin_window, button, end_frame
4) 脚本模板集（可直接起步）
覆盖采集、预处理、推理、推流四类最常见工程骨架。复制后替换参数即可。
模板 A：采集 + 显示
最小可跑链路，先用它验证运行环境。
import visiong
cam = visiong.Camera(640, 360, 'bgr')
out = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if frame.is_valid():
out.display(frame)
finally:
out.release(); cam.release()模板 B：预处理 + 显示
在实时循环里接入 IVE 算子。
import visiong
cam = visiong.Camera(640, 360, 'gray')
ive = visiong.IVE()
out = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
src = cam.snapshot()
if not src.is_valid():
continue
edge = ive.canny(src, high_thresh=120, low_thresh=40)
out.display(edge)
finally:
out.release(); cam.release()模板 C：推理 + 绘制 + 显示
通用 NPU 检测骨架。
import visiong
cam = visiong.Camera(640, 360, 'rgb')
npu = visiong.NPU('yolov5', '/path/to/model.rknn', '/path/to/labels.txt')
udp = visiong.DisplayUDP()
try:
cam.skip()
while True:
img = cam.snapshot()
if not img.is_valid():
continue
for det in npu.infer(img, model_format='rgb'):
x, y, w, h = det.box
img.draw_rectangle(x, y, w, h, color=(0, 255, 0), thickness=2)
udp.display(img)
finally:
udp.release(); cam.release()模板 D：推流服务
将采集流发布为 HTTP/RTSP。
import visiong
cam = visiong.Camera(640, 360, 'yuv')
stream = visiong.DisplayRTSP(port=554, path='/live/0', codec='h265', fps=25)
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if frame.is_valid():
stream.display(frame)
finally:
stream.stop(); cam.release()5) 大量实战示例
手写案例按难度分层组织，从“最小跑通”到“组合实战”。建议按顺序练习。
当前示例数量：80（手写） + 453（逐函数自动示例）
模块函数覆盖
7/7
类方法覆盖
95/308
类覆盖
17/43
查看覆盖薄弱点
未覆盖模块函数：无
覆盖较低的类（手写案例）：
Blob：0/1Circle：0/1Detection：0/1Line：0/1MotionVector：0/1OCRResult：0/1PinAltFunction：0/1PinId：0/1PinMuxRegisterInfo：0/1QRCode：0/1TouchPoint：0/1ImageBuffer：4/39
1. L1 入门基础：先跑通采集与显示 初级
10 个案例
掌握 Camera、DisplayUDP、ImageBuffer 的最小闭环，先看到稳定视频流。
案例 1-1：最小实时预览
确认摄像头和 UDP 显示链路可用。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(10)
while True:
img = cam.snapshot()
if img.is_valid():
udp.display(img)
finally:
udp.release()
cam.release()案例 1-2：抓一帧并保存 JPG
学习 snapshot + save 的最小离线流程。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
try:
cam.skip(5)
img = cam.snapshot()
if img.is_valid():
img.save('frame.jpg', quality=90)
print('saved frame.jpg')
finally:
cam.release()案例 1-3：灰度转换并显示
把彩色帧转灰度，理解 to_grayscale。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
gray = frame.to_grayscale()
udp.display(gray)
finally:
udp.release()
cam.release()案例 1-4：缩放 + letterbox
统一输出尺寸，避免直接拉伸变形。
import visiong
cam = visiong.Camera(1280, 720, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
src = cam.snapshot()
if not src.is_valid():
continue
fit = src.letterbox(640, 360, color=(0, 0, 0))
udp.display(fit)
finally:
udp.release()
cam.release()案例 1-5：基础绘制（线/框/圆/文字）
熟悉 draw_line、draw_rectangle、draw_circle、draw_string。
import visiong
img = visiong.ImageBuffer.create(640, 360, 'bgr888', (18, 24, 30))
img.draw_line(20, 20, 220, 20, color=(0, 255, 0), thickness=2)
img.draw_rectangle(40, 60, 180, 100, color=(255, 255, 0), thickness=2)
img.draw_circle(380, 180, 70, color=(255, 120, 0), thickness=3)
img.draw_string(30, 300, 'VisionG Drawing Demo', color=(255, 255, 255), scale=1.0, thickness=2)
img.save('draw_demo.jpg', quality=90)案例 1-6：NumPy 回路（拷贝）
to_numpy + from_numpy 的安全互通。
import numpy as np
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
img = cam.snapshot()
if not img.is_valid():
continue
arr = img.to_numpy(copy=True)
arr[:, :, 1] = np.clip(arr[:, :, 1] + 25, 0, 255)
out = visiong.ImageBuffer.from_numpy(arr, format='bgr888', copy=True)
udp.display(out)
finally:
udp.release()
cam.release()案例 1-7：NumPy 零拷贝导入
from_numpy_zero_copy 的最小示例（需 uint8、C 连续、偶数宽高）。
import numpy as np
import visiong
arr = np.zeros((240, 320, 3), dtype=np.uint8)
arr[:, :, 2] = 180
img = visiong.ImageBuffer.from_numpy_zero_copy(arr, format='bgr888')
img.save('zero_copy.jpg', quality=90)案例 1-8：裁剪 + 旋转 + 翻转
几何变换链路的常见写法。
import visiong
src = visiong.ImageBuffer.load('/path/to/input.jpg')
roi = src.crop(120, 60, 320, 220)
rot = roi.rotate(90)
out = rot.flip(horizontal=True, vertical=False)
out.save('geom_out.jpg', quality=90)案例 1-9：实时调亮度/对比度/饱和度
演示 Camera 参数调节对画面的影响。
import time
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
cam.set_brightness(150)
cam.set_contrast(140)
cam.set_saturation(140)
t0 = time.time()
while time.time() - t0 < 15:
img = cam.snapshot()
if img.is_valid():
udp.display(img)
finally:
udp.release()
cam.release()案例 1-10：自动对焦与手动对焦
set_focus_mode / set_manual_focus / get_focus_position。
import time
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
cam.set_focus_mode('continuous')
t0 = time.time()
while time.time() - t0 < 5:
img = cam.snapshot()
if img.is_valid():
img.draw_string(10, 10, 'AF', color=(0, 255, 0), scale=1.0, thickness=2)
udp.display(img)
cam.set_manual_focus(512)
print('focus position =', cam.get_focus_position())
finally:
udp.release()
cam.release()
2. L2 图像分析：从阈值到几何特征 初级→中级
10 个案例
使用 ImageBuffer 检测接口，完成简单视觉任务。
案例 2-1：二值化（Otsu）
binarize 的最小实用流程。
import visiong
img = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
bw = img.binarize(method='otsu')
bw.save('otsu.jpg', quality=90)案例 2-2：颜色块检测 find_blobs
在实时画面中定位阈值颜色块。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
# 深蓝
thresholds = [(0, 131, 161, 255, 0, 255)]
try:
cam.skip(8)
while True:
img = cam.snapshot()
if not img.is_valid():
continue
for b in img.find_blobs(thresholds, area_threshold=150, pixels_threshold=150):
img.draw_rectangle(b.x, b.y, b.w, b.h, color=(0, 255, 0), thickness=2)
udp.display(img)
finally:
udp.release()
cam.release()案例 2-3：直线检测 find_lines
霍夫线检测并绘制。
import visiong
cam = visiong.Camera(640, 360, format='gray')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
img = cam.snapshot()
if not img.is_valid():
continue
vis = img.to_format('bgr888')
lines = img.find_lines(0, 0, img.width, img.height, threshold=70)
for ln in lines:
vis.draw_line(ln.x1, ln.y1, ln.x2, ln.y2, color=(0, 255, 0), thickness=2)
udp.display(vis)
finally:
udp.release()
cam.release()案例 2-4：圆检测 find_circles
检测圆形目标并标注圆心。
import visiong
img = visiong.ImageBuffer.load('/path/to/circle_input.jpg').to_grayscale()
vis = img.to_format('bgr888')
circles = img.find_circles(0, 0, img.width, img.height, threshold=45, r_min=8, r_max=80)
for c in circles:
vis.draw_circle(c.cx, c.cy, c.r, color=(255, 220, 0), thickness=2)
vis.draw_cross(c.cx, c.cy, color=(255, 220, 0), size=8, thickness=2)
vis.save('circles.jpg', quality=90)案例 2-5：透视矫正 warp_perspective
把斜拍区域拉正到固定尺寸。
import visiong
src = visiong.ImageBuffer.load('/path/to/doc.jpg')
quad = [(50, 60), (540, 40), (580, 330), (40, 340)]
flat = src.warp_perspective(quad, 480, 320)
flat.save('doc_flat.jpg', quality=90)案例 2-6：方块检测 find_squares
阈值+轮廓组合的稳定写法。
import visiong
img = visiong.ImageBuffer.load('/path/to/square_input.jpg').to_grayscale()
squares = img.find_squares((0, 0, img.width, img.height), threshold_val=120, min_area=400)
print('square count =', len(squares))案例 2-7：多边形检测 find_polygons
统计多边形候选目标数量。
import visiong
img = visiong.ImageBuffer.load('/path/to/polygon_input.jpg').to_grayscale()
items = img.find_polygons(0, 0, img.width, img.height, min_area=100, max_area=100000, min_sides=3, max_sides=10, accuracy='normal')
print('polygon count =', len(items))案例 2-8：二维码定位与连线绘制
读取 payload，同时用四角点连线。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
while True:
img = cam.snapshot()
if not img.is_valid():
continue
for qr in img.find_qrcodes():
c = qr.corners
img.draw_line(c[0][0], c[0][1], c[1][0], c[1][1], color=(0, 255, 255), thickness=2)
img.draw_line(c[1][0], c[1][1], c[2][0], c[2][1], color=(0, 255, 255), thickness=2)
img.draw_line(c[2][0], c[2][1], c[3][0], c[3][1], color=(0, 255, 255), thickness=2)
img.draw_line(c[3][0], c[3][1], c[0][0], c[0][1], color=(0, 255, 255), thickness=2)
txt = qr.payload.decode('utf-8', 'ignore')
img.draw_string(10, 10, txt, color=(0, 255, 255), scale=1.0, thickness=2)
udp.display(img)
finally:
udp.release()
cam.release()案例 2-9：限定 ROI 的直线检测
只在中间区域做检测，降低噪声干扰。
import visiong
cam = visiong.Camera(640, 360, format='gray')
udp = visiong.DisplayUDP()
roi = (120, 60, 400, 240)
try:
cam.skip(8)
while True:
g = cam.snapshot()
if not g.is_valid():
continue
vis = g.to_format('bgr888')
vis.draw_rectangle(*roi, color=(0, 255, 255), thickness=1)
for ln in g.find_lines(*roi, threshold=65):
vis.draw_line(ln.x1, ln.y1, ln.x2, ln.y2, color=(0, 255, 0), thickness=2)
udp.display(vis)
finally:
udp.release()
cam.release()案例 2-10：离线批量预处理
load -> resize -> save，常用于数据准备。
import visiong
inputs = ['/path/to/a.jpg', '/path/to/b.jpg', '/path/to/c.jpg']
for i, p in enumerate(inputs, 1):
img = visiong.ImageBuffer.load(p)
out = img.resize(320, 320)
out.save(f'out_{i:02d}.jpg', quality=92)
3. L3 IVE 专题：硬件算子组合 中级
10 个案例
熟悉 IVE 常用算子组合，搭建可复用预处理链路。
案例 3-1：5x5 卷积滤波
IVE.filter 处理锐化/平滑核。
import visiong
ive = visiong.IVE()
src = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
sharpen = [0, -1, 0, -1, 5, -1, 0, -1, 0]
dst = ive.filter(src, sharpen)
dst.save('ive_filter.jpg', quality=90)案例 3-2：Sobel + 16bit 转 8bit
sobel 输出再 cast_16bit_to_8bit 可视化。
import visiong
ive = visiong.IVE()
gray = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
gx, gy = ive.sobel(gray, visiong.SobelOutCtrl.BOTH, visiong.ImageTypeIVE.S16C1)
gx8 = ive.cast_16bit_to_8bit(gx, visiong.Cast16to8Mode.S16_TO_U8_ABS)
gy8 = ive.cast_16bit_to_8bit(gy, visiong.Cast16to8Mode.S16_TO_U8_ABS)
gx8.save('sobel_x.jpg', quality=90)
gy8.save('sobel_y.jpg', quality=90)案例 3-3：Canny 边缘
ive.canny 直接输出边缘图。
import visiong
ive = visiong.IVE()
gray = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
edge = ive.canny(gray, high_thresh=120, low_thresh=40)
edge.save('canny.jpg', quality=90)案例 3-4：膨胀 + 腐蚀
形态学后处理去噪。
import visiong
ive = visiong.IVE()
gray = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
bw = gray.binarize(method='otsu')
clean = ive.erode(bw, kernel_size=3)
clean = ive.dilate(clean, kernel_size=3)
clean.save('morph.jpg', quality=90)案例 3-5：图像相加/相减/逻辑与
ive.add / ive.sub / ive.logic_op 常见组合。
import visiong
ive = visiong.IVE()
a = visiong.ImageBuffer.load('/path/to/a.jpg').to_grayscale()
b = visiong.ImageBuffer.load('/path/to/b.jpg').to_grayscale()
mix = ive.add(a, b)
diff = ive.sub(a, b, mode=visiong.SubMode.ABS)
mask = ive.logic_op(a, b, op=visiong.LogicOp.AND)
mix.save('ive_add.jpg', quality=90)
diff.save('ive_sub.jpg', quality=90)
mask.save('ive_and.jpg', quality=90)案例 3-6：角点 + 光流
st_corner 找特征点，再 lk_optical_flow 跟踪。
import visiong
ive = visiong.IVE()
prev = visiong.ImageBuffer.load('/path/to/frame_prev.jpg').to_grayscale()
curr = visiong.ImageBuffer.load('/path/to/frame_curr.jpg').to_grayscale()
pts = ive.st_corner(prev, max_corners=120, min_dist=8, quality_level=20)
mvs = ive.lk_optical_flow(prev, curr, pts)
print('track points =', len(mvs))案例 3-7：直方图与均衡化
ive.hist + ive.equalize_hist。
import visiong
ive = visiong.IVE()
gray = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
hist = ive.hist(gray)
print('hist bins =', len(hist))
eq = ive.equalize_hist(gray)
eq.save('equalized.jpg', quality=90)案例 3-8：LUT 映射增强
ive.map 对灰度图进行查表处理。
import visiong
ive = visiong.IVE()
src = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
lut = [min(255, int(i * 1.2)) for i in range(256)]
dst = ive.map(src, lut)
dst.save('lut_map.jpg', quality=90)案例 3-9：模板相似度 ncc
比较两张图的归一化相关系数。
import visiong
ive = visiong.IVE()
a = visiong.ImageBuffer.load('/path/to/template.jpg').to_grayscale()
b = visiong.ImageBuffer.load('/path/to/search.jpg').to_grayscale()
score = ive.ncc(a, b)
print('ncc score =', score)案例 3-10：多尺度金字塔
ive.create_pyramid 生成多尺度图像序列。
import visiong
ive = visiong.IVE()
src = visiong.ImageBuffer.load('/path/to/input.jpg').to_grayscale()
levels = ive.create_pyramid(src, levels=4)
for i, lv in enumerate(levels):
lv.save(f'pyr_{i}.jpg', quality=90)
4. L4 传输与录制：从本地显示到网络推流 中级
10 个案例
掌握 UDP/FB/RTSP/HTTP 与编码录制能力。
案例 4-1：UDP 指定 IP 与质量
DisplayUDP 初始化参数的常用写法。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
udp = visiong.DisplayUDP(udp_ip='192.168.1.100', udp_port=8000, jpeg_quality=80)
try:
cam.skip(8)
while True:
img = cam.snapshot()
if img.is_valid():
udp.display(img)
finally:
udp.release()
cam.release()案例 4-2：Framebuffer 显示
本地屏显示，包含 roi 参数。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
fb = visiong.DisplayFB()
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if frame.is_valid():
fb.display(frame, (0, 0, frame.width, frame.height))
finally:
fb.release()
cam.release()案例 4-3：RTSP 推流
DisplayRTSP 持续推流，退出前 stop。
import visiong
cam = visiong.Camera(640, 360, format='yuv')
rtsp = visiong.DisplayRTSP(port=554, path='/live/0', codec='h265', quality=72, fps=25)
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if frame.is_valid():
rtsp.display(frame)
finally:
rtsp.stop()
cam.release()案例 4-4：HTTP MJPEG
浏览器访问设备 IP:8080 查看实时画面。
import visiong
cam = visiong.Camera(640, 360, format='bgr')
http = visiong.DisplayHTTP(port=8080, quality=75, mode='jpg')
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if frame.is_valid():
http.display(frame)
finally:
http.stop()
cam.release()案例 4-5：HTTP-FLV
DisplayHTTPFLV 发布 /live.flv。
import visiong
cam = visiong.Camera(640, 360, format='yuv')
flv = visiong.DisplayHTTPFLV(port=8080, path='/live.flv', codec='h264', fps=25, quality=70)
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if frame.is_valid():
flv.display(frame)
finally:
flv.stop()
cam.release()案例 4-6：VencRecorder 录 MP4
录制固定时长，结束 close。
import visiong
cam = visiong.Camera(640, 360, format='yuv')
rec = visiong.VencRecorder('demo.mp4', codec='h264', container='mp4', quality=72, fps=25)
try:
cam.skip(8)
for _ in range(300):
frame = cam.snapshot()
if frame.is_valid():
rec.write(frame)
finally:
rec.close()
cam.release()案例 4-7：单帧硬编 JPG
save_venc_jpg 在不跑循环时快速验证编码。
import visiong
img = visiong.ImageBuffer.load('/path/to/input.jpg')
img.save_venc_jpg('out.jpg', quality=80)案例 4-8：单帧硬编 H264/H265
save_venc_h264 与 save_venc_h265 对比。
import visiong
img = visiong.ImageBuffer.load('/path/to/input.jpg').to_format('yuv420sp')
img.save_venc_h264('out.h264', quality=75, rc_mode='cbr', fps=25, append=False)
img.save_venc_h265('out.h265', quality=75, rc_mode='cbr', fps=25, append=False)案例 4-9：运行中动态调推流参数
set_fps / set_quality 用于在线调优。
import time
import visiong
cam = visiong.Camera(640, 360, format='yuv')
rtsp = visiong.DisplayRTSP(port=554, path='/live/0', codec='h264', quality=75, fps=30)
try:
cam.skip(8)
switched = False
t0 = time.time()
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
if (not switched) and time.time() - t0 > 10:
rtsp.set_fps(20)
rtsp.set_quality(65)
switched = True
rtsp.display(frame)
finally:
rtsp.stop()
cam.release()案例 4-10：按路径关闭录制器
close_venc_recorder 在异常恢复时很实用。
import visiong
visiong.close_venc_recorder('demo.mp4')
5. L5 NPU 实战：检测、识别、跟踪 中级→高级
10 个案例
从高层 NPU 到 LowLevelNPU，覆盖常见模型流程。
案例 5-1：YOLOv5 检测并绘框
NPU.infer 返回 Detection 列表。
import visiong
MODEL = '/path/to/yolov5s.rknn'
LABELS = '/path/to/coco_80_labels_list.txt'
cam = visiong.Camera(640, 360, format='rgb')
udp = visiong.DisplayUDP()
det = visiong.NPU('yolov5', MODEL, LABELS, box=0.25, nms=0.45)
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
for r in det.infer(frame, model_format='rgb'):
x, y, w, h = r.box
out.draw_rectangle(x, y, w, h, color=(0, 255, 0), thickness=2)
out.draw_string(x, max(0, y - 20), f'{r.label} {r.score:.2f}', color=(0, 255, 0), scale=0.9, thickness=2)
udp.display(out)
finally:
udp.release()
cam.release()案例 5-2：FaceNet 特征比对
get_face_feature + get_feature_distance。
import visiong
net = visiong.NPU('facenet', '/path/to/facenet.rknn')
a = visiong.ImageBuffer.load('/path/to/face_a.jpg').to_format('rgb888')
b = visiong.ImageBuffer.load('/path/to/face_b.jpg').to_format('rgb888')
fa = net.get_face_feature(a)
fb = net.get_face_feature(b)
dist = visiong.NPU.get_feature_distance(fa, fb)
print('distance =', dist)案例 5-3：LPRNet 车牌识别
NPU.recognize_plate 最小调用。
import visiong
lpr = visiong.NPU('lprnet', '/path/to/lprnet.rknn')
img = visiong.ImageBuffer.load('/path/to/plate.jpg').to_format('rgb888')
text = lpr.recognize_plate(img)
print('plate =', text)案例 5-4：PPOCR 文本识别
det + rec 结果输出 text 与框。
import visiong
ocr = visiong.PPOCR(
det_model_path='/path/to/ppocrv3_det.rknn',
rec_model_path='/path/to/ppocrv4_rec.rknn',
dict_path='/path/to/ppocr_keys_v1.txt',
model_input_format='rgb'
)
img = visiong.ImageBuffer.load('/path/to/text.jpg').to_format('rgb888')
for item in ocr.infer(img):
print(item.text, item.text_score, item.rect)案例 5-5：NanoTrack 单目标跟踪
初始化 bbox 后持续 track。
import visiong
tracker = visiong.NanoTrack('/path/to/T.rknn', '/path/to/X.rknn', '/path/to/H.rknn')
cam = visiong.Camera(640, 360, format='rgb')
udp = visiong.DisplayUDP()
try:
cam.skip(8)
first = cam.snapshot()
tracker.init(first, (220, 120, 100, 100))
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
tr = tracker.track(frame)
x, y, w, h = tr.box
vis = frame.to_format('bgr888')
vis.draw_rectangle(x, y, w, h, color=(0, 255, 0), thickness=2)
vis.draw_string(10, 10, f'score={tr.score:.2f}', color=(0, 255, 0), scale=1.0, thickness=2)
udp.display(vis)
finally:
udp.release()
cam.release()案例 5-6：LowLevelNPU 数组输入
set_input_array + run + output_array。
import numpy as np
import visiong
ll = visiong.LowLevelNPU('/path/to/model.rknn')
x = np.zeros((1, 224, 224, 3), dtype=np.uint8)
ll.set_input_array(0, x, quantize_if_needed=True, zero_pad=True, sync_to_device=True)
ll.run(sync_outputs=True)
y = ll.output_array(0, dequantize_if_needed=True, sync_from_device=True)
print('output shape =', getattr(y, 'shape', None))案例 5-7：ROI 检测推理
infer 传 roi，减少无关区域。
import visiong
npu = visiong.NPU('yolov5', '/path/to/model.rknn', '/path/to/labels.txt')
img = visiong.ImageBuffer.load('/path/to/input.jpg').to_format('rgb888')
res = npu.infer(img, roi=(80, 40, 480, 280), model_format='rgb')
print('det count =', len(res))案例 5-8：LowLevelNPU 图像输入
set_input_image + output_float。
import visiong
ll = visiong.LowLevelNPU('/path/to/model.rknn')
img = visiong.ImageBuffer.load('/path/to/input.jpg').to_format('rgb888')
ll.set_input_image(0, img, color_order='rgb', keep_aspect=True, pad_value=114, driver_convert=True)
ll.run(sync_outputs=True)
out = ll.output_float(0, dequantize_if_needed=True, sync_from_device=True)
print('len(out) =', len(out))案例 5-9：打印 tensor 元信息
快速核对模型输入输出维度、量化信息。
import visiong
ll = visiong.LowLevelNPU('/path/to/model.rknn')
for t in ll.input_tensors():
print('in', t.index, t.name, t.dims, t.type, t.format, t.scale, t.zero_point)
for t in ll.output_tensors():
print('out', t.index, t.name, t.dims, t.type, t.format, t.scale, t.zero_point)案例 5-10：异步 run + wait
non_block 推理模式的最小模板。
import numpy as np
import visiong
ll = visiong.LowLevelNPU('/path/to/model.rknn')
x = np.zeros((1, 224, 224, 3), dtype=np.uint8)
ll.set_input_array(0, x)
ll.run(sync_outputs=False, non_block=True, timeout_ms=0)
ok = ll.wait(timeout_ms=1000)
print('wait ok =', ok)
y = ll.output_bytes(0, with_stride=False, sync_from_device=True)
print('output bytes =', len(y))
6. L6 系统与交互：PinMux、时钟、触控、GUI 高级
10 个案例
补齐系统控制与交互能力，形成完整端侧应用闭环。
案例 6-1：Pin 名解析与复用查询
parse_pin + get_mux + get_function_name。
import visiong
pm = visiong.PinMux()
try:
pid = pm.parse_pin('GPIO0_C3')
print('bank/pin =', pid.bank, pid.pin)
print('mux =', pm.get_mux(pid.bank, pid.pin))
print('func =', pm.get_function_name(pid.bank, pid.pin))
finally:
pm.close()案例 6-2：GPIO 输出闪烁
gpio_request_line / gpio_set_value / gpio_release_line。
import time
import visiong
pm = visiong.PinMux()
pid = pm.parse_pin('GPIO0_C3')
cfg = visiong.GpioLineConfig()
cfg.direction = 'out'
cfg.default_value = 0
try:
pm.gpio_request_line(pid.bank, pid.pin, cfg)
for i in range(10):
pm.gpio_set_value(pid.bank, pid.pin, i % 2)
time.sleep(0.2)
finally:
pm.gpio_release_line(pid.bank, pid.pin)
pm.close()案例 6-3：ADC 采样
list_adc_channels + read_adc。
import visiong
pm = visiong.PinMux()
try:
chs = pm.list_adc_channels()
print('adc channels =', chs)
s = pm.read_adc(0)
print('raw =', s.raw, 'mv =', s.millivolts)
finally:
pm.close()案例 6-4：接口冲突检查
check_conflict + ensure_interface 用于切换外设功能。
import visiong
pm = visiong.PinMux()
pid = pm.parse_pin('GPIO0_C3')
try:
rep = pm.check_conflict(pid.bank, pid.pin, target_function_or_group='uart4')
print('conflict =', rep.conflict, rep.reason)
ok = pm.ensure_interface('uart4')
print('ensure uart4 =', ok)
finally:
pm.close()案例 6-5：NPU 时钟查看与设置
NpuClock.status + set_rate_mhz。
import visiong
clk = visiong.NpuClock()
st = clk.status()
print('current hz =', st.current_rate_hz)
ret = clk.set_rate_mhz(500, update_cru_clk500m_src=True, unbind_rebind_npu=False, allow_unsafe_rate=False)
print('ok =', ret.ok, 'assigned =', ret.assigned_rate_hz, 'message =', ret.message)案例 6-6：GUI 最小窗口
GUI 基础帧循环。
import visiong
W, H = 320, 240
gui = visiong.GUI(W, H)
touch = visiong.Touch()
canvas_img = visiong.ImageBuffer.create(W, H, 'bgr888', (20, 20, 20))
while True:
gui.begin_frame(touch)
if gui.begin_window('Demo', 10, 10, 280, 200):
gui.layout_row_dynamic(28, 1)
gui.label('Hello VisionG GUI', 'left')
gui.button('OK')
gui.end_window()
gui.end_frame(canvas_img)案例 6-7：GPIO 输入读取
gpio_get_value + gpio_get_status。
import visiong
pm = visiong.PinMux()
pid = pm.parse_pin('GPIO0_C4')
cfg = visiong.GpioLineConfig()
cfg.direction = 'in'
try:
pm.gpio_request_line(pid.bank, pid.pin, cfg)
val = pm.gpio_get_value(pid.bank, pid.pin)
st = pm.gpio_get_status(pid.bank, pid.pin)
print('value =', val, 'requested =', st.requested)
finally:
pm.gpio_release_line(pid.bank, pid.pin)
pm.close()案例 6-8：驱动能力配置
set_drive_strength / set_pull / set_input_schmitt。
import visiong
pm = visiong.PinMux()
pid = pm.parse_pin('GPIO0_C3')
try:
pm.set_drive_strength(pid.bank, pid.pin, 3)
pm.set_pull(pid.bank, pid.pin, 'up')
pm.set_input_schmitt(pid.bank, pid.pin, True)
print('drive =', pm.get_drive_strength(pid.bank, pid.pin).level)
print('pull =', pm.get_pull(pid.bank, pid.pin).mode)
print('schmitt =', pm.get_input_schmitt(pid.bank, pid.pin).enabled)
finally:
pm.close()案例 6-9：触摸坐标读取与映射配置
Touch.configure_geometry + read。
import time
import visiong
touch = visiong.Touch()
touch.configure_geometry(240, 320, 270)
try:
for _ in range(200):
pts = touch.read()
if pts:
print([(p.x, p.y) for p in pts])
time.sleep(0.02)
finally:
touch.release()案例 6-10：GUI 滑条控制亮度
GUI + Camera + DisplayFB 联动最小示例。
import visiong
W, H = 320, 240
cam = visiong.Camera(W, H, format='bgr')
fb = visiong.DisplayFB()
touch = visiong.Touch()
gui = visiong.GUI(W, H)
brightness = 128
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
gui.begin_frame(touch)
if gui.begin_window('Camera Control', 8, 8, 304, 110):
gui.layout_row_dynamic(28, 1)
brightness = int(gui.slider('Brightness', brightness, 0, 255, 1))
cam.set_brightness(brightness)
gui.end_window()
gui.end_frame(frame)
fb.display(frame, (0, 0, frame.width, frame.height))
finally:
fb.release()
touch.release()
cam.release()
7. L7 运行诊断与模块函数 中级→高级
10 个案例
补齐模块级函数与运行诊断能力，方便线上排错与自检。
案例 7-1：设备唯一 ID
用 get_unique_id 标识设备实例。
import visiong
uid = visiong.get_unique_id()
print('unique id =', uid)案例 7-2：兼容值解密
decrypt_legacy_value 的最小调用。
import visiong
plain = visiong.decrypt_legacy_value('your-secret')
print('decrypted =', plain)案例 7-3：读取 DMA 统计
dma_state_metrics 用于观察 DMA 状态。
import visiong
m = visiong.dma_state_metrics(reset=False)
print('dma metrics =', m)案例 7-4：重置 DMA 统计
清空历史统计，重新开始采样。
import visiong
visiong.dma_state_reset_metrics()案例 7-5：导出 DMA 报告
将统计信息导出为文件留档。
import visiong
ok = visiong.dma_state_dump_metrics(output_path='dma_metrics.json', reset_after_dump=False)
print('dump ok =', ok)案例 7-6：关闭指定录制器
防止异常中断后句柄残留。
import visiong
visiong.close_venc_recorder('demo.mp4')案例 7-7：关闭全部录制器
服务重启前做统一清理。
import visiong
visiong.close_all_venc_recorders()案例 7-8：模块可用性检查
通过 hasattr 检查构建开关是否启用。
import visiong
mods = ['IVE', 'NPU', 'PPOCR', 'NanoTrack', 'GUI', 'PinMux']
for name in mods:
print(name, hasattr(visiong, name))案例 7-9：关键组件初始化自检
启动前快速检查核心能力是否可用。
import visiong
cam = visiong.Camera(320, 240, format='bgr')
udp = visiong.DisplayUDP()
npu = visiong.NPU('yolov5', '/path/to/model.rknn', '/path/to/labels.txt')
print('cam ok =', cam.is_initialized())
print('udp ok =', udp.is_initialized())
print('npu ok =', npu.is_initialized())
udp.release()
cam.release()案例 7-10：统一资源清理骨架
学习者可复用的 try/finally 模板。
import visiong
cam = udp = rtsp = rec = None
try:
cam = visiong.Camera(640, 360, format='yuv')
udp = visiong.DisplayUDP()
rtsp = visiong.DisplayRTSP(port=554, path='/live/0')
rec = visiong.VencRecorder('demo.mp4')
# ... your loop ...
finally:
if rec is not None:
rec.close()
if rtsp is not None:
rtsp.stop()
if udp is not None:
udp.release()
if cam is not None:
cam.release()
visiong.close_all_venc_recorders()
8. L8 综合项目模板（端到端） 高级
10 个案例
给出可直接扩展的端到端脚手架，缩短从 Demo 到项目的距离。
案例 8-1：采集 + IVE 边缘 + UDP
实时边缘流模板。
import visiong
cam = visiong.Camera(640, 360, format='gray')
udp = visiong.DisplayUDP()
ive = visiong.IVE()
try:
cam.skip(8)
while True:
g = cam.snapshot()
if not g.is_valid():
continue
edge = ive.canny(g, high_thresh=120, low_thresh=40)
udp.display(edge)
finally:
udp.release()
cam.release()案例 8-2：采集 + 检测 + RTSP
目标检测结果叠加后推 RTSP。
import visiong
cam = visiong.Camera(640, 360, format='rgb')
rtsp = visiong.DisplayRTSP(port=554, path='/live/0', codec='h264', fps=25)
det = visiong.NPU('yolov5', '/path/to/model.rknn', '/path/to/labels.txt')
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
vis = frame.to_format('yuv420sp')
rtsp.display(vis)
finally:
rtsp.stop()
cam.release()案例 8-3：采集 + PPOCR + 结果叠字
OCR 结果直接叠加在画面上。
import visiong
cam = visiong.Camera(640, 360, format='rgb')
udp = visiong.DisplayUDP()
ocr = visiong.PPOCR('/path/to/det.rknn', '/path/to/rec.rknn', '/path/to/keys.txt', model_input_format='rgb')
try:
cam.skip(8)
while True:
img = cam.snapshot()
if not img.is_valid():
continue
vis = img.to_format('bgr888')
for item in ocr.infer(img):
x, y, w, h = item.rect
vis.draw_rectangle(x, y, w, h, color=(0, 255, 255), thickness=2)
vis.draw_string(x, max(0, y - 20), item.text, color=(0, 255, 255), scale=1.0, thickness=2)
udp.display(vis)
finally:
udp.release()
cam.release()案例 8-4：采集 + NanoTrack + 丢失重置
跟踪低分时自动 reset 的基础逻辑。
import visiong
cam = visiong.Camera(640, 360, format='rgb')
udp = visiong.DisplayUDP()
tracker = visiong.NanoTrack('/path/to/T.rknn', '/path/to/X.rknn', '/path/to/H.rknn')
try:
cam.skip(8)
first = cam.snapshot()
tracker.init(first, (220, 120, 90, 90))
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
tr = tracker.track(frame)
if tr.score < 0.4:
tracker.reset()
tracker.init(frame, (220, 120, 90, 90))
vis = frame.to_format('bgr888')
x, y, w, h = tr.box
vis.draw_rectangle(x, y, w, h, color=(0, 255, 0), thickness=2)
udp.display(vis)
finally:
udp.release()
cam.release()案例 8-5：采集 + LowLevelNPU + 结果日志
底层推理循环模板。
import visiong
cam = visiong.Camera(640, 360, format='rgb')
ll = visiong.LowLevelNPU('/path/to/model.rknn')
try:
cam.skip(8)
while True:
img = cam.snapshot()
if not img.is_valid():
continue
ll.set_input_image(0, img, color_order='rgb', keep_aspect=True, pad_value=114, driver_convert=True)
ll.run(sync_outputs=True)
out = ll.output_float(0)
print('out len =', len(out))
finally:
cam.release()案例 8-6：HTTP 直播 + 定时截图
一边网页直播，一边周期保存样本图。
import time
import visiong
cam = visiong.Camera(640, 360, format='bgr')
http = visiong.DisplayHTTP(port=8080, mode='jpg', quality=75)
last_save = 0.0
idx = 0
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
http.display(frame)
now = time.time()
if now - last_save > 5:
frame.save(f'snap_{idx:04d}.jpg', quality=90)
idx += 1
last_save = now
finally:
http.stop()
cam.release()案例 8-7：离线数据集预处理 + 推理
批量 load/letterbox/infer 的模板。
import visiong
npu = visiong.NPU('yolov5', '/path/to/model.rknn', '/path/to/labels.txt')
images = ['/path/to/a.jpg', '/path/to/b.jpg']
for p in images:
img = visiong.ImageBuffer.load(p).to_format('rgb888')
inp = img.letterbox(640, 640, color=(114, 114, 114))
dets = npu.infer(inp, model_format='rgb')
print(p, 'det=', len(dets))案例 8-8：GPIO 触发抓图
读取输入脚状态触发 snapshot 保存。
import time
import visiong
pm = visiong.PinMux()
pid = pm.parse_pin('GPIO0_C4')
cfg = visiong.GpioLineConfig()
cfg.direction = 'in'
cam = visiong.Camera(640, 360, format='bgr')
try:
pm.gpio_request_line(pid.bank, pid.pin, cfg)
cam.skip(8)
idx = 0
while True:
if pm.gpio_get_value(pid.bank, pid.pin) == 1:
img = cam.snapshot()
if img.is_valid():
img.save(f'trigger_{idx:03d}.jpg', quality=90)
idx += 1
time.sleep(0.2)
finally:
pm.gpio_release_line(pid.bank, pid.pin)
pm.close()
cam.release()案例 8-9：NPU 时钟调优 + 基准测试
简单比较不同时钟下的推理吞吐。
import time
import visiong
clk = visiong.NpuClock()
npu = visiong.NPU('yolov5', '/path/to/model.rknn', '/path/to/labels.txt')
img = visiong.ImageBuffer.load('/path/to/bench.jpg').to_format('rgb888')
for mhz in [400, 500]:
ret = clk.set_rate_mhz(mhz)
t0 = time.time()
for _ in range(20):
npu.infer(img, model_format='rgb')
dt = time.time() - t0
print(mhz, 'MHz', 'ok=', ret.ok, 'fps=', 20 / max(1e-6, dt))案例 8-10：GUI 相机应用骨架
预览+按钮的最小应用框架。
import visiong
W, H = 320, 240
cam = visiong.Camera(W, H, format='bgr')
fb = visiong.DisplayFB()
touch = visiong.Touch()
gui = visiong.GUI(W, H)
try:
cam.skip(8)
while True:
frame = cam.snapshot()
if not frame.is_valid():
continue
gui.begin_frame(touch)
if gui.begin_window('App', 8, 8, 304, 90):
gui.layout_row_dynamic(28, 2)
if gui.button('Snap'):
frame.save('gui_snap.jpg', quality=90)
gui.button('Exit')
gui.end_window()
gui.end_frame(frame)
fb.display(frame, (0, 0, frame.width, frame.height))
finally:
fb.release()
touch.release()
cam.release()6) 按类最小片段
每个类给一个最小调用片段，适合先复制再改参数。
Camera · 最小调用片段
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.init(target_width=640, target_height=360)
result = cam.skip(10)
result = cam.snapshot()IVEModel · 最小调用片段
import visiong
model = visiong.IVEModel(width=640, height=360)
# IVEModel 暂无公开方法MotionVector · 最小调用片段
import visiong
mv = visiong.MotionVector()
# MotionVector 暂无公开方法IVE · 最小调用片段
import visiong
ive = visiong.IVE()
visiong.IVE.set_log_enabled(enabled=True)
result = visiong.IVE.is_log_enabled()
result = ive.filter(src=img, mask=[[0, -1, 0], [-1, 5, -1], [0, -1, 0]])Blob · 最小调用片段
import visiong
blob = visiong.Blob(x=10, y=10, w=120, h=90, cx=160, cy=120, pixels=0)
# Blob 暂无公开方法Line · 最小调用片段
import visiong
line = visiong.Line()
# Line 暂无公开方法Circle · 最小调用片段
import visiong
circle = visiong.Circle()
# Circle 暂无公开方法QRCode · 最小调用片段
import visiong
qRCode = visiong.QRCode()
# QRCode 暂无公开方法ImageBuffer · 最小调用片段
import visiong
img = visiong.ImageBuffer()
result = visiong.ImageBuffer.create(width=640, height=360, format='bgr888')
result = img.to_numpy(copy=True)
result = img.numpy_view()DisplayUDP · 最小调用片段
import visiong
udp = visiong.DisplayUDP()
result = udp.init(ip_address='127.0.0.1', port=8080)
udp.display(img_buf=img)
udp.release()TouchPoint · 最小调用片段
import visiong
touchPoint = visiong.TouchPoint()
# TouchPoint 暂无公开方法Touch · 最小调用片段
import visiong
touch = visiong.Touch()
touch.release()
result = touch.is_pressed()
result = touch.read()DisplayFB · 最小调用片段
import visiong
fb = visiong.DisplayFB()
fb.display(img_buf=img, roi=(0, 0, 320, 240))
fb.release()
result = fb.is_initialized()Detection · 最小调用片段
import visiong
detection = visiong.Detection()
# Detection 暂无公开方法NPU · 最小调用片段
import visiong
npu = visiong.NPU(model_type='yolov5', model_path='/path/to/model.rknn')
result = npu.infer(img_buf=img)
result = npu.get_face_feature(face_image=img)
result = npu.recognize_plate(plate_image=img)LowLevelTensorInfo · 最小调用片段
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
# LowLevelTensorInfo 暂无公开方法LowLevelNPU · 最小调用片段
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.is_initialized()
result = llnpu.num_inputs()
result = llnpu.num_outputs()OCRResult · 最小调用片段
import visiong
oCRResult = visiong.OCRResult()
# OCRResult 暂无公开方法PPOCR · 最小调用片段
import visiong
ocr = visiong.PPOCR(det_model_path='/path/to/det.rknn', rec_model_path='/path/to/rec.rknn')
result = ocr.infer(img_buf=img)
result = ocr.is_initialized()NanoTrackResult · 最小调用片段
import visiong
nanoTrackResult = visiong.NanoTrackResult()
# NanoTrackResult 暂无公开方法NanoTrack · 最小调用片段
import visiong
tracker = visiong.NanoTrack(template_model=0, search_model=0, head_model='/path/to/head.rknn')
result = tracker.init(img_buf=img, bbox=(60, 40, 120, 90))
result = tracker.track(img_buf=img)
result = tracker.is_initialized()Canvas · 最小调用片段
import visiong
canvas = visiong.Canvas()
# Canvas 暂无公开方法GUI · 最小调用片段
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.begin_frame(touch_device=None)
result = gui.begin_window(title='Demo', x=10, y=10, w=120, h=90)
result = gui.end_window()PinId · 最小调用片段
import visiong
pinId = visiong.PinId()
# PinId 暂无公开方法PinMuxRegisterInfo · 最小调用片段
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
# PinMuxRegisterInfo 暂无公开方法PinAltFunction · 最小调用片段
import visiong
pinAltFunction = visiong.PinAltFunction()
# PinAltFunction 暂无公开方法PinRuntimeStatus · 最小调用片段
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
# PinRuntimeStatus 暂无公开方法PinConflictReport · 最小调用片段
import visiong
pinConflictReport = visiong.PinConflictReport()
# PinConflictReport 暂无公开方法FunctionInterfaceStatus · 最小调用片段
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
# FunctionInterfaceStatus 暂无公开方法AdcChannelStatus · 最小调用片段
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
# AdcChannelStatus 暂无公开方法GpioLineConfig · 最小调用片段
import visiong
gpioLineConfig = visiong.GpioLineConfig()
# GpioLineConfig 暂无公开方法GpioLineStatus · 最小调用片段
import visiong
gpioLineStatus = visiong.GpioLineStatus()
# GpioLineStatus 暂无公开方法DriveStrengthStatus · 最小调用片段
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
# DriveStrengthStatus 暂无公开方法PullStatus · 最小调用片段
import visiong
pullStatus = visiong.PullStatus()
# PullStatus 暂无公开方法SchmittStatus · 最小调用片段
import visiong
schmittStatus = visiong.SchmittStatus()
# SchmittStatus 暂无公开方法PinMux · 最小调用片段
import visiong
pm = visiong.PinMux()
result = pm.is_open()
pm.close()
result = pm.parse_pin(pin_name='GPIO0_C3')NpuClockStatus · 最小调用片段
import visiong
npuClockStatus = visiong.NpuClockStatus()
# NpuClockStatus 暂无公开方法NpuClockApplyResult · 最小调用片段
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
# NpuClockApplyResult 暂无公开方法NpuClock · 最小调用片段
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.status()
result = npu_clock.supported_rates_hz()
result = npu_clock.supported_rates_mhz()DisplayHTTP · 最小调用片段
import visiong
http = visiong.DisplayHTTP()
http.stop()
http.display(img=img)
result = http.is_running()DisplayRTSP · 最小调用片段
import visiong
rtsp = visiong.DisplayRTSP()
rtsp.stop()
rtsp.set_fps(fps=30)
result = rtsp.get_fps()VencRecorder · 最小调用片段
import visiong
rec = visiong.VencRecorder(filepath='/path/to/demo.jpg')
rec.write(img=img)
rec.close()
result = rec.is_open()DisplayHTTPFLV · 最小调用片段
import visiong
flv = visiong.DisplayHTTPFLV()
flv.stop()
flv.display(img=img)
result = flv.is_running()7) 逐函数示例库（自动生成）
覆盖全部模块函数、类方法和字段。每一项都提供最小可复制代码片段，便于学习者直接改参数后运行。
自动示例总数：453
模块级函数
visiong.close_venc_recorder(filepath)
Closes a cached MP4 writer for filepath (finalize MP4).
import visiong
visiong.close_venc_recorder('demo.mp4')visiong.close_all_venc_recorders()
Closes all cached MP4 writers (finalize MP4).
import visiong
visiong.close_all_venc_recorders()visiong.get_unique_id()
Reads the unique 6-byte chip ID from the RV1106 OTP area and returns it as a 12-character hex string.
import visiong
uid = visiong.get_unique_id()visiong.decrypt_legacy_value(key)
Decrypts the stored legacy value with a 64-character hex key.
import visiong
key = 'your-secret'
value = visiong.decrypt_legacy_value(key)visiong.dma_state_metrics(reset=...)
Returns dma-buf state-machine counters as a dict. Set reset=True to clear counters after reading.
import visiong
metrics = visiong.dma_state_metrics(reset=False)visiong.dma_state_reset_metrics()
Resets dma-buf state-machine counters.
import visiong
visiong.dma_state_reset_metrics()visiong.dma_state_dump_metrics(output_path=..., reset_after_dump=...)
Returns dma-buf state-machine counters as JSON and optionally writes them to output_path.
import visiong
dump = visiong.dma_state_dump_metrics(output_path='metrics.json')Camera
Camera.init(target_width, target_height, format=..., hdr=..., crop_mode=...)
Initializes the camera. format defaults to 'yuv'. Supported values: 'bgr', 'rgb', 'yuv'/'yuv420', or 'gray'. crop_mode: 'auto' (default, follows the camera max-resolution aspect ratio), 'off', or any ratio such as '16:9', '4:3', '1:1', o...
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.init(target_width=640, target_height=360)Camera.skip(num_frames=...)
(Reads and discards a specified number of frames from the camera.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.skip(10)Camera.snapshot()
Captures a single frame from the camera and returns an ImageBuffer.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.snapshot()Camera.release()
Releases the camera and frees resources. Safe to call even if not initialized.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.release()Camera.is_initialized()
Returns True if the camera has been successfully initialized.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.is_initialized()Camera.get_capture_width()
Returns the actual capture width (alias for actual_width).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_capture_width()Camera.get_capture_height()
Returns the actual capture height (alias for actual_height).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_capture_height()Camera.set_saturation(value)
Sets the image saturation. Range: [0, 255]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_saturation(value=0)Camera.set_contrast(value)
Sets the image contrast. Range: [0, 255]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_contrast(value=0)Camera.set_brightness(value)
Sets the image brightness. Range: [0, 255]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_brightness(value=0)Camera.set_sharpness(value)
Sets the image sharpness. Range: [0, 100]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_sharpness(value=0)Camera.set_hue(value)
Sets the image hue. Range: [0, 255]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_hue(value=0)Camera.set_white_balance_mode(mode)
Sets white balance mode ('auto' or 'manual'). Raises ValueError on invalid mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_white_balance_mode(mode='auto')Camera.set_white_balance_temperature(temp)
Sets white balance color temperature in manual mode. Raises ValueError on invalid input or RuntimeError if not in manual mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_white_balance_temperature(temp=4500)Camera.set_exposure_mode(mode)
Sets exposure mode ('auto' or 'manual'). Raises ValueError on invalid mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_exposure_mode(mode='auto')Camera.set_exposure_time(time_in_seconds)
Sets manual exposure time in seconds. Must be positive. Raises RuntimeError if not in manual mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_exposure_time(time_in_seconds=0.01)Camera.set_exposure_gain(gain)
Sets manual exposure gain. Typical range: [0, 127]. Raises ValueError on invalid input or RuntimeError if not in manual mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_exposure_gain(gain=32)Camera.set_spatial_denoise_level(level)
Sets the spatial (2D) denoise level. Range: [0, 100]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_spatial_denoise_level(level=3)Camera.set_temporal_denoise_level(level)
Sets the temporal (3D) denoise level. Range: [0, 100]. Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_temporal_denoise_level(level=3)Camera.set_frame_rate(fps)
Sets the camera frame rate. Range: [10, 60] (or 0 for auto). Raises ValueError on invalid input.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_frame_rate(fps=30)Camera.set_power_line_frequency(mode)
Sets anti-flicker mode ('50hz', '60hz', or 'off'). Raises ValueError on invalid mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_power_line_frequency(mode='auto')Camera.set_flip(flip, mirror)
Sets image flip (vertical) and mirror (horizontal).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_flip(flip=True, mirror=True)Camera.get_saturation()
Gets the current image saturation (0-255).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_saturation()Camera.get_contrast()
Gets the current image contrast (0-255).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_contrast()Camera.get_brightness()
Gets the current image brightness (0-255).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_brightness()Camera.get_sharpness()
Gets the current image sharpness (0-100).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_sharpness()Camera.get_hue()
Gets the current image hue (0-255).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_hue()Camera.get_white_balance_mode()
Gets the current white balance mode ('auto' or 'manual').
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_white_balance_mode()Camera.get_white_balance_temperature()
Gets the current white balance color temperature.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_white_balance_temperature()Camera.get_exposure_mode()
Gets the current exposure mode ('auto' or 'manual').
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_exposure_mode()Camera.get_exposure_time()
Gets the current exposure time in seconds.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_exposure_time()Camera.get_exposure_gain()
Gets the current exposure gain.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_exposure_gain()Camera.lock_focus()
Locks the autofocus at its current position. Prevents further automatic adjustments.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.lock_focus()Camera.unlock_focus()
Unlocks the autofocus, allowing it to resume its configured mode (e.g., continuous focus).
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.unlock_focus()Camera.trigger_focus()
Performs a single, one-shot autofocus search. This is an action, not a mode.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.trigger_focus()Camera.set_focus_mode(mode)
Sets the autofocus mode. Supported modes:
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_focus_mode(mode='continuous')Camera.set_manual_focus(position)
Moves the lens to a specific motor code position. This implicitly sets the mode to 'manual'.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
cam.set_manual_focus(position=512)Camera.get_focus_position()
Returns the current motor code position of the lens. Returns -1 on failure.
import visiong
cam = visiong.Camera(target_width=640, target_height=360)
result = cam.get_focus_position()IVEModel
该类没有公开方法示例。
MotionVector
MotionVector.__repr__()
<MotionVector status=
import visiong
mv = visiong.MotionVector()
text = repr(mv)MotionVector.status · field
readonly_field
import visiong
mv = visiong.MotionVector()
value = mv.statusMotionVector.mv_x · field
readonly_field
import visiong
mv = visiong.MotionVector()
value = mv.mv_xMotionVector.mv_y · field
readonly_field
import visiong
mv = visiong.MotionVector()
value = mv.mv_yIVE
IVE.set_log_enabled(enabled) · static
Enable or disable IVE internal logs (default off; env VISIONG_IVE_LOG=1 enables).
import visiong
visiong.IVE.set_log_enabled(enabled=True)IVE.is_log_enabled() · static
Returns True if IVE internal logs are enabled.
import visiong
result = visiong.IVE.is_log_enabled()IVE.filter(src, mask)
5x5 filter. src: GRAY8. mask: 25 ints. Returns GRAY8 - do not pass to cast_16bit_to_8bit. Use a 5x5 Gaussian mask for blur (no separate gaussian_filter).
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.filter(src=img, mask=[[0, -1, 0], [-1, 5, -1], [0, -1, 0]])IVE.sobel(src, out_ctrl=..., out_format=...)
Sobel edge detection. src: GRAY8. Returns (horizontal_edges, vertical_edges). When out_format=S16C1 each result is S16C1 - unpack and pass to cast_16bit_to_8bit for display; when U8C1 each result is GRAY8.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.sobel(src=img)IVE.canny(src, high_thresh, low_thresh)
Canny edge detection. src must be GRAY8. high_thresh and low_thresh typically in 0-255.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.canny(src=img, high_thresh=120, low_thresh=40)IVE.mag_and_ang(src, threshold=..., return_magnitude=...)
Gradient magnitude/angle. return_magnitude=True -> S16C1 (use cast_16bit_to_8bit for display); False -> U8C1.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.mag_and_ang(src=img)IVE.dilate(src, kernel_size=...)
Performs morphological dilation. `src` must be GRAY8. `kernel_size` can be 3 or 5.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.dilate(src=img)IVE.erode(src, kernel_size=...)
Performs morphological erosion. `src` must be GRAY8. `kernel_size` can be 3 or 5.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.erode(src=img)IVE.ordered_stat_filter(src, mode)
Performs an ordered statistic filter (Median, Max, Min). `src` must be GRAY8.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.ordered_stat_filter(src=img, mode=visiong.OrdStatFilterMode.MEDIAN)IVE.add(src1, src2)
Pixel-wise addition. Returns GRAY8; do not pass to cast_16bit_to_8bit.
import visiong
ive = visiong.IVE()
img1 = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.add(src1=img1, src2=img2)IVE.sub(src1, src2, mode=...)
Pixel-wise subtraction. Returns GRAY8; do not pass to cast_16bit_to_8bit. Mode: ABS or SHIFT.
import visiong
ive = visiong.IVE()
img1 = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.sub(src1=img1, src2=img2)IVE.logic_op(src1, src2, op)
Performs a pixel-wise logical operation (AND, OR, XOR) on two images.
import visiong
ive = visiong.IVE()
img1 = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.logic_op(src1=img1, src2=img2, op=visiong.LogicOp.AND)IVE.threshold(src)
Thresholding on GRAY8. kwargs accept low/low_thresh, high/high_thresh, mode. Returns GRAY8; do not pass to cast_16bit_to_8bit.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.threshold(src=img)IVE.cast_16bit_to_8bit(src, mode)
Casts 16-bit (S16C1/U16C1) to 8-bit. Input must be from sobel(S16C1), mag_and_ang(magnitude), norm_grad, or sad (first element); not from filter, add, sub, threshold, gmm.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.cast_16bit_to_8bit(src=img, mode=visiong.Cast16to8Mode.S16_TO_U8_ABS)IVE.hist(src)
Calculates the histogram of a GRAY8 image. Returns a list of 256 integers.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.hist(src=img)IVE.equalize_hist(src)
Performs histogram equalization on a GRAY8 image.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.equalize_hist(src=img)IVE.integral(src, mode=...)
Integral image. mode: COMBINE, SUM, or SQSUM. Output U64C1.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.integral(src=img)IVE.ccl(src, min_area=...)
Connected Components Labeling on a binarized GRAY8 image. Blobs with area < min_area are filtered. Returns list of Blob.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.ccl(src=img)IVE.ncc(src1, src2)
Normalized cross-correlation. If sizes differ, set auto_resize=True (default) to resize src2 to src1 size.
import visiong
ive = visiong.IVE()
img1 = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.ncc(src1=img1, src2=img2)IVE.csc(src, mode)
Performs Color Space Conversion (e.g., YUV to RGB).
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.csc(src=img, mode=visiong.CscMode.YUV2RGB_BT601_LIMITED)IVE.yuv_to_rgb(src)
Converts YUV420SP (or YUV420SP_VU) to RGB. Use full_range=True/False, or pass mode=visiong.CscMode.YUV2RGB_*.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.yuv_to_rgb(src=img)IVE.yuv_to_hsv(src)
Converts YUV420SP (or YUV420SP_VU) to HSV. Use full_range=True/False, or pass mode=visiong.CscMode.YUV2HSV_*.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.yuv_to_hsv(src=img)IVE.rgb_to_yuv(src, full_range=...)
Converts RGB/BGR image to YUV420SP.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.rgb_to_yuv(src=img)IVE.rgb_to_hsv(src, full_range=...)
Converts RGB/BGR image to HSV. H:[0,180], S:[0,255], V:[0,255].
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.rgb_to_hsv(src=img)IVE.dma(src, mode=...)
Performs a direct memory access operation (e.g., copy).
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.dma(src=img)IVE.cast_8bit_to_8bit(src, bias, numerator, denominator)
Scales an 8-bit image using the formula: dst = (src * numerator / denominator) + bias.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.cast_8bit_to_8bit(src=img, bias=0, numerator=1, denominator=1)IVE.map(src, lut)
Pixel LUT mapping. lut: list of 256 integers in [0,255]; output[x] = lut[src[x]].
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
lut = [i for i in range(256)]
result = ive.map(src=img, lut=lut)IVE.gmm(src, model, first_frame=...)
GMM background subtraction. Returns (foreground, background); both GRAY8 - do not pass to cast_16bit_to_8bit.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
model = visiong.IVEModel(640, 360, 0)
result = ive.gmm(src=img, model=model)IVE.gmm2(src, factor, model, first_frame=...)
GMM2 with factor image. Returns (foreground, background); both GRAY8.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
model = visiong.IVEModel(640, 360, 0)
result = ive.gmm2(src=img, factor=0.5, model=model)IVE.lbp(src, abs_mode=..., threshold=...)
Calculates the Local Binary Pattern of an image.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.lbp(src=img)IVE.norm_grad(src)
Normalized gradient. Returns (horizontal_grad, vertical_grad); both S16C1 - use cast_16bit_to_8bit for display.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.norm_grad(src=img)IVE.lk_optical_flow(prev_img, next_img, points)
Performs Lucas-Kanade optical flow. points: list of (x,y). Returns list of MotionVector; mv_x/mv_y are S9.7 (divide by 128 for pixels). Up to ~500 points.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
points = [(120, 80), (180, 120)]
result = ive.lk_optical_flow(prev_img=img, next_img=img, points=points)IVE.st_corner(src, max_corners=..., min_dist=..., quality_level=...)
Performs Shi-Tomasi corner detection. Returns a list of (x, y) corner tuples.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.st_corner(src=img)IVE.match_bg_model(current_img, bg_model, frame_num)
Matches the current image against a background model. Returns a foreground flag image.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.match_bg_model(current_img=img, bg_model=img2, frame_num=img)IVE.update_bg_model(current_img, fg_flag, bg_model, frame_num)
Updates the background model. Returns the new background image.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.update_bg_model(current_img=img, fg_flag=img2, bg_model=img2, frame_num=img)IVE.sad(src1, src2, mode, threshold, min_val=..., max_val=...)
Sum of Absolute Differences. Returns (sad_image U16C1, threshold_image U8C1); cast sad_image for 8-bit display.
import visiong
ive = visiong.IVE()
img1 = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = ive.sad(src1=img1, src2=img2, mode=visiong.SadMode.MB_8X8, threshold=120)IVE.create_pyramid(src, levels)
Builds an image pyramid with levels levels. Returns list of ImageBuffer from full size down.
import visiong
ive = visiong.IVE()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ive.create_pyramid(src=img, levels=3)Blob
Blob.__repr__()
<Blob rect=(
import visiong
blob = visiong.Blob(x=10, y=10, w=120, h=90, cx=160, cy=120, pixels=0)
text = repr(blob)Line
Line.__repr__()
类方法最小示例
import visiong
line = visiong.Line()
text = repr(line)Line.x1 · field
readonly_field
import visiong
line = visiong.Line()
value = line.x1Line.y1 · field
readonly_field
import visiong
line = visiong.Line()
value = line.y1Line.x2 · field
readonly_field
import visiong
line = visiong.Line()
value = line.x2Line.y2 · field
readonly_field
import visiong
line = visiong.Line()
value = line.y2Line.magnitude · field
readonly_field
import visiong
line = visiong.Line()
value = line.magnitudeLine.theta · field
readonly_field
import visiong
line = visiong.Line()
value = line.thetaLine.rho · field
readonly_field
import visiong
line = visiong.Line()
value = line.rhoCircle
Circle.__repr__()
<Circle center=(
import visiong
circle = visiong.Circle()
text = repr(circle)Circle.cx · field
readonly_field
import visiong
circle = visiong.Circle()
value = circle.cxCircle.cy · field
readonly_field
import visiong
circle = visiong.Circle()
value = circle.cyCircle.r · field
readonly_field
import visiong
circle = visiong.Circle()
value = circle.rCircle.magnitude · field
readonly_field
import visiong
circle = visiong.Circle()
value = circle.magnitudeQRCode
QRCode.__repr__()
from_numpy_zero_copy: width and height must both be even for zero-copy hardware-friendly ImageBuffer wrapping.
import visiong
qRCode = visiong.QRCode()
text = repr(qRCode)QRCode.corners · field
property_readonly
import visiong
qRCode = visiong.QRCode()
value = qRCode.cornersQRCode.payload · field
property_readonly
import visiong
qRCode = visiong.QRCode()
value = qRCode.payloadImageBuffer
ImageBuffer.create(width, height, format, color=...) · static
Creates a new ImageBuffer filled with the given color. format: e.g. 'rgb888', 'bgr888', 'gray8'. color: (R,G,B) or (R,G,B,A).
import visiong
result = visiong.ImageBuffer.create(width=640, height=360, format='bgr888')ImageBuffer.__array__(copy=..., dtype=...)
Allows NumPy conversion via np.asarray(img). copy=False returns a read-only array pinned to the ImageBuffer lifetime; color exports use a BGR-shaped view over the current CPU-readable storage, which may be a cached converted backing stor...
import visiong
import numpy as np
img = visiong.ImageBuffer()
arr = np.asarray(img)ImageBuffer.to_numpy(copy=...)
Returns a NumPy array: 2D (H,W) for grayscale, 3D (H,W,3) BGR-shaped for color. copy=False returns a read-only lifetime-pinned view over the current CPU-readable storage.
import visiong
img = visiong.ImageBuffer()
result = img.to_numpy(copy=True)ImageBuffer.numpy_view()
Returns a read-only lifetime-pinned NumPy view over the current CPU-readable storage.
import visiong
img = visiong.ImageBuffer()
result = img.numpy_view()ImageBuffer.from_numpy(array, format=..., copy=...) · static
Creates an ImageBuffer from a NumPy array. format='auto' infers: 2D or 1-channel -> GRAY8, 3-channel -> BGR888, 4-channel -> BGRA8888. copy=False uses zero-copy when the input is uint8, C-contiguous, and has even width/height.
import visiong
import numpy as np
arr = np.zeros((1, 224, 224, 3), dtype=np.uint8)
result = visiong.ImageBuffer.from_numpy(array=arr)ImageBuffer.from_numpy_zero_copy(array, format=...) · static
Strict zero-copy import from a NumPy array. Requires dtype=uint8, C-contiguous layout, and even width/height; otherwise raises instead of silently copying.
import visiong
import numpy as np
arr = np.zeros((1, 224, 224, 3), dtype=np.uint8)
result = visiong.ImageBuffer.from_numpy_zero_copy(array=arr)ImageBuffer.is_valid()
Returns True if the buffer holds valid image data.
import visiong
img = visiong.ImageBuffer()
result = img.is_valid()ImageBuffer.copy()
Creates and returns a deep copy of the image buffer.
import visiong
img = visiong.ImageBuffer()
result = img.copy()ImageBuffer.load(filepath) · static
Loads an image from a file into a new ImageBuffer. Supports JPEG, PNG, and BMP (via stb_image).
import visiong
result = visiong.ImageBuffer.load(filepath='/path/to/demo.jpg')ImageBuffer.save(filepath, quality=...)
Saves the ImageBuffer to a file. Supports JPEG, PNG, and BMP. Quality (1-100) applies to JPEG/PNG. Uses software encoding (stb_image).
import visiong
img = visiong.ImageBuffer()
result = img.save(filepath='/path/to/demo.jpg')ImageBuffer.save_hsv_bin(filepath)
Converts YUV420SP image to HSV using IVE hardware and saves raw HSV binary data to the given filepath. Input must be YUV420SP or YUV420SP_VU.
import visiong
img = visiong.ImageBuffer()
result = img.save_hsv_bin(filepath='/path/to/demo.jpg')ImageBuffer.save_venc_jpg(filepath, quality=...)
it checks if size and format match; otherwise, it auto-initializes the encoder.
import visiong
img = visiong.ImageBuffer()
result = img.save_venc_jpg(filepath='/path/to/demo.jpg')ImageBuffer.save_venc_h264(filepath, quality=..., rc_mode=..., fps=..., append=..., container=..., mp4_faststart=...)
it muxes into MP4 and caches the writer until close_venc_recorder/close_all_venc_recorders or process exit.
import visiong
img = visiong.ImageBuffer()
result = img.save_venc_h264(filepath='/path/to/demo.jpg')ImageBuffer.save_venc_h265(filepath, quality=..., rc_mode=..., fps=..., append=..., container=..., mp4_faststart=...)
it muxes into MP4 and caches the writer until close_venc_recorder/close_all_venc_recorders or process exit.
import visiong
img = visiong.ImageBuffer()
result = img.save_venc_h265(filepath='/path/to/demo.jpg')ImageBuffer.to_format(new_format)
Converts the image to the given pixel format string (e.g. 'rgb888', 'bgr888', 'gray8', 'yuv420sp').
import visiong
img = visiong.ImageBuffer()
result = img.to_format(new_format='rgb')ImageBuffer.to_grayscale()
Converts the image to GRAY8. Uses RGA for color images.
import visiong
img = visiong.ImageBuffer()
result = img.to_grayscale()ImageBuffer.resize(new_width, new_height)
Resizes the image to new_width x new_height using RGA.
import visiong
img = visiong.ImageBuffer()
result = img.resize(new_width=320, new_height=240)ImageBuffer.crop(x, y, w, h)
Crops to the region (x, y, w, h). rect_tuple: (x, y, w, h). Uses RGA.
import visiong
img = visiong.ImageBuffer()
result = img.crop(x=10, y=10, w=120, h=90)ImageBuffer.rotate(angle_degrees)
Rotates the image by 90, 180, or 270 degrees using hardware acceleration.
import visiong
img = visiong.ImageBuffer()
result = img.rotate(angle_degrees=90)ImageBuffer.flip(horizontal, vertical)
Flips the image horizontally and/or vertically using hardware acceleration.
import visiong
img = visiong.ImageBuffer()
result = img.flip(horizontal=True, vertical=False)ImageBuffer.find_blobs(thresholds, invert=..., roi=..., x_stride=..., y_stride=..., area_threshold=..., pixels_threshold=..., merge=..., margin=..., mode=..., erode_size=..., dilate_size=...)
(Finds blobs by color thresholds. thresholds: list of 6-tuples (H_min,H_max,S_min,S_max,V_min,V_max) for HSV; mode 0=HSV, 1=LAB. For grayscale use the overload with [(gray_min, gray_max)].)
import visiong
img = visiong.ImageBuffer()
result = img.find_blobs(thresholds=[(0, 80)])ImageBuffer.find_lines(x, y, w, h, x_stride=..., y_stride=..., threshold=..., rho_resolution_px=..., theta_resolution_deg=..., canny_low_thresh=..., canny_high_thresh=...)
theta_resolution_deg
import visiong
img = visiong.ImageBuffer()
result = img.find_lines(x=10, y=10, w=120, h=90)ImageBuffer.find_circles(x, y, w, h, x_stride=..., y_stride=..., threshold=..., r_min=..., r_max=..., r_step=..., canny_low_thresh=..., canny_high_thresh=...)
canny_high_thresh
import visiong
img = visiong.ImageBuffer()
result = img.find_circles(x=10, y=10, w=120, h=90)ImageBuffer.find_polygons(x, y, w, h, min_area=..., max_area=..., min_sides=..., max_sides=..., accuracy=...)
accuracy must be 'fast', 'normal', or 'accurate'.
import visiong
img = visiong.ImageBuffer()
result = img.find_polygons(x=10, y=10, w=120, h=90)ImageBuffer.find_qrcodes()
Finds and decodes QR codes in the image. The image is automatically converted to grayscale if needed.
import visiong
img = visiong.ImageBuffer()
result = img.find_qrcodes()ImageBuffer.find_squares(roi=..., threshold_val, min_area=..., approx_epsilon=..., corner_sample_radius=..., corner_ratio_thresh=..., edge_check_offset=..., area_sample_points=..., area_white_thresh=..., area_morph_close_kernel_size=..., duplicate_center_thresh=..., duplicate_area_thresh=...)
(Finds squares in the image using a robust corner-based algorithm.
import visiong
img = visiong.ImageBuffer()
result = img.find_squares(threshold_val=120)ImageBuffer.binarize(method=..., threshold_range=..., invert=..., adaptive_block_size=..., adaptive_c=..., pre_blur_kernel_size=..., post_morph_kernel_size=...)
(Performs image binarization with adjustable denoising strength.
import visiong
img = visiong.ImageBuffer()
result = img.binarize(method='otsu')ImageBuffer.warp_perspective(quad, out_width, out_height)
(Performs a perspective warp transformation.
import visiong
img = visiong.ImageBuffer()
result = img.warp_perspective(quad=[(0, 0), (319, 0), (319, 239), (0, 239)], out_width=320, out_height=240)ImageBuffer.letterbox(target_width, target_height, color=...)
Scales the image to fit inside target dimensions while preserving aspect ratio, then pads with color to fill target_width x target_height. Uses RGA.
import visiong
img = visiong.ImageBuffer()
result = img.letterbox(target_width=640, target_height=360)ImageBuffer.draw_line(x0, y0, x1, y1, color=..., thickness=...)
Draws a line on the image in-place and returns itself.
import visiong
img = visiong.ImageBuffer()
img.draw_line(x0=10, y0=10, x1=120, y1=90)ImageBuffer.draw_rectangle(x, y, w, h, color=..., thickness=..., fill=...)
Draws a rectangle on the image in-place and returns itself.
import visiong
img = visiong.ImageBuffer()
img.draw_rectangle(x=10, y=10, w=120, h=90)ImageBuffer.draw_circle(cx, cy, radius, color=..., thickness=..., fill=...)
Draws a circle on the image in-place and returns itself.
import visiong
img = visiong.ImageBuffer()
img.draw_circle(cx=160, cy=120, radius=30)ImageBuffer.draw_string(x, y, text, color=..., scale=..., thickness=...)
Draws text at (x,y). color (R,G,B). scale and thickness affect size. In-place, returns self.
import visiong
img = visiong.ImageBuffer()
img.draw_string(x=10, y=10, text='hello visiong')ImageBuffer.set_text_font(font_path=..., predefine_chars=..., glyph_budget=...) · static
Configures the shared UTF-8 font used by draw_string (e.g. for Chinese).
import visiong
value = 0
visiong.ImageBuffer.set_text_font(value)ImageBuffer.clear_text_font() · static
Clears draw_string shared font configuration.
import visiong
visiong.ImageBuffer.clear_text_font()ImageBuffer.draw_cross(cx, cy, color=..., size=..., thickness=...)
Draws a cross on the image in-place and returns itself.
import visiong
img = visiong.ImageBuffer()
img.draw_cross(cx=160, cy=120)ImageBuffer.paste(img_to_paste, x, y)
Pastes another image onto this one at the specified (x, y) coordinates.
import visiong
img = visiong.ImageBuffer()
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = img.paste(img_to_paste=img2, x=10, y=10)ImageBuffer.blend(img_to_blend, x=..., y=...)
Blends an RGBA image onto this image using its alpha channel. This is a CPU operation.
import visiong
img = visiong.ImageBuffer()
img2 = visiong.ImageBuffer.create(320, 240, "bgr888", (255, 255, 255))
result = img.blend(img_to_blend=img2)ImageBuffer.__repr__()
<ImageBuffer
import visiong
img = visiong.ImageBuffer()
text = repr(img)DisplayUDP
DisplayUDP.init(ip_address, port, jpeg_quality=...)
Initializes or re-initializes the UDP sender (target IP, port, JPEG quality 1-100).
import visiong
udp = visiong.DisplayUDP()
result = udp.init(ip_address='127.0.0.1', port=8080)DisplayUDP.display(img_buf)
Encodes the ImageBuffer to JPEG (VENC) and sends it via UDP to the configured address. The DisplayUDP-local lock may convert color format and black-pad smaller frames before encoding.
import visiong
udp = visiong.DisplayUDP()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
udp.display(img_buf=img)DisplayUDP.release()
Releases DisplayUDP resources.
import visiong
udp = visiong.DisplayUDP()
udp.release()DisplayUDP.is_initialized()
Checks if DisplayUDP is initialized.
import visiong
udp = visiong.DisplayUDP()
result = udp.is_initialized()TouchPoint
TouchPoint.__repr__()
<TouchPoint x=
import visiong
touchPoint = visiong.TouchPoint()
text = repr(touchPoint)TouchPoint.x · field
readonly_field
import visiong
touchPoint = visiong.TouchPoint()
value = touchPoint.xTouchPoint.y · field
readonly_field
import visiong
touchPoint = visiong.TouchPoint()
value = touchPoint.yTouch
Touch.release()
Releases the touch device (closes I2C).
import visiong
touch = visiong.Touch()
touch.release()Touch.is_pressed()
Returns True if at least one finger is on the screen.
import visiong
touch = visiong.Touch()
result = touch.is_pressed()Touch.read()
Reads all active touch points and returns a list of TouchPoint objects.
import visiong
touch = visiong.Touch()
result = touch.read()Touch.configure_geometry(original_width, original_height, rotation_degrees)
Re-configures the screen geometry and coordinate rotation at runtime.
import visiong
touch = visiong.Touch()
result = touch.configure_geometry(original_width=640, original_height=360, rotation_degrees=270)DisplayFB
DisplayFB.display(img_buf, roi)
Displays the full image on the framebuffer (non-blocking). Returns True on success.
import visiong
fb = visiong.DisplayFB()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
fb.display(img_buf=img, roi=(0, 0, 320, 240))DisplayFB.release()
Releases framebuffer resources.
import visiong
fb = visiong.DisplayFB()
fb.release()DisplayFB.is_initialized()
Returns True if the framebuffer is initialized.
import visiong
fb = visiong.DisplayFB()
result = fb.is_initialized()DisplayFB.__repr__()
DisplayFB(screen_width={}, screen_height={})
import visiong
fb = visiong.DisplayFB()
text = repr(fb)Detection
Detection.__repr__()
<Detection label='
import visiong
detection = visiong.Detection()
text = repr(detection)Detection.box · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.boxDetection.score · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.scoreDetection.class_id · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.class_idDetection.label · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.labelDetection.landmarks · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.landmarksDetection.keypoints · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.keypointsDetection.mask_points · field
readonly_field
import visiong
detection = visiong.Detection()
value = detection.mask_pointsNPU
NPU.infer(img_buf, roi=..., model_format=...)
Runs inference. For detection/pose models (YOLOv5, RetinaFace, YOLO11, YOLO11_SEG, YOLO11_POSE) returns a list of Detection.
import visiong
npu = visiong.NPU(model_type='yolov5', model_path='/path/to/model.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = npu.infer(img_buf=img)NPU.get_face_feature(face_image)
Extracts a 128-dimensional feature vector from a cropped face image. Requires FACENET model_type; raises RuntimeError otherwise.
import visiong
npu = visiong.NPU(model_type='yolov5', model_path='/path/to/model.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = npu.get_face_feature(face_image=img)NPU.recognize_plate(plate_image)
Recognizes a license plate from a cropped image. For use with LPRNET models. Returns a string.
import visiong
npu = visiong.NPU(model_type='yolov5', model_path='/path/to/model.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = npu.recognize_plate(plate_image=img)NPU.get_feature_distance(feature1, feature2) · static
Euclidean distance between two 128-D face feature vectors. Returns 100.0 if either length is not 128.
import visiong
feature1 = [0.0] * 512
feature2 = [0.0] * 512
result = visiong.NPU.get_feature_distance(feature1=feature1, feature2=feature2)NPU.is_initialized()
Checks if the NPU is initialized.
import visiong
npu = visiong.NPU(model_type='yolov5', model_path='/path/to/model.rknn')
result = npu.is_initialized()LowLevelTensorInfo
该类没有公开方法示例。
LowLevelTensorInfo.index · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.indexLowLevelTensorInfo.name · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.nameLowLevelTensorInfo.dims · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.dimsLowLevelTensorInfo.format · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.formatLowLevelTensorInfo.type · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.typeLowLevelTensorInfo.quant_type · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.quant_typeLowLevelTensorInfo.zero_point · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.zero_pointLowLevelTensorInfo.scale · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.scaleLowLevelTensorInfo.num_elements · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.num_elementsLowLevelTensorInfo.size_bytes · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.size_bytesLowLevelTensorInfo.size_with_stride_bytes · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.size_with_stride_bytesLowLevelTensorInfo.w_stride · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.w_strideLowLevelTensorInfo.h_stride · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.h_strideLowLevelTensorInfo.pass_through · field
readonly_field
import visiong
lowLevelTensorInfo = visiong.LowLevelTensorInfo()
value = lowLevelTensorInfo.pass_throughLowLevelNPU
LowLevelNPU.is_initialized()
is_initialized
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.is_initialized()LowLevelNPU.num_inputs()
类方法最小示例
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.num_inputs()LowLevelNPU.num_outputs()
类方法最小示例
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.num_outputs()LowLevelNPU.input_tensors()
input_tensors
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.input_tensors()LowLevelNPU.output_tensors()
output_tensors
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_tensors()LowLevelNPU.input_tensor(index)
input_tensor
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.input_tensor(index=0)LowLevelNPU.output_tensor(index)
output_tensor
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_tensor(index=0)LowLevelNPU.input_shape(index)
类方法最小示例
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.input_shape(index=0)LowLevelNPU.output_shape(index)
output_shape
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_shape(index=0)LowLevelNPU.sdk_versions()
sdk_versions
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.sdk_versions()LowLevelNPU.set_core_mask(core_mask)
Sets RKNN core mask. Accepts: 'auto', '0', '1', '2', '0_1', '0_1_2'.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
llnpu.set_core_mask(core_mask='auto')LowLevelNPU.set_input_attr(index, tensor_type, tensor_format, pass_through)
Rebinds one input tensor attr. Example: set_input_attr(0, 'uint8', 'nhwc', False).
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
llnpu.set_input_attr(index=0, tensor_type=0, tensor_format='rgb', pass_through=True)LowLevelNPU.reset_input_attr(index)
Resets an input tensor attr to its startup value.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.reset_input_attr(index=0)LowLevelNPU.set_input_bytes(index, payload, zero_pad=..., sync_to_device=...)
Writes raw bytes into an input tensor buffer.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
llnpu.set_input_bytes(index=0, payload='payload')LowLevelNPU.set_input_array(index, array, quantize_if_needed=..., zero_pad=..., sync_to_device=...)
Writes a numpy array into input tensor memory. Float arrays can be auto-quantized.
import visiong
import numpy as np
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
arr = np.zeros((1, 224, 224, 3), dtype=np.uint8)
llnpu.set_input_array(index=0, array=arr)LowLevelNPU.set_input_image(index, image, color_order=..., keep_aspect=..., pad_value=..., driver_convert=...)
Writes ImageBuffer into input tensor memory using RGA path when possible.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
llnpu.set_input_image(index=0, image=img)LowLevelNPU.sync_input_to_device(index)
sync_input_to_device
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.sync_input_to_device(index=0)LowLevelNPU.sync_output_from_device(index)
sync_output_from_device
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.sync_output_from_device(index=0)LowLevelNPU.sync_all_outputs_from_device()
sync_all_outputs_from_device
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.sync_all_outputs_from_device()LowLevelNPU.run(sync_outputs=..., non_block=..., timeout_ms=...)
Runs RKNN. Optional sync_outputs controls output cache sync.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.run(sync_outputs=True)LowLevelNPU.wait(timeout_ms=...)
类方法最小示例
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.wait(timeout_ms=0)LowLevelNPU.output_bytes(index, with_stride=..., sync_from_device=...)
Returns raw output bytes.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_bytes(index=0)LowLevelNPU.output_float(index, dequantize_if_needed=..., sync_from_device=...)
Returns output tensor as float32 numpy array (dequantized when possible).
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_float(index=0)LowLevelNPU.output_array(index, dequantize_if_needed=..., sync_from_device=...)
Returns output as float array (default) or raw uint8 vector when dequantize_if_needed=False.
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_array(index=0)LowLevelNPU.input_dma_fd(index)
input_dma_fd
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.input_dma_fd(index=0)LowLevelNPU.output_dma_fd(index)
output_dma_fd
import visiong
llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
result = llnpu.output_dma_fd(index=0)OCRResult
OCRResult.__repr__()
<OCRResult text='
import visiong
oCRResult = visiong.OCRResult()
text = repr(oCRResult)OCRResult.quad · field
readonly_field
import visiong
oCRResult = visiong.OCRResult()
value = oCRResult.quadOCRResult.rect · field
readonly_field
import visiong
oCRResult = visiong.OCRResult()
value = oCRResult.rectOCRResult.det_score · field
readonly_field
import visiong
oCRResult = visiong.OCRResult()
value = oCRResult.det_scoreOCRResult.text · field
readonly_field
import visiong
oCRResult = visiong.OCRResult()
value = oCRResult.textOCRResult.text_score · field
readonly_field
import visiong
oCRResult = visiong.OCRResult()
value = oCRResult.text_scorePPOCR
PPOCR.infer(img_buf)
Runs DET+REC OCR on one image and returns a list of OCRResult.
import visiong
ocr = visiong.PPOCR(det_model_path='/path/to/det.rknn', rec_model_path='/path/to/rec.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = ocr.infer(img_buf=img)PPOCR.is_initialized()
Checks whether PPOCR runtime is initialized.
import visiong
ocr = visiong.PPOCR(det_model_path='/path/to/det.rknn', rec_model_path='/path/to/rec.rknn')
result = ocr.is_initialized()NanoTrackResult
该类没有公开方法示例。
NanoTrackResult.box · field
readonly_field
import visiong
nanoTrackResult = visiong.NanoTrackResult()
value = nanoTrackResult.boxNanoTrackResult.score · field
readonly_field
import visiong
nanoTrackResult = visiong.NanoTrackResult()
value = nanoTrackResult.scoreNanoTrack
NanoTrack.init(img_buf, bbox)
Initializes tracker state with the first frame and initial bbox (x, y, w, h).
import visiong
tracker = visiong.NanoTrack(template_model=0, search_model=0, head_model='/path/to/head.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = tracker.init(img_buf=img, bbox=(60, 40, 120, 90))NanoTrack.track(img_buf)
Tracks target on a new frame and returns NanoTrackResult.
import visiong
tracker = visiong.NanoTrack(template_model=0, search_model=0, head_model='/path/to/head.rknn')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = tracker.track(img_buf=img)NanoTrack.is_initialized()
Returns True after init() succeeds.
import visiong
tracker = visiong.NanoTrack(template_model=0, search_model=0, head_model='/path/to/head.rknn')
result = tracker.is_initialized()NanoTrack.reset()
Clears tracker state. You must call init() again before track().
import visiong
tracker = visiong.NanoTrack(template_model=0, search_model=0, head_model='/path/to/head.rknn')
result = tracker.reset()Canvas
该类没有公开方法示例。
GUI
GUI.begin_frame(touch_device)
Starts a new frame. Pass Touch device or None. Call before any widgets.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.begin_frame(touch_device=None)GUI.begin_window(title, x, y, w, h, flags=...)
Begins a window. x,y,w,h are floats. Returns True if the window is visible.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.begin_window(title='Demo', x=10, y=10, w=120, h=90)GUI.end_window()
Ends the current window.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.end_window()GUI.end_frame(target_image)
Ends the frame and renders the GUI into the given ImageBuffer.
import visiong
gui = visiong.GUI(width=640, height=360)
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = gui.end_frame(target_image=img)GUI.layout_row_dynamic(height, cols=...)
Starts a row with dynamic column widths. cols: number of widgets in the row.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.layout_row_dynamic(height=360)GUI.layout_row_static(height, item_width, cols)
Starts a row with fixed widget width.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.layout_row_static(height=360, item_width=640, cols=1)GUI.layout_row_begin(format, row_height, cols)
Begins a row with custom format (e.g. percentage). Use layout_row_push then layout_row_end.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.layout_row_begin(format='bgr888', row_height=360, cols=1)GUI.layout_row_push(value)
Pushes a column size (after layout_row_begin).
import visiong
gui = visiong.GUI(width=640, height=360)
gui.layout_row_push(value=0)GUI.layout_row_end()
Ends the row started by layout_row_begin.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.layout_row_end()GUI.group_begin(title, flags=...)
Begins a group. Returns True if the group is visible. Must call group_end after.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.group_begin(title='Demo')GUI.group_end()
Ends the current group.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.group_end()GUI.label(text, align=...)
Draws a label. align: 'left', 'right', 'center'.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.label(text='hello visiong')GUI.label_wrap(text)
Draws a label with word wrapping.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.label_wrap(text='hello visiong')GUI.button(label)
Returns True if the button was clicked.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.button(label='object')GUI.slider(label, value, min, max, step)
Slider widget. Touch interaction is relative-drag based: press anywhere in range, then slide horizontally to adjust.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.slider(label='object', value=0, min=0, max=100, step=0)GUI.checkbox(label, is_checked)
Checkbox. Returns the new checked state (bool).
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.checkbox(label='object', is_checked=True)GUI.option(label, is_active)
Radio option. Returns True if this option is selected.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.option(label='object', is_active=True)GUI.edit_string(text, max_len=...)
Single-line text edit. Returns (changed: bool, text: str).
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.edit_string(text='hello visiong')GUI.progress(current, max, is_modifyable=...)
Progress bar. If modifiable, touch interaction is relative-drag based rather than absolute jump-to-position.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.progress(current=30, max=100)GUI.button_image(image)
Creates a clickable button from an ImageBuffer. Returns True if clicked.
import visiong
gui = visiong.GUI(width=640, height=360)
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
result = gui.button_image(image=img)GUI.tree_node(title, is_expanded)
Begins a tree node. Returns True if expanded. Call tree_pop when done with children.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.tree_node(title='Demo', is_expanded=True)GUI.tree_pop()
Ends the current tree node.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.tree_pop()GUI.property_int(name, value, min, max, step, inc_per_pixel=...)
Integer property control. Touch drag adjusts by horizontal delta instead of jumping to an absolute position.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.property_int(name='demo', value=0, min=0, max=100, step=0)GUI.property_float(name, value, min, max, step, inc_per_pixel=...)
Float property control. Touch drag adjusts by horizontal delta instead of jumping to an absolute position.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.property_float(name='demo', value=0, min=0, max=100, step=0)GUI.combo_begin(text, width, height)
Begins a combo box. Popup placement is touch-first and constrained within the parent window.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.combo_begin(text='hello visiong', width=640, height=360)GUI.combo_item(text)
Adds an item to the current combo. Returns True if selected.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.combo_item(text='hello visiong')GUI.combo_end()
Ends the combo box.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.combo_end()GUI.contextual_begin(width, height)
Begins a contextual menu. On touch devices this is typically opened via long press, with popup placement constrained to the parent window.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.contextual_begin(width=640, height=360)GUI.contextual_item(text)
Adds an item to the contextual menu. Returns True if clicked.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.contextual_item(text='hello visiong')GUI.contextual_end()
Ends the contextual menu.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.contextual_end()GUI.chart_begin(type, count, min_val, max_val)
Begins a chart section. Type can be 'lines' or 'columns'.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.chart_begin(type='lines', count=0, min_val=0, max_val=100)GUI.chart_push(value)
Pushes a new value to the active chart.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.chart_push(value=0)GUI.chart_end()
Ends the chart section.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.chart_end()GUI.menubar_begin()
Begins a menubar at the top of the current window.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.menubar_begin()GUI.menubar_end()
Ends the menubar section.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.menubar_end()GUI.menu_begin(label, width, height)
Begins a dropdown menu.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.menu_begin(label='object', width=640, height=360)GUI.input_is_pointer_down_in_rect(rect, primary_pointer=...)
Touch-first alias of input_is_mouse_down_in_rect. Returns True if the primary pointer is held inside the rect.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.input_is_pointer_down_in_rect(rect=(60, 40, 120, 90))GUI.input_is_pointer_dragging_in_rect()
Touch-first alias of input_is_mouse_dragging_in_rect. Returns drag state and delta, including momentum when active.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.input_is_pointer_dragging_in_rect()GUI.is_title_bar_active()
Touch-first alias of is_title_bar_pressed. Returns True while the title bar is held or being dragged.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.is_title_bar_active()GUI.get_scroll_delta_y()
Touch-first alias of get_smart_scroll_dy. Returns current scroll delta, including momentum when active.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.get_scroll_delta_y()GUI.menu_item(label)
Adds a clickable item to a menu.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.menu_item(label='object')GUI.menu_end()
Ends the menu section.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.menu_end()GUI.tooltip(text)
Shows a tooltip for the previously declared widget. On touch devices it appears on long press.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.tooltip(text='hello visiong')GUI.get_canvas()
Returns the current window's Canvas for custom drawing (stroke_line, fill_rect, draw_text, etc.).
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.get_canvas()GUI.widget_bounds(canvas)
Returns (x, y, w, h) of the last laid-out widget. Pass the canvas from get_canvas.
import visiong
gui = visiong.GUI(width=640, height=360)
canvas = visiong.ImageBuffer.create(320, 240, "bgr888", (20, 20, 20))
result = gui.widget_bounds(canvas=canvas)GUI.input_is_mouse_down_in_rect(rect, left_mouse=...)
Returns True if the primary pointer is held inside the given (x,y,w,h) rect. On touch devices this follows finger contact.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.input_is_mouse_down_in_rect(rect=(60, 40, 120, 90))GUI.window_set_focus(name)
Sets focus to the window with the given name.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.window_set_focus(name='demo')GUI.window_drag_from_pos(canvas)
Direct-manipulation window drag. On touch devices this moves the current window while dragging its title bar.
import visiong
gui = visiong.GUI(width=640, height=360)
canvas = visiong.ImageBuffer.create(320, 240, "bgr888", (20, 20, 20))
gui.window_drag_from_pos(canvas=canvas)GUI.window_set_scroll(scroll_y)
Sets the current window's vertical scroll offset.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.window_set_scroll(scroll_y=0)GUI.input_is_mouse_dragging_in_rect()
Returns (is_dragging, scroll_dy, (x,y,w,h) content_rect). On touch devices scroll_dy includes locked-axis drag and fling momentum.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.input_is_mouse_dragging_in_rect()GUI.is_title_bar_pressed()
Returns True if the current window title bar is actively held or dragged.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.is_title_bar_pressed()GUI.get_content_height()
Returns the content area height of the current layout.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.get_content_height()GUI.push_style_vec2(name, x, y)
Pushes a vec2 style. name: 'padding' or 'spacing'. Must be popped with pop_style.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.push_style_vec2(name='demo', x=10, y=10)GUI.pop_style()
Pops the last pushed style (e.g. vec2).
import visiong
gui = visiong.GUI(width=640, height=360)
gui.pop_style()GUI.get_smart_scroll_dy()
Returns touch-first scroll delta for the current window, including momentum when active.
import visiong
gui = visiong.GUI(width=640, height=360)
result = gui.get_smart_scroll_dy()GUI.stroke_line(canvas, x0, y0, x1, y1, thickness, color)
Draws a line on the canvas. color: (R,G,B,A).
import visiong
gui = visiong.GUI(width=640, height=360)
canvas = visiong.ImageBuffer.create(320, 240, "bgr888", (20, 20, 20))
result = gui.stroke_line(canvas=canvas, x0=10, y0=10, x1=120, y1=90, thickness=1, color=(0, 255, 0))GUI.stroke_rect(canvas, x, y, w, h, rounding, thickness, color)
Draws a rectangle outline. color: (R,G,B,A).
import visiong
gui = visiong.GUI(width=640, height=360)
canvas = visiong.ImageBuffer.create(320, 240, "bgr888", (20, 20, 20))
result = gui.stroke_rect(canvas=canvas, x=10, y=10, w=120, h=90, rounding=4, thickness=1, color=(0, 255, 0))GUI.fill_rect(canvas, x, y, w, h, rounding, color)
Fills a rectangle. color: (R,G,B,A).
import visiong
gui = visiong.GUI(width=640, height=360)
canvas = visiong.ImageBuffer.create(320, 240, "bgr888", (20, 20, 20))
result = gui.fill_rect(canvas=canvas, x=10, y=10, w=120, h=90, rounding=4, color=(0, 255, 0))GUI.draw_text(canvas, x, y, text, color)
Draws text at (x,y). color: (R,G,B,A).
import visiong
gui = visiong.GUI(width=640, height=360)
canvas = visiong.ImageBuffer.create(320, 240, "bgr888", (20, 20, 20))
gui.draw_text(canvas=canvas, x=10, y=10, text='hello visiong', color=(0, 255, 0))GUI.set_style_color(property_name, color)
Sets a theme color. property_name: 'text', 'header_bg', 'button_normal', 'button_hover', 'button_active'. color: (R,G,B,A).
import visiong
gui = visiong.GUI(width=640, height=360)
gui.set_style_color(property_name='name', color=(0, 255, 0))GUI.set_style_button_rounding(rounding)
Sets button corner rounding radius.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.set_style_button_rounding(rounding=4)GUI.set_style_window_rounding(rounding)
Sets window corner rounding radius.
import visiong
gui = visiong.GUI(width=640, height=360)
gui.set_style_window_rounding(rounding=4)GUI.set_window_background_color(color)
Sets the current window background color. color: (R,G,B,A).
import visiong
gui = visiong.GUI(width=640, height=360)
gui.set_window_background_color(color=(0, 255, 0))PinId
PinId.__repr__()
类方法最小示例
import visiong
pinId = visiong.PinId()
text = repr(pinId)PinId.bank · field
readonly_field
import visiong
pinId = visiong.PinId()
value = pinId.bankPinId.pin · field
readonly_field
import visiong
pinId = visiong.PinId()
value = pinId.pinPinMuxRegisterInfo
PinMuxRegisterInfo.__repr__()
PinMuxRegisterInfo(domain='
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
text = repr(pinMuxRegisterInfo)PinMuxRegisterInfo.domain · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.domainPinMuxRegisterInfo.base_addr · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.base_addrPinMuxRegisterInfo.reg_offset · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.reg_offsetPinMuxRegisterInfo.absolute_addr · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.absolute_addrPinMuxRegisterInfo.bit · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.bitPinMuxRegisterInfo.width · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.widthPinMuxRegisterInfo.mask · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.maskPinMuxRegisterInfo.gpio_only · field
readonly_field
import visiong
pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
value = pinMuxRegisterInfo.gpio_onlyPinAltFunction
PinAltFunction.__repr__()
PinAltFunction(function='
import visiong
pinAltFunction = visiong.PinAltFunction()
text = repr(pinAltFunction)PinAltFunction.function · field
readonly_field
import visiong
pinAltFunction = visiong.PinAltFunction()
value = pinAltFunction.functionPinAltFunction.group · field
readonly_field
import visiong
pinAltFunction = visiong.PinAltFunction()
value = pinAltFunction.groupPinAltFunction.mux · field
readonly_field
import visiong
pinAltFunction = visiong.PinAltFunction()
value = pinAltFunction.muxPinRuntimeStatus
该类没有公开方法示例。
PinRuntimeStatus.found · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.foundPinRuntimeStatus.bank · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.bankPinRuntimeStatus.pin · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.pinPinRuntimeStatus.mux_owner · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.mux_ownerPinRuntimeStatus.gpio_owner · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.gpio_ownerPinRuntimeStatus.function · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.functionPinRuntimeStatus.group · field
readonly_field
import visiong
pinRuntimeStatus = visiong.PinRuntimeStatus()
value = pinRuntimeStatus.groupPinConflictReport
该类没有公开方法示例。
PinConflictReport.conflict · field
readonly_field
import visiong
pinConflictReport = visiong.PinConflictReport()
value = pinConflictReport.conflictPinConflictReport.reason · field
readonly_field
import visiong
pinConflictReport = visiong.PinConflictReport()
value = pinConflictReport.reasonPinConflictReport.runtime · field
readonly_field
import visiong
pinConflictReport = visiong.PinConflictReport()
value = pinConflictReport.runtimeFunctionInterfaceStatus
该类没有公开方法示例。
FunctionInterfaceStatus.request · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.requestFunctionInterfaceStatus.function · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.functionFunctionInterfaceStatus.group · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.groupFunctionInterfaceStatus.owner · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.ownerFunctionInterfaceStatus.owner_bound · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.owner_boundFunctionInterfaceStatus.interfaces · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.interfacesFunctionInterfaceStatus.note · field
readonly_field
import visiong
functionInterfaceStatus = visiong.FunctionInterfaceStatus()
value = functionInterfaceStatus.noteAdcChannelStatus
该类没有公开方法示例。
AdcChannelStatus.available · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.availableAdcChannelStatus.channel · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.channelAdcChannelStatus.raw · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.rawAdcChannelStatus.scale · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.scaleAdcChannelStatus.millivolts · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.millivoltsAdcChannelStatus.device · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.deviceAdcChannelStatus.raw_path · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.raw_pathAdcChannelStatus.scale_path · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.scale_pathAdcChannelStatus.pin_hint · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.pin_hintAdcChannelStatus.note · field
readonly_field
import visiong
adcChannelStatus = visiong.AdcChannelStatus()
value = adcChannelStatus.noteGpioLineConfig
该类没有公开方法示例。
GpioLineStatus
该类没有公开方法示例。
GpioLineStatus.requested · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.requestedGpioLineStatus.value · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.valueGpioLineStatus.bank · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.bankGpioLineStatus.pin · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.pinGpioLineStatus.gpiochip · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.gpiochipGpioLineStatus.config · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.configGpioLineStatus.note · field
readonly_field
import visiong
gpioLineStatus = visiong.GpioLineStatus()
value = gpioLineStatus.noteDriveStrengthStatus
该类没有公开方法示例。
DriveStrengthStatus.available · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.availableDriveStrengthStatus.level · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.levelDriveStrengthStatus.raw · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.rawDriveStrengthStatus.reg_offset · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.reg_offsetDriveStrengthStatus.absolute_addr · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.absolute_addrDriveStrengthStatus.domain · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.domainDriveStrengthStatus.note · field
readonly_field
import visiong
driveStrengthStatus = visiong.DriveStrengthStatus()
value = driveStrengthStatus.notePullStatus
该类没有公开方法示例。
PullStatus.available · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.availablePullStatus.mode · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.modePullStatus.raw · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.rawPullStatus.reg_offset · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.reg_offsetPullStatus.absolute_addr · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.absolute_addrPullStatus.domain · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.domainPullStatus.note · field
readonly_field
import visiong
pullStatus = visiong.PullStatus()
value = pullStatus.noteSchmittStatus
该类没有公开方法示例。
SchmittStatus.available · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.availableSchmittStatus.enabled · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.enabledSchmittStatus.raw · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.rawSchmittStatus.reg_offset · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.reg_offsetSchmittStatus.absolute_addr · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.absolute_addrSchmittStatus.domain · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.domainSchmittStatus.note · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.noteSchmittStatus.bank · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.bankSchmittStatus.pin · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.pinSchmittStatus.drive_supported · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.drive_supportedSchmittStatus.pull_supported · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.pull_supportedSchmittStatus.schmitt_supported · field
readonly_field
import visiong
schmittStatus = visiong.SchmittStatus()
value = schmittStatus.schmitt_supportedPinMux
PinMux.is_open()
Returns True if memory mappings are active.
import visiong
pm = visiong.PinMux()
result = pm.is_open()PinMux.close()
Closes /dev/mem mappings.
import visiong
pm = visiong.PinMux()
pm.close()PinMux.parse_pin(pin_name)
Parses a pin string like 'GPIO1_C4', 'gpio1-20', or '1:20'.
import visiong
pm = visiong.PinMux()
result = pm.parse_pin(pin_name='GPIO0_C3')PinMux.get_mux(bank, pin)
Reads current mux value from register field.
import visiong
pm = visiong.PinMux()
result = pm.get_mux(bank=0, pin=0)PinMux.set_mux(bank, pin, mux)
Writes mux value using Rockchip write-mask semantics (no reboot required).
import visiong
pm = visiong.PinMux()
pm.set_mux(bank=0, pin=0, mux=0)PinMux.get_register_info(bank, pin)
Returns register address/bitfield info used for this pin.
import visiong
pm = visiong.PinMux()
result = pm.get_register_info(bank=0, pin=0)PinMux.list_functions(bank, pin)
Lists available alternate functions by parsing /proc/device-tree/pinctrl.
import visiong
pm = visiong.PinMux()
result = pm.list_functions(bank=0, pin=0)PinMux.get_runtime_status(bank, pin)
Reads mux/gpio owner and current function/group from debugfs pinctrl.
import visiong
pm = visiong.PinMux()
result = pm.get_runtime_status(bank=0, pin=0)PinMux.check_conflict(bank, pin, target_function_or_group=...)
Checks whether switching this pin may conflict with current mux/gpio owners.
import visiong
pm = visiong.PinMux()
result = pm.check_conflict(bank=0, pin=0)PinMux.release_conflict(bank, pin)
Attempts to unbind current mux owner device. Returns False if release is incomplete.
import visiong
pm = visiong.PinMux()
pm.release_conflict(bank=0, pin=0)PinMux.get_interface_status(function_or_group)
Reports whether Linux has exposed usable interfaces (/dev/* or /sys/class/*) for the function.
import visiong
pm = visiong.PinMux()
result = pm.get_interface_status(function_or_group='uart4')PinMux.ensure_interface(function_or_group)
Attempts to bind the inferred owner device and re-check userspace interface visibility.
import visiong
pm = visiong.PinMux()
result = pm.ensure_interface(function_or_group='uart4')PinMux.list_overlays()
Lists currently active device-tree overlays from configfs.
import visiong
pm = visiong.PinMux()
result = pm.list_overlays()PinMux.apply_overlay(dtbo_path, overlay_name=...)
Applies a DT overlay (.dtbo) through configfs and returns created overlay entry name.
import visiong
pm = visiong.PinMux()
result = pm.apply_overlay(dtbo_path='/path/to/overlay.dtbo')PinMux.remove_overlay(overlay_name)
Removes an applied configfs overlay by name.
import visiong
pm = visiong.PinMux()
result = pm.remove_overlay(overlay_name='demo')PinMux.list_adc_channels()
Lists available SARADC channels from IIO sysfs and reads current values.
import visiong
pm = visiong.PinMux()
result = pm.list_adc_channels()PinMux.read_adc(channel)
Reads one ADC channel by numeric index.
import visiong
pm = visiong.PinMux()
result = pm.read_adc(channel=0)PinMux.gpio_request_line(bank, pin, config=...)
Requests one GPIO line with direction/bias/drive options.
import visiong
pm = visiong.PinMux()
result = pm.gpio_request_line(bank=0, pin=0)PinMux.gpio_release_line(bank, pin)
Releases a previously requested GPIO line.
import visiong
pm = visiong.PinMux()
result = pm.gpio_release_line(bank=0, pin=0)PinMux.gpio_set_value(bank, pin, value)
Sets value on a requested GPIO output line.
import visiong
pm = visiong.PinMux()
result = pm.gpio_set_value(bank=0, pin=0, value=0)PinMux.gpio_get_value(bank, pin)
Reads value from a requested GPIO line.
import visiong
pm = visiong.PinMux()
result = pm.gpio_get_value(bank=0, pin=0)PinMux.gpio_get_status(bank, pin)
Returns runtime status of requested GPIO line.
import visiong
pm = visiong.PinMux()
result = pm.gpio_get_status(bank=0, pin=0)PinMux.set_drive_strength(bank, pin, level)
Sets RV1106 IOC drive strength level (0..7) for a pin.
import visiong
pm = visiong.PinMux()
pm.set_drive_strength(bank=0, pin=0, level=3)PinMux.get_drive_strength(bank, pin)
Reads RV1106 IOC drive strength level/raw register for a pin.
import visiong
pm = visiong.PinMux()
result = pm.get_drive_strength(bank=0, pin=0)PinMux.set_pull(bank, pin, mode)
Sets pull mode (disable/pull_up/pull_down/bus_hold or 0..3).
import visiong
pm = visiong.PinMux()
pm.set_pull(bank=0, pin=0, mode='up')PinMux.get_pull(bank, pin)
Reads pull mode/raw register for a pin.
import visiong
pm = visiong.PinMux()
result = pm.get_pull(bank=0, pin=0)PinMux.set_input_schmitt(bank, pin, enable)
Enables/disables input schmitt for a pin.
import visiong
pm = visiong.PinMux()
pm.set_input_schmitt(bank=0, pin=0, enable=True)PinMux.get_input_schmitt(bank, pin)
Reads input schmitt state/raw register for a pin.
import visiong
pm = visiong.PinMux()
result = pm.get_input_schmitt(bank=0, pin=0)PinMux.probe_electrical_capability(bank, pin, active_test=...)
Probes drive/pull/schmitt capability for one pin. active_test=True performs write-restore checks.
import visiong
pm = visiong.PinMux()
result = pm.probe_electrical_capability(bank=0, pin=0)PinMux.probe_electrical_capabilities(active_test=...)
Probes drive/pull/schmitt capability for all pins.
import visiong
pm = visiong.PinMux()
result = pm.probe_electrical_capabilities()PinMux.get_function_name(bank, pin)
Returns best-effort function name matching current mux.
import visiong
pm = visiong.PinMux()
result = pm.get_function_name(bank=0, pin=0)PinMux.set_function(bank, pin, function_or_group)
Sets mux by function name (e.g. 'uart4', 'pwm1') or group name (e.g. 'uart4m1-xfer').
import visiong
pm = visiong.PinMux()
pm.set_function(bank=0, pin=0, function_or_group='uart4')NpuClockStatus
该类没有公开方法示例。
NpuClockStatus.npu_node_present · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.npu_node_presentNpuClockStatus.debugfs_available · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.debugfs_availableNpuClockStatus.overlay_configfs_available · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.overlay_configfs_availableNpuClockStatus.assigned_rate_hz · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.assigned_rate_hzNpuClockStatus.current_rate_hz · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.current_rate_hzNpuClockStatus.npu_root_rate_hz · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.npu_root_rate_hzNpuClockStatus.clk500m_src_rate_hz · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.clk500m_src_rate_hzNpuClockStatus.note · field
readonly_field
import visiong
npuClockStatus = visiong.NpuClockStatus()
value = npuClockStatus.noteNpuClockApplyResult
该类没有公开方法示例。
NpuClockApplyResult.ok · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.okNpuClockApplyResult.rebind_attempted · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.rebind_attemptedNpuClockApplyResult.rebind_ok · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.rebind_okNpuClockApplyResult.reboot_required · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.reboot_requiredNpuClockApplyResult.requested_rate_hz · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.requested_rate_hzNpuClockApplyResult.assigned_rate_hz · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.assigned_rate_hzNpuClockApplyResult.current_rate_hz · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.current_rate_hzNpuClockApplyResult.npu_root_rate_hz · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.npu_root_rate_hzNpuClockApplyResult.clk500m_src_rate_hz · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.clk500m_src_rate_hzNpuClockApplyResult.overlay_name · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.overlay_nameNpuClockApplyResult.message · field
readonly_field
import visiong
npuClockApplyResult = visiong.NpuClockApplyResult()
value = npuClockApplyResult.messageNpuClock
NpuClock.status()
Reads assigned/runtime NPU clock status.
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.status()NpuClock.supported_rates_hz()
Returns conservative validated NPU rates in Hz.
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.supported_rates_hz()NpuClock.supported_rates_mhz()
Returns conservative validated NPU rates in MHz.
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.supported_rates_mhz()NpuClock.list_overlays(prefix=...)
Lists active DT overlays with the given prefix.
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.list_overlays()NpuClock.remove_overlay(overlay_name)
Removes one DT overlay by name.
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.remove_overlay(overlay_name='demo')NpuClock.set_rate_hz(rate_hz, update_cru_clk500m_src=..., unbind_rebind_npu=..., allow_unsafe_rate=...)
Applies NPU assigned-clock-rates in Hz. Can optionally update CRU CLK_500M_SRC and rebind NPU driver.
import visiong
npu_clock = visiong.NpuClock()
npu_clock.set_rate_hz(rate_hz=500000000)NpuClock.set_rate_mhz(rate_mhz, update_cru_clk500m_src=..., unbind_rebind_npu=..., allow_unsafe_rate=...)
Applies NPU assigned-clock-rates in MHz.
import visiong
npu_clock = visiong.NpuClock()
npu_clock.set_rate_mhz(rate_mhz=500)NpuClock.request_reboot()
Requests immediate system reboot (sync + reboot).
import visiong
npu_clock = visiong.NpuClock()
result = npu_clock.request_reboot()DisplayHTTP
DisplayHTTP.stop()
Stops the HTTP server and disconnects all clients. This is automatically called when the object is garbage collected.
import visiong
http = visiong.DisplayHTTP()
http.stop()DisplayHTTP.display(img)
Encodes the ImageBuffer to JPEG (VENC) and pushes the frame to all connected MJPEG clients. In mode='jpg', DisplayHTTP keeps a local JPEG lock and may convert color format or black-pad smaller frames before encoding.
import visiong
http = visiong.DisplayHTTP()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
http.display(img=img)DisplayHTTP.is_running()
Returns True if the server is currently running.
import visiong
http = visiong.DisplayHTTP()
result = http.is_running()DisplayHTTP.set_fps(fps)
Sets max FPS. 0 disables limiting.
import visiong
http = visiong.DisplayHTTP()
http.set_fps(fps=30)DisplayHTTP.get_fps()
Returns current max FPS.
import visiong
http = visiong.DisplayHTTP()
result = http.get_fps()DisplayHTTP.set_quality(quality)
Sets JPEG quality (1-100).
import visiong
http = visiong.DisplayHTTP()
http.set_quality(quality=75)DisplayHTTP.get_quality()
Returns current JPEG quality (1-100).
import visiong
http = visiong.DisplayHTTP()
result = http.get_quality()DisplayRTSP
DisplayRTSP.stop()
Stops the RTSP server. This is automatically called when the object is garbage collected.
import visiong
rtsp = visiong.DisplayRTSP()
rtsp.stop()DisplayRTSP.set_fps(fps)
Sets max frames per second. 0 disables limiting.
import visiong
rtsp = visiong.DisplayRTSP()
rtsp.set_fps(fps=30)DisplayRTSP.get_fps()
Returns current max frames per second.
import visiong
rtsp = visiong.DisplayRTSP()
result = rtsp.get_fps()DisplayRTSP.set_quality(quality)
Sets encoding quality (1-100). Takes effect on the next display() call.
import visiong
rtsp = visiong.DisplayRTSP()
rtsp.set_quality(quality=75)DisplayRTSP.get_quality()
Returns current encoding quality (1-100).
import visiong
rtsp = visiong.DisplayRTSP()
result = rtsp.get_quality()DisplayRTSP.set_rc_mode(rc_mode)
Sets rate control mode: 'cbr' or 'vbr'. Takes effect on next display() call.
import visiong
rtsp = visiong.DisplayRTSP()
rtsp.set_rc_mode(rc_mode='cbr')DisplayRTSP.get_rc_mode()
Returns current rate control mode as string ('cbr' or 'vbr').
import visiong
rtsp = visiong.DisplayRTSP()
result = rtsp.get_rc_mode()DisplayRTSP.set_logs(logs)
Sets logs: 1 enable, 0 suppress.
import visiong
rtsp = visiong.DisplayRTSP()
rtsp.set_logs(logs=1)DisplayRTSP.get_logs()
Returns 1 if logs are enabled, otherwise 0.
import visiong
rtsp = visiong.DisplayRTSP()
result = rtsp.get_logs()DisplayRTSP.display(img)
Encodes the ImageBuffer and pushes the frame to all connected RTSP clients.
import visiong
rtsp = visiong.DisplayRTSP()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
rtsp.display(img=img)DisplayRTSP.is_running()
Returns True if the server is currently running.
import visiong
rtsp = visiong.DisplayRTSP()
result = rtsp.is_running()VencRecorder
VencRecorder.write(img)
Encodes and writes one frame.
import visiong
rec = visiong.VencRecorder(filepath='/path/to/demo.jpg')
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
rec.write(img=img)VencRecorder.close()
Closes and finalizes the output file (required for MP4).
import visiong
rec = visiong.VencRecorder(filepath='/path/to/demo.jpg')
rec.close()VencRecorder.is_open()
Returns True if recorder is open.
import visiong
rec = visiong.VencRecorder(filepath='/path/to/demo.jpg')
result = rec.is_open()VencRecorder.path()
Returns output filepath.
import visiong
rec = visiong.VencRecorder(filepath='/path/to/demo.jpg')
result = rec.path()DisplayHTTPFLV
DisplayHTTPFLV.stop()
Stops the server.
import visiong
flv = visiong.DisplayHTTPFLV()
flv.stop()DisplayHTTPFLV.display(img)
Encodes and pushes one frame to all connected viewers.
import visiong
flv = visiong.DisplayHTTPFLV()
img = visiong.ImageBuffer.create(320, 240, "bgr888", (0, 0, 0))
flv.display(img=img)DisplayHTTPFLV.is_running()
Returns True if server is running.
import visiong
flv = visiong.DisplayHTTPFLV()
result = flv.is_running()DisplayHTTPFLV.set_fps(fps)
Sets max FPS. 0 disables limiting.
import visiong
flv = visiong.DisplayHTTPFLV()
flv.set_fps(fps=30)DisplayHTTPFLV.get_fps()
Returns current max FPS.
import visiong
flv = visiong.DisplayHTTPFLV()
result = flv.get_fps()DisplayHTTPFLV.set_quality(quality)
Sets encoding quality 1-100.
import visiong
flv = visiong.DisplayHTTPFLV()
flv.set_quality(quality=75)DisplayHTTPFLV.get_quality()
Returns current encoding quality.
import visiong
flv = visiong.DisplayHTTPFLV()
result = flv.get_quality()DisplayHTTPFLV.set_rc_mode(rc_mode)
Sets rate control mode: 'cbr' or 'vbr'. Takes effect on next display() call.
import visiong
flv = visiong.DisplayHTTPFLV()
flv.set_rc_mode(rc_mode='cbr')DisplayHTTPFLV.get_rc_mode()
Returns current rate control mode as string ('cbr' or 'vbr').
import visiong
flv = visiong.DisplayHTTPFLV()
result = flv.get_rc_mode()8) 全量 API（逐函数调用模板）
支持搜索类名、函数名、字段名。每个模块函数、类方法、字段都提供了调用模板。
CameraIVEModelMotionVectorIVEBlobLineCircleQRCodeImageBufferDisplayUDPTouchPointTouchDisplayFBDetectionNPULowLevelTensorInfoLowLevelNPUOCRResultPPOCRNanoTrackResultNanoTrackCanvasGUIPinIdPinMuxRegisterInfoPinAltFunctionPinRuntimeStatusPinConflictReportFunctionInterfaceStatusAdcChannelStatusGpioLineConfigGpioLineStatusDriveStrengthStatusPullStatusSchmittStatusPinMuxNpuClockStatusNpuClockApplyResultNpuClockDisplayHTTPDisplayRTSPVencRecorderDisplayHTTPFLV
8.1 模块级函数
| 函数 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
close_venc_recorder |
(filepath) |
None | Closes a cached MP4 writer for filepath (finalize MP4). | visiong.close_venc_recorder('demo.mp4') |
close_all_venc_recorders |
() |
None | Closes all cached MP4 writers (finalize MP4). | visiong.close_all_venc_recorders() |
get_unique_id |
() |
对象 / 数值 | Reads the unique 6-byte chip ID from the RV1106 OTP area and returns it as a 12-character hex string. | uid = visiong.get_unique_id() |
decrypt_legacy_value |
(key) |
对象 / 数值 | Decrypts the stored legacy value with a 64-character hex key. | value = visiong.decrypt_legacy_value(key) |
dma_state_metrics |
(reset=...) |
dict / 统计对象 | Returns dma-buf state-machine counters as a dict. Set reset=True to clear counters after reading. | metrics = visiong.dma_state_metrics(reset=False) |
dma_state_reset_metrics |
() |
None | Resets dma-buf state-machine counters. | visiong.dma_state_reset_metrics() |
dma_state_dump_metrics |
(output_path=..., reset_after_dump=...) |
字符串 / 文件路径 | Returns dma-buf state-machine counters as JSON and optionally writes them to output_path. | dump = visiong.dma_state_dump_metrics(output_path='metrics.json') |
8.2 类与方法
Camera
无类级描述。
构造模板： cam = visiong.Camera(target_width=640, target_height=360)
构造签名： (target_width, target_height, format=..., hdr=..., crop_mode=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
init |
(target_width, target_height, format=..., hdr=..., crop_mode=...) |
对象 / 数值 | Initializes the camera. format defaults to 'yuv'. Supported values: 'bgr', 'rgb', 'yuv'/'yuv420', or 'gray'. crop_mode: 'auto' (default, follows the camera max-resolution aspect ratio), 'off', or any ratio such as '16:9', '4:3', '1:1', o... | result = cam.init(target_width=640, target_height=360) |
skip |
(num_frames=...) |
对象 / 数值 | (Reads and discards a specified number of frames from the camera. | result = cam.skip(10) |
snapshot |
() |
图像 / 数组对象 | Captures a single frame from the camera and returns an ImageBuffer. | result = cam.snapshot() |
release |
() |
None / 状态码 | Releases the camera and frees resources. Safe to call even if not initialized. | cam.release() |
is_initialized |
() |
bool | Returns True if the camera has been successfully initialized. | result = cam.is_initialized() |
get_capture_width |
() |
对象 / 数值 | Returns the actual capture width (alias for actual_width). | result = cam.get_capture_width() |
get_capture_height |
() |
对象 / 数值 | Returns the actual capture height (alias for actual_height). | result = cam.get_capture_height() |
set_saturation |
(value) |
None / 状态码 | Sets the image saturation. Range: [0, 255]. Raises ValueError on invalid input. | cam.set_saturation(value=0) |
set_contrast |
(value) |
None / 状态码 | Sets the image contrast. Range: [0, 255]. Raises ValueError on invalid input. | cam.set_contrast(value=0) |
set_brightness |
(value) |
None / 状态码 | Sets the image brightness. Range: [0, 255]. Raises ValueError on invalid input. | cam.set_brightness(value=0) |
set_sharpness |
(value) |
None / 状态码 | Sets the image sharpness. Range: [0, 100]. Raises ValueError on invalid input. | cam.set_sharpness(value=0) |
set_hue |
(value) |
None / 状态码 | Sets the image hue. Range: [0, 255]. Raises ValueError on invalid input. | cam.set_hue(value=0) |
set_white_balance_mode |
(mode) |
None / 状态码 | Sets white balance mode ('auto' or 'manual'). Raises ValueError on invalid mode. | cam.set_white_balance_mode(mode='auto') |
set_white_balance_temperature |
(temp) |
None / 状态码 | Sets white balance color temperature in manual mode. Raises ValueError on invalid input or RuntimeError if not in manual mode. | cam.set_white_balance_temperature(temp=4500) |
set_exposure_mode |
(mode) |
None / 状态码 | Sets exposure mode ('auto' or 'manual'). Raises ValueError on invalid mode. | cam.set_exposure_mode(mode='auto') |
set_exposure_time |
(time_in_seconds) |
None / 状态码 | Sets manual exposure time in seconds. Must be positive. Raises RuntimeError if not in manual mode. | cam.set_exposure_time(time_in_seconds=0.01) |
set_exposure_gain |
(gain) |
None / 状态码 | Sets manual exposure gain. Typical range: [0, 127]. Raises ValueError on invalid input or RuntimeError if not in manual mode. | cam.set_exposure_gain(gain=32) |
set_spatial_denoise_level |
(level) |
None / 状态码 | Sets the spatial (2D) denoise level. Range: [0, 100]. Raises ValueError on invalid input. | cam.set_spatial_denoise_level(level=3) |
set_temporal_denoise_level |
(level) |
None / 状态码 | Sets the temporal (3D) denoise level. Range: [0, 100]. Raises ValueError on invalid input. | cam.set_temporal_denoise_level(level=3) |
set_frame_rate |
(fps) |
None / 状态码 | Sets the camera frame rate. Range: [10, 60] (or 0 for auto). Raises ValueError on invalid input. | cam.set_frame_rate(fps=30) |
set_power_line_frequency |
(mode) |
None / 状态码 | Sets anti-flicker mode ('50hz', '60hz', or 'off'). Raises ValueError on invalid mode. | cam.set_power_line_frequency(mode='auto') |
set_flip |
(flip, mirror) |
None / 状态码 | Sets image flip (vertical) and mirror (horizontal). | cam.set_flip(flip=True, mirror=True) |
get_saturation |
() |
对象 / 数值 | Gets the current image saturation (0-255). | result = cam.get_saturation() |
get_contrast |
() |
对象 / 数值 | Gets the current image contrast (0-255). | result = cam.get_contrast() |
get_brightness |
() |
对象 / 数值 | Gets the current image brightness (0-255). | result = cam.get_brightness() |
get_sharpness |
() |
对象 / 数值 | Gets the current image sharpness (0-100). | result = cam.get_sharpness() |
get_hue |
() |
对象 / 数值 | Gets the current image hue (0-255). | result = cam.get_hue() |
get_white_balance_mode |
() |
对象 / 数值 | Gets the current white balance mode ('auto' or 'manual'). | result = cam.get_white_balance_mode() |
get_white_balance_temperature |
() |
对象 / 数值 | Gets the current white balance color temperature. | result = cam.get_white_balance_temperature() |
get_exposure_mode |
() |
对象 / 数值 | Gets the current exposure mode ('auto' or 'manual'). | result = cam.get_exposure_mode() |
get_exposure_time |
() |
对象 / 数值 | Gets the current exposure time in seconds. | result = cam.get_exposure_time() |
get_exposure_gain |
() |
对象 / 数值 | Gets the current exposure gain. | result = cam.get_exposure_gain() |
lock_focus |
() |
对象 / 数值 | Locks the autofocus at its current position. Prevents further automatic adjustments. | result = cam.lock_focus() |
unlock_focus |
() |
对象 / 数值 | Unlocks the autofocus, allowing it to resume its configured mode (e.g., continuous focus). | result = cam.unlock_focus() |
trigger_focus |
() |
对象 / 数值 | Performs a single, one-shot autofocus search. This is an action, not a mode. | result = cam.trigger_focus() |
set_focus_mode |
(mode) |
None / 状态码 | Sets the autofocus mode. Supported modes: | cam.set_focus_mode(mode='continuous') |
set_manual_focus |
(position) |
None / 状态码 | Moves the lens to a specific motor code position. This implicitly sets the mode to 'manual'. | cam.set_manual_focus(position=512) |
get_focus_position |
() |
对象 / 数值 | Returns the current motor code position of the lens. Returns -1 on failure. | result = cam.get_focus_position() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
IVEModel
Manages a persistent memory block for stateful IVE algorithms like GMM or background modeling.
构造模板： model = visiong.IVEModel(width=640, height=360)
构造签名： (width, height, model_size=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
MotionVector
Holds the result of a Lucas-Kanade optical flow tracking point.
构造模板： mv = visiong.MotionVector()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | <MotionVector status= | text = repr(mv) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
status |
readonly_field | value = mv.status |
mv_x |
readonly_field | value = mv.mv_x |
mv_y |
readonly_field | value = mv.mv_y |
IVE
Hardware-accelerated image processing using Rockchip IVE.
构造模板： ive = visiong.IVE()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
static set_log_enabled |
(enabled) |
None / 状态码 | Enable or disable IVE internal logs (default off; env VISIONG_IVE_LOG=1 enables). | visiong.IVE.set_log_enabled(enabled=True) |
static is_log_enabled |
() |
bool | Returns True if IVE internal logs are enabled. | result = visiong.IVE.is_log_enabled() |
filter |
(src, mask) |
对象 / 数值 | 5x5 filter. src: GRAY8. mask: 25 ints. Returns GRAY8 - do not pass to cast_16bit_to_8bit. Use a 5x5 Gaussian mask for blur (no separate gaussian_filter). | result = ive.filter(src=img, mask=[[0, -1, 0], [-1, 5, -1], [0, -1, 0]]) |
sobel |
(src, out_ctrl=..., out_format=...) |
对象 / 数值 | Sobel edge detection. src: GRAY8. Returns (horizontal_edges, vertical_edges). When out_format=S16C1 each result is S16C1 - unpack and pass to cast_16bit_to_8bit for display; when U8C1 each result is GRAY8. | result = ive.sobel(src=img) |
canny |
(src, high_thresh, low_thresh) |
对象 / 数值 | Canny edge detection. src must be GRAY8. high_thresh and low_thresh typically in 0-255. | result = ive.canny(src=img, high_thresh=120, low_thresh=40) |
mag_and_ang |
(src, threshold=..., return_magnitude=...) |
对象 / 数值 | Gradient magnitude/angle. return_magnitude=True -> S16C1 (use cast_16bit_to_8bit for display); False -> U8C1. | result = ive.mag_and_ang(src=img) |
dilate |
(src, kernel_size=...) |
对象 / 数值 | Performs morphological dilation. `src` must be GRAY8. `kernel_size` can be 3 or 5. | result = ive.dilate(src=img) |
erode |
(src, kernel_size=...) |
对象 / 数值 | Performs morphological erosion. `src` must be GRAY8. `kernel_size` can be 3 or 5. | result = ive.erode(src=img) |
ordered_stat_filter |
(src, mode) |
对象 / 数值 | Performs an ordered statistic filter (Median, Max, Min). `src` must be GRAY8. | result = ive.ordered_stat_filter(src=img, mode=visiong.OrdStatFilterMode.MEDIAN) |
add |
(src1, src2) |
对象 / 数值 | Pixel-wise addition. Returns GRAY8; do not pass to cast_16bit_to_8bit. | result = ive.add(src1=img1, src2=img2) |
sub |
(src1, src2, mode=...) |
对象 / 数值 | Pixel-wise subtraction. Returns GRAY8; do not pass to cast_16bit_to_8bit. Mode: ABS or SHIFT. | result = ive.sub(src1=img1, src2=img2) |
logic_op |
(src1, src2, op) |
对象 / 数值 | Performs a pixel-wise logical operation (AND, OR, XOR) on two images. | result = ive.logic_op(src1=img1, src2=img2, op=visiong.LogicOp.AND) |
threshold |
(src) |
对象 / 数值 | Thresholding on GRAY8. kwargs accept low/low_thresh, high/high_thresh, mode. Returns GRAY8; do not pass to cast_16bit_to_8bit. | result = ive.threshold(src=img) |
cast_16bit_to_8bit |
(src, mode) |
对象 / 数值 | Casts 16-bit (S16C1/U16C1) to 8-bit. Input must be from sobel(S16C1), mag_and_ang(magnitude), norm_grad, or sad (first element); not from filter, add, sub, threshold, gmm. | result = ive.cast_16bit_to_8bit(src=img, mode=visiong.Cast16to8Mode.S16_TO_U8_ABS) |
hist |
(src) |
对象 / 数值 | Calculates the histogram of a GRAY8 image. Returns a list of 256 integers. | result = ive.hist(src=img) |
equalize_hist |
(src) |
对象 / 数值 | Performs histogram equalization on a GRAY8 image. | result = ive.equalize_hist(src=img) |
integral |
(src, mode=...) |
对象 / 数值 | Integral image. mode: COMBINE, SUM, or SQSUM. Output U64C1. | result = ive.integral(src=img) |
ccl |
(src, min_area=...) |
对象 / 数值 | Connected Components Labeling on a binarized GRAY8 image. Blobs with area < min_area are filtered. Returns list of Blob. | result = ive.ccl(src=img) |
ncc |
(src1, src2) |
对象 / 数值 | Normalized cross-correlation. If sizes differ, set auto_resize=True (default) to resize src2 to src1 size. | result = ive.ncc(src1=img1, src2=img2) |
csc |
(src, mode) |
对象 / 数值 | Performs Color Space Conversion (e.g., YUV to RGB). | result = ive.csc(src=img, mode=visiong.CscMode.YUV2RGB_BT601_LIMITED) |
yuv_to_rgb |
(src) |
对象 / 数值 | Converts YUV420SP (or YUV420SP_VU) to RGB. Use full_range=True/False, or pass mode=visiong.CscMode.YUV2RGB_*. | result = ive.yuv_to_rgb(src=img) |
yuv_to_hsv |
(src) |
对象 / 数值 | Converts YUV420SP (or YUV420SP_VU) to HSV. Use full_range=True/False, or pass mode=visiong.CscMode.YUV2HSV_*. | result = ive.yuv_to_hsv(src=img) |
rgb_to_yuv |
(src, full_range=...) |
对象 / 数值 | Converts RGB/BGR image to YUV420SP. | result = ive.rgb_to_yuv(src=img) |
rgb_to_hsv |
(src, full_range=...) |
对象 / 数值 | Converts RGB/BGR image to HSV. H:[0,180], S:[0,255], V:[0,255]. | result = ive.rgb_to_hsv(src=img) |
dma |
(src, mode=...) |
对象 / 数值 | Performs a direct memory access operation (e.g., copy). | result = ive.dma(src=img) |
cast_8bit_to_8bit |
(src, bias, numerator, denominator) |
对象 / 数值 | Scales an 8-bit image using the formula: dst = (src * numerator / denominator) + bias. | result = ive.cast_8bit_to_8bit(src=img, bias=0, numerator=1, denominator=1) |
map |
(src, lut) |
对象 / 数值 | Pixel LUT mapping. lut: list of 256 integers in [0,255]; output[x] = lut[src[x]]. | result = ive.map(src=img, lut=lut) |
gmm |
(src, model, first_frame=...) |
对象 / 数值 | GMM background subtraction. Returns (foreground, background); both GRAY8 - do not pass to cast_16bit_to_8bit. | result = ive.gmm(src=img, model=model) |
gmm2 |
(src, factor, model, first_frame=...) |
对象 / 数值 | GMM2 with factor image. Returns (foreground, background); both GRAY8. | result = ive.gmm2(src=img, factor=0.5, model=model) |
lbp |
(src, abs_mode=..., threshold=...) |
对象 / 数值 | Calculates the Local Binary Pattern of an image. | result = ive.lbp(src=img) |
norm_grad |
(src) |
对象 / 数值 | Normalized gradient. Returns (horizontal_grad, vertical_grad); both S16C1 - use cast_16bit_to_8bit for display. | result = ive.norm_grad(src=img) |
lk_optical_flow |
(prev_img, next_img, points) |
对象 / 数值 | Performs Lucas-Kanade optical flow. points: list of (x,y). Returns list of MotionVector; mv_x/mv_y are S9.7 (divide by 128 for pixels). Up to ~500 points. | result = ive.lk_optical_flow(prev_img=img, next_img=img, points=points) |
st_corner |
(src, max_corners=..., min_dist=..., quality_level=...) |
对象 / 数值 | Performs Shi-Tomasi corner detection. Returns a list of (x, y) corner tuples. | result = ive.st_corner(src=img) |
match_bg_model |
(current_img, bg_model, frame_num) |
对象 / 数值 | Matches the current image against a background model. Returns a foreground flag image. | result = ive.match_bg_model(current_img=img, bg_model=img2, frame_num=img) |
update_bg_model |
(current_img, fg_flag, bg_model, frame_num) |
对象 / 数值 | Updates the background model. Returns the new background image. | result = ive.update_bg_model(current_img=img, fg_flag=img2, bg_model=img2, frame_num=img) |
sad |
(src1, src2, mode, threshold, min_val=..., max_val=...) |
对象 / 数值 | Sum of Absolute Differences. Returns (sad_image U16C1, threshold_image U8C1); cast sad_image for 8-bit display. | result = ive.sad(src1=img1, src2=img2, mode=visiong.SadMode.MB_8X8, threshold=120) |
create_pyramid |
(src, levels) |
对象 / 数值 | Builds an image pyramid with levels levels. Returns list of ImageBuffer from full size down. | result = ive.create_pyramid(src=img, levels=3) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
Blob
Connected component from find_blobs or IVE ccl: bounding box (x,y,w,h), center (cx,cy), pixel count, optional label code.
构造模板： blob = visiong.Blob(x=10, y=10, w=120, h=90, cx=160, cy=120, pixels=0)
构造签名： (x, y, w, h, cx, cy, pixels, code=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | <Blob rect=( | text = repr(blob) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
Line
Line segment from find_lines: endpoints (x1,y1)-(x2,y2), magnitude, theta (angle), rho.
构造模板： line = visiong.Line()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | - | text = repr(line) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
x1 |
readonly_field | value = line.x1 |
y1 |
readonly_field | value = line.y1 |
x2 |
readonly_field | value = line.x2 |
y2 |
readonly_field | value = line.y2 |
magnitude |
readonly_field | value = line.magnitude |
theta |
readonly_field | value = line.theta |
rho |
readonly_field | value = line.rho |
Circle
Circle from find_circles: center (cx,cy), radius r, magnitude.
构造模板： circle = visiong.Circle()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | <Circle center=( | text = repr(circle) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
cx |
readonly_field | value = circle.cx |
cy |
readonly_field | value = circle.cy |
r |
readonly_field | value = circle.r |
magnitude |
readonly_field | value = circle.magnitude |
QRCode
无类级描述。
构造模板： qRCode = visiong.QRCode()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | from_numpy_zero_copy: width and height must both be even for zero-copy hardware-friendly ImageBuffer wrapping. | text = repr(qRCode) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
corners |
property_readonly | value = qRCode.corners |
payload |
property_readonly | value = qRCode.payload |
ImageBuffer
无类级描述。
构造模板： img = visiong.ImageBuffer()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
static create |
(width, height, format, color=...) |
对象 / 数值 | Creates a new ImageBuffer filled with the given color. format: e.g. 'rgb888', 'bgr888', 'gray8'. color: (R,G,B) or (R,G,B,A). | result = visiong.ImageBuffer.create(width=640, height=360, format='bgr888') |
__array__ |
(copy=..., dtype=...) |
对象 / 数值 | Allows NumPy conversion via np.asarray(img). copy=False returns a read-only array pinned to the ImageBuffer lifetime; color exports use a BGR-shaped view over the current CPU-readable storage, which may be a cached converted backing stor... | arr = np.asarray(img) |
to_numpy |
(copy=...) |
图像 / 数组对象 | Returns a NumPy array: 2D (H,W) for grayscale, 3D (H,W,3) BGR-shaped for color. copy=False returns a read-only lifetime-pinned view over the current CPU-readable storage. | result = img.to_numpy(copy=True) |
numpy_view |
() |
对象 / 数值 | Returns a read-only lifetime-pinned NumPy view over the current CPU-readable storage. | result = img.numpy_view() |
static from_numpy |
(array, format=..., copy=...) |
图像 / 数组对象 | Creates an ImageBuffer from a NumPy array. format='auto' infers: 2D or 1-channel -> GRAY8, 3-channel -> BGR888, 4-channel -> BGRA8888. copy=False uses zero-copy when the input is uint8, C-contiguous, and has even width/height. | result = visiong.ImageBuffer.from_numpy(array=arr) |
static from_numpy_zero_copy |
(array, format=...) |
图像 / 数组对象 | Strict zero-copy import from a NumPy array. Requires dtype=uint8, C-contiguous layout, and even width/height; otherwise raises instead of silently copying. | result = visiong.ImageBuffer.from_numpy_zero_copy(array=arr) |
is_valid |
() |
bool | Returns True if the buffer holds valid image data. | result = img.is_valid() |
copy |
() |
图像 / 数组对象 | Creates and returns a deep copy of the image buffer. | result = img.copy() |
static load |
(filepath) |
图像 / 数组对象 | Loads an image from a file into a new ImageBuffer. Supports JPEG, PNG, and BMP (via stb_image). | result = visiong.ImageBuffer.load(filepath='/path/to/demo.jpg') |
save |
(filepath, quality=...) |
对象 / 数值 | Saves the ImageBuffer to a file. Supports JPEG, PNG, and BMP. Quality (1-100) applies to JPEG/PNG. Uses software encoding (stb_image). | result = img.save(filepath='/path/to/demo.jpg') |
save_hsv_bin |
(filepath) |
对象 / 数值 | Converts YUV420SP image to HSV using IVE hardware and saves raw HSV binary data to the given filepath. Input must be YUV420SP or YUV420SP_VU. | result = img.save_hsv_bin(filepath='/path/to/demo.jpg') |
save_venc_jpg |
(filepath, quality=...) |
对象 / 数值 | it checks if size and format match; otherwise, it auto-initializes the encoder. | result = img.save_venc_jpg(filepath='/path/to/demo.jpg') |
save_venc_h264 |
(filepath, quality=..., rc_mode=..., fps=..., append=..., container=..., mp4_faststart=...) |
对象 / 数值 | it muxes into MP4 and caches the writer until close_venc_recorder/close_all_venc_recorders or process exit. | result = img.save_venc_h264(filepath='/path/to/demo.jpg') |
save_venc_h265 |
(filepath, quality=..., rc_mode=..., fps=..., append=..., container=..., mp4_faststart=...) |
对象 / 数值 | it muxes into MP4 and caches the writer until close_venc_recorder/close_all_venc_recorders or process exit. | result = img.save_venc_h265(filepath='/path/to/demo.jpg') |
to_format |
(new_format) |
图像 / 数组对象 | Converts the image to the given pixel format string (e.g. 'rgb888', 'bgr888', 'gray8', 'yuv420sp'). | result = img.to_format(new_format='rgb') |
to_grayscale |
() |
图像 / 数组对象 | Converts the image to GRAY8. Uses RGA for color images. | result = img.to_grayscale() |
resize |
(new_width, new_height) |
图像 / 数组对象 | Resizes the image to new_width x new_height using RGA. | result = img.resize(new_width=320, new_height=240) |
crop ×2 |
(x, y, w, h) |
图像 / 数组对象 | Crops to the region (x, y, w, h). rect_tuple: (x, y, w, h). Uses RGA. | result = img.crop(x=10, y=10, w=120, h=90) |
rotate |
(angle_degrees) |
图像 / 数组对象 | Rotates the image by 90, 180, or 270 degrees using hardware acceleration. | result = img.rotate(angle_degrees=90) |
flip |
(horizontal, vertical) |
图像 / 数组对象 | Flips the image horizontally and/or vertically using hardware acceleration. | result = img.flip(horizontal=True, vertical=False) |
find_blobs ×2 |
(thresholds, invert=..., roi=..., x_stride=..., y_stride=..., area_threshold=..., pixels_threshold=..., merge=..., margin=..., mode=..., erode_size=..., dilate_size=...) |
结果列表 / 检测对象 | (Finds blobs by color thresholds. thresholds: list of 6-tuples (H_min,H_max,S_min,S_max,V_min,V_max) for HSV; mode 0=HSV, 1=LAB. For grayscale use the overload with [(gray_min, gray_max)].) | result = img.find_blobs(thresholds=[(0, 80)]) |
find_lines ×2 |
(x, y, w, h, x_stride=..., y_stride=..., threshold=..., rho_resolution_px=..., theta_resolution_deg=..., canny_low_thresh=..., canny_high_thresh=...) |
结果列表 / 检测对象 | theta_resolution_deg | result = img.find_lines(x=10, y=10, w=120, h=90) |
find_circles ×2 |
(x, y, w, h, x_stride=..., y_stride=..., threshold=..., r_min=..., r_max=..., r_step=..., canny_low_thresh=..., canny_high_thresh=...) |
结果列表 / 检测对象 | canny_high_thresh | result = img.find_circles(x=10, y=10, w=120, h=90) |
find_polygons ×2 |
(x, y, w, h, min_area=..., max_area=..., min_sides=..., max_sides=..., accuracy=...) |
结果列表 / 检测对象 | accuracy must be 'fast', 'normal', or 'accurate'. | result = img.find_polygons(x=10, y=10, w=120, h=90) |
find_qrcodes |
() |
结果列表 / 检测对象 | Finds and decodes QR codes in the image. The image is automatically converted to grayscale if needed. | result = img.find_qrcodes() |
find_squares |
(roi=..., threshold_val, min_area=..., approx_epsilon=..., corner_sample_radius=..., corner_ratio_thresh=..., edge_check_offset=..., area_sample_points=..., area_white_thresh=..., area_morph_close_kernel_size=..., duplicate_center_thresh=..., duplicate_area_thresh=...) |
结果列表 / 检测对象 | (Finds squares in the image using a robust corner-based algorithm. | result = img.find_squares(threshold_val=120) |
binarize |
(method=..., threshold_range=..., invert=..., adaptive_block_size=..., adaptive_c=..., pre_blur_kernel_size=..., post_morph_kernel_size=...) |
对象 / 数值 | (Performs image binarization with adjustable denoising strength. | result = img.binarize(method='otsu') |
warp_perspective |
(quad, out_width, out_height) |
对象 / 数值 | (Performs a perspective warp transformation. | result = img.warp_perspective(quad=[(0, 0), (319, 0), (319, 239), (0, 239)], out_width=320, out_height=240) |
letterbox |
(target_width, target_height, color=...) |
对象 / 数值 | Scales the image to fit inside target dimensions while preserving aspect ratio, then pads with color to fill target_width x target_height. Uses RGA. | result = img.letterbox(target_width=640, target_height=360) |
draw_line |
(x0, y0, x1, y1, color=..., thickness=...) |
None / 状态码 | Draws a line on the image in-place and returns itself. | img.draw_line(x0=10, y0=10, x1=120, y1=90) |
draw_rectangle ×2 |
(x, y, w, h, color=..., thickness=..., fill=...) |
None / 状态码 | Draws a rectangle on the image in-place and returns itself. | img.draw_rectangle(x=10, y=10, w=120, h=90) |
draw_circle |
(cx, cy, radius, color=..., thickness=..., fill=...) |
None / 状态码 | Draws a circle on the image in-place and returns itself. | img.draw_circle(cx=160, cy=120, radius=30) |
draw_string |
(x, y, text, color=..., scale=..., thickness=...) |
None / 状态码 | Draws text at (x,y). color (R,G,B). scale and thickness affect size. In-place, returns self. | img.draw_string(x=10, y=10, text='hello visiong') |
static set_text_font |
(font_path=..., predefine_chars=..., glyph_budget=...) |
None / 状态码 | Configures the shared UTF-8 font used by draw_string (e.g. for Chinese). | visiong.ImageBuffer.set_text_font(value) |
static clear_text_font |
() |
对象 / 数值 | Clears draw_string shared font configuration. | visiong.ImageBuffer.clear_text_font() |
draw_cross |
(cx, cy, color=..., size=..., thickness=...) |
None / 状态码 | Draws a cross on the image in-place and returns itself. | img.draw_cross(cx=160, cy=120) |
paste |
(img_to_paste, x, y) |
对象 / 数值 | Pastes another image onto this one at the specified (x, y) coordinates. | result = img.paste(img_to_paste=img2, x=10, y=10) |
blend |
(img_to_blend, x=..., y=...) |
对象 / 数值 | Blends an RGBA image onto this image using its alpha channel. This is a CPU operation. | result = img.blend(img_to_blend=img2) |
__repr__ |
() |
对象 / 数值 | <ImageBuffer | text = repr(img) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
DisplayUDP
无类级描述。
构造模板： udp = visiong.DisplayUDP()
构造签名： (udp_ip=..., udp_port=..., jpeg_quality=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
init |
(ip_address, port, jpeg_quality=...) |
对象 / 数值 | Initializes or re-initializes the UDP sender (target IP, port, JPEG quality 1-100). | result = udp.init(ip_address='127.0.0.1', port=8080) |
display |
(img_buf) |
None / 状态码 | Encodes the ImageBuffer to JPEG (VENC) and sends it via UDP to the configured address. The DisplayUDP-local lock may convert color format and black-pad smaller frames before encoding. | udp.display(img_buf=img) |
release |
() |
None / 状态码 | Releases DisplayUDP resources. | udp.release() |
is_initialized |
() |
bool | Checks if DisplayUDP is initialized. | result = udp.is_initialized() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
TouchPoint
Represents a single touch coordinate.
构造模板： touchPoint = visiong.TouchPoint()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | <TouchPoint x= | text = repr(touchPoint) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
x |
readonly_field | value = touchPoint.x |
y |
readonly_field | value = touchPoint.y |
Touch
Interface for I2C touch screen devices.
构造模板： touch = visiong.Touch()
构造签名： (chip_model=..., i2c_bus=..., original_width=..., original_height=..., rotation_degrees=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
release |
() |
None / 状态码 | Releases the touch device (closes I2C). | touch.release() |
is_pressed |
() |
bool | Returns True if at least one finger is on the screen. | result = touch.is_pressed() |
read |
() |
对象 / 数值 | Reads all active touch points and returns a list of TouchPoint objects. | result = touch.read() |
configure_geometry |
(original_width, original_height, rotation_degrees) |
对象 / 数值 | Re-configures the screen geometry and coordinate rotation at runtime. | result = touch.configure_geometry(original_width=640, original_height=360, rotation_degrees=270) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
DisplayFB
Framebuffer display. mode: 'high' (default) or 'low' refresh.
构造模板： fb = visiong.DisplayFB()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
display ×2 |
(img_buf, roi) |
None / 状态码 | Displays the full image on the framebuffer (non-blocking). Returns True on success. | fb.display(img_buf=img, roi=(0, 0, 320, 240)) |
release |
() |
None / 状态码 | Releases framebuffer resources. | fb.release() |
is_initialized |
() |
bool | Returns True if the framebuffer is initialized. | result = fb.is_initialized() |
__repr__ |
() |
对象 / 数值 | DisplayFB(screen_width={}, screen_height={}) | text = repr(fb) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
Detection
Single detection from YOLOv5/RetinaFace/YOLO11/YOLO11_SEG/YOLO11_POSE inference.
构造模板： detection = visiong.Detection()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | <Detection label=' | text = repr(detection) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
box |
readonly_field | value = detection.box |
score |
readonly_field | value = detection.score |
class_id |
readonly_field | value = detection.class_id |
label |
readonly_field | value = detection.label |
landmarks |
readonly_field | value = detection.landmarks |
keypoints |
readonly_field | value = detection.keypoints |
mask_points |
readonly_field | value = detection.mask_points |
NPU
无类级描述。
构造模板： npu = visiong.NPU(model_type='yolov5', model_path='/path/to/model.rknn')
构造签名： (model_type, model_path, label_path=..., box=..., nms=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
infer |
(img_buf, roi=..., model_format=...) |
结果列表 / 检测对象 | Runs inference. For detection/pose models (YOLOv5, RetinaFace, YOLO11, YOLO11_SEG, YOLO11_POSE) returns a list of Detection. | result = npu.infer(img_buf=img) |
get_face_feature |
(face_image) |
对象 / 数值 | Extracts a 128-dimensional feature vector from a cropped face image. Requires FACENET model_type; raises RuntimeError otherwise. | result = npu.get_face_feature(face_image=img) |
recognize_plate |
(plate_image) |
对象 / 数值 | Recognizes a license plate from a cropped image. For use with LPRNET models. Returns a string. | result = npu.recognize_plate(plate_image=img) |
static get_feature_distance |
(feature1, feature2) |
对象 / 数值 | Euclidean distance between two 128-D face feature vectors. Returns 100.0 if either length is not 128. | result = visiong.NPU.get_feature_distance(feature1=feature1, feature2=feature2) |
is_initialized |
() |
bool | Checks if the NPU is initialized. | result = npu.is_initialized() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
LowLevelTensorInfo
Low-level RKNN tensor metadata.
构造模板： lowLevelTensorInfo = visiong.LowLevelTensorInfo()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
index |
readonly_field | value = lowLevelTensorInfo.index |
name |
readonly_field | value = lowLevelTensorInfo.name |
dims |
readonly_field | value = lowLevelTensorInfo.dims |
format |
readonly_field | value = lowLevelTensorInfo.format |
type |
readonly_field | value = lowLevelTensorInfo.type |
quant_type |
readonly_field | value = lowLevelTensorInfo.quant_type |
zero_point |
readonly_field | value = lowLevelTensorInfo.zero_point |
scale |
readonly_field | value = lowLevelTensorInfo.scale |
num_elements |
readonly_field | value = lowLevelTensorInfo.num_elements |
size_bytes |
readonly_field | value = lowLevelTensorInfo.size_bytes |
size_with_stride_bytes |
readonly_field | value = lowLevelTensorInfo.size_with_stride_bytes |
w_stride |
readonly_field | value = lowLevelTensorInfo.w_stride |
h_stride |
readonly_field | value = lowLevelTensorInfo.h_stride |
pass_through |
readonly_field | value = lowLevelTensorInfo.pass_through |
LowLevelNPU
Low-level RKNN runtime wrapper for teaching and custom tensor IO.
构造模板： llnpu = visiong.LowLevelNPU(model_path='/path/to/model.rknn')
构造签名： (model_path, init_flags=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
is_initialized |
() |
bool | is_initialized | result = llnpu.is_initialized() |
num_inputs |
() |
对象 / 数值 | - | result = llnpu.num_inputs() |
num_outputs |
() |
对象 / 数值 | - | result = llnpu.num_outputs() |
input_tensors |
() |
对象 / 数值 | input_tensors | result = llnpu.input_tensors() |
output_tensors |
() |
对象 / 数值 | output_tensors | result = llnpu.output_tensors() |
input_tensor |
(index) |
对象 / 数值 | input_tensor | result = llnpu.input_tensor(index=0) |
output_tensor |
(index) |
对象 / 数值 | output_tensor | result = llnpu.output_tensor(index=0) |
input_shape |
(index) |
对象 / 数值 | - | result = llnpu.input_shape(index=0) |
output_shape |
(index) |
对象 / 数值 | output_shape | result = llnpu.output_shape(index=0) |
sdk_versions |
() |
对象 / 数值 | sdk_versions | result = llnpu.sdk_versions() |
set_core_mask |
(core_mask) |
None / 状态码 | Sets RKNN core mask. Accepts: 'auto', '0', '1', '2', '0_1', '0_1_2'. | llnpu.set_core_mask(core_mask='auto') |
set_input_attr |
(index, tensor_type, tensor_format, pass_through) |
None / 状态码 | Rebinds one input tensor attr. Example: set_input_attr(0, 'uint8', 'nhwc', False). | llnpu.set_input_attr(index=0, tensor_type=0, tensor_format='rgb', pass_through=True) |
reset_input_attr |
(index) |
对象 / 数值 | Resets an input tensor attr to its startup value. | result = llnpu.reset_input_attr(index=0) |
set_input_bytes |
(index, payload, zero_pad=..., sync_to_device=...) |
None / 状态码 | Writes raw bytes into an input tensor buffer. | llnpu.set_input_bytes(index=0, payload='payload') |
set_input_array |
(index, array, quantize_if_needed=..., zero_pad=..., sync_to_device=...) |
None / 状态码 | Writes a numpy array into input tensor memory. Float arrays can be auto-quantized. | llnpu.set_input_array(index=0, array=arr) |
set_input_image |
(index, image, color_order=..., keep_aspect=..., pad_value=..., driver_convert=...) |
None / 状态码 | Writes ImageBuffer into input tensor memory using RGA path when possible. | llnpu.set_input_image(index=0, image=img) |
sync_input_to_device |
(index) |
对象 / 数值 | sync_input_to_device | result = llnpu.sync_input_to_device(index=0) |
sync_output_from_device |
(index) |
对象 / 数值 | sync_output_from_device | result = llnpu.sync_output_from_device(index=0) |
sync_all_outputs_from_device |
() |
对象 / 数值 | sync_all_outputs_from_device | result = llnpu.sync_all_outputs_from_device() |
run |
(sync_outputs=..., non_block=..., timeout_ms=...) |
执行结果 | Runs RKNN. Optional sync_outputs controls output cache sync. | result = llnpu.run(sync_outputs=True) |
wait |
(timeout_ms=...) |
执行结果 | - | result = llnpu.wait(timeout_ms=0) |
output_bytes |
(index, with_stride=..., sync_from_device=...) |
对象 / 数值 | Returns raw output bytes. | result = llnpu.output_bytes(index=0) |
output_float |
(index, dequantize_if_needed=..., sync_from_device=...) |
对象 / 数值 | Returns output tensor as float32 numpy array (dequantized when possible). | result = llnpu.output_float(index=0) |
output_array |
(index, dequantize_if_needed=..., sync_from_device=...) |
对象 / 数值 | Returns output as float array (default) or raw uint8 vector when dequantize_if_needed=False. | result = llnpu.output_array(index=0) |
input_dma_fd |
(index) |
对象 / 数值 | input_dma_fd | result = llnpu.input_dma_fd(index=0) |
output_dma_fd |
(index) |
对象 / 数值 | output_dma_fd | result = llnpu.output_dma_fd(index=0) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
OCRResult
Single OCR result with text and quadrilateral location.
构造模板： oCRResult = visiong.OCRResult()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | <OCRResult text=' | text = repr(oCRResult) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
quad |
readonly_field | value = oCRResult.quad |
rect |
readonly_field | value = oCRResult.rect |
det_score |
readonly_field | value = oCRResult.det_score |
text |
readonly_field | value = oCRResult.text |
text_score |
readonly_field | value = oCRResult.text_score |
PPOCR
无类级描述。
构造模板： ocr = visiong.PPOCR(det_model_path='/path/to/det.rknn', rec_model_path='/path/to/rec.rknn')
构造签名： (det_model_path, rec_model_path, dict_path=..., det_threshold=..., box_threshold=..., use_dilate=..., rec_fast_model_path=..., rec_fast_max_ratio=..., rec_fast_enable_fallback=..., rec_fast_fallback_score_thresh=..., model_input_format=..., det_unclip_ratio=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
infer |
(img_buf) |
结果列表 / 检测对象 | Runs DET+REC OCR on one image and returns a list of OCRResult. | result = ocr.infer(img_buf=img) |
is_initialized |
() |
bool | Checks whether PPOCR runtime is initialized. | result = ocr.is_initialized() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
NanoTrackResult
Single frame tracking output from NanoTrack.
构造模板： nanoTrackResult = visiong.NanoTrackResult()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
box |
readonly_field | value = nanoTrackResult.box |
score |
readonly_field | value = nanoTrackResult.score |
NanoTrack
无类级描述。
构造模板： tracker = visiong.NanoTrack(template_model=0, search_model=0, head_model='/path/to/head.rknn')
构造签名： (template_model, search_model, head_model)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
init |
(img_buf, bbox) |
对象 / 数值 | Initializes tracker state with the first frame and initial bbox (x, y, w, h). | result = tracker.init(img_buf=img, bbox=(60, 40, 120, 90)) |
track |
(img_buf) |
结果列表 / 检测对象 | Tracks target on a new frame and returns NanoTrackResult. | result = tracker.track(img_buf=img) |
is_initialized |
() |
bool | Returns True after init() succeeds. | result = tracker.is_initialized() |
reset |
() |
对象 / 数值 | Clears tracker state. You must call init() again before track(). | result = tracker.reset() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
Canvas
An opaque handle to a Nuklear drawing canvas.
构造模板： canvas = visiong.Canvas()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
GUI
Nuklear GUI Manager
构造模板： gui = visiong.GUI(width=640, height=360)
构造签名： (width, height, font=..., predefine_chars=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
begin_frame |
(touch_device) |
对象 / 数值 | Starts a new frame. Pass Touch device or None. Call before any widgets. | result = gui.begin_frame(touch_device=None) |
begin_window |
(title, x, y, w, h, flags=...) |
对象 / 数值 | Begins a window. x,y,w,h are floats. Returns True if the window is visible. | result = gui.begin_window(title='Demo', x=10, y=10, w=120, h=90) |
end_window |
() |
对象 / 数值 | Ends the current window. | result = gui.end_window() |
end_frame |
(target_image) |
对象 / 数值 | Ends the frame and renders the GUI into the given ImageBuffer. | result = gui.end_frame(target_image=img) |
layout_row_dynamic |
(height, cols=...) |
对象 / 数值 | Starts a row with dynamic column widths. cols: number of widgets in the row. | gui.layout_row_dynamic(height=360) |
layout_row_static |
(height, item_width, cols) |
对象 / 数值 | Starts a row with fixed widget width. | gui.layout_row_static(height=360, item_width=640, cols=1) |
layout_row_begin |
(format, row_height, cols) |
对象 / 数值 | Begins a row with custom format (e.g. percentage). Use layout_row_push then layout_row_end. | gui.layout_row_begin(format='bgr888', row_height=360, cols=1) |
layout_row_push |
(value) |
对象 / 数值 | Pushes a column size (after layout_row_begin). | gui.layout_row_push(value=0) |
layout_row_end |
() |
对象 / 数值 | Ends the row started by layout_row_begin. | gui.layout_row_end() |
group_begin |
(title, flags=...) |
对象 / 数值 | Begins a group. Returns True if the group is visible. Must call group_end after. | gui.group_begin(title='Demo') |
group_end |
() |
对象 / 数值 | Ends the current group. | gui.group_end() |
label |
(text, align=...) |
对象 / 数值 | Draws a label. align: 'left', 'right', 'center'. | result = gui.label(text='hello visiong') |
label_wrap |
(text) |
对象 / 数值 | Draws a label with word wrapping. | result = gui.label_wrap(text='hello visiong') |
button |
(label) |
对象 / 数值 | Returns True if the button was clicked. | result = gui.button(label='object') |
slider |
(label, value, min, max, step) |
对象 / 数值 | Slider widget. Touch interaction is relative-drag based: press anywhere in range, then slide horizontally to adjust. | result = gui.slider(label='object', value=0, min=0, max=100, step=0) |
checkbox |
(label, is_checked) |
对象 / 数值 | Checkbox. Returns the new checked state (bool). | result = gui.checkbox(label='object', is_checked=True) |
option |
(label, is_active) |
对象 / 数值 | Radio option. Returns True if this option is selected. | result = gui.option(label='object', is_active=True) |
edit_string |
(text, max_len=...) |
对象 / 数值 | Single-line text edit. Returns (changed: bool, text: str). | result = gui.edit_string(text='hello visiong') |
progress |
(current, max, is_modifyable=...) |
对象 / 数值 | Progress bar. If modifiable, touch interaction is relative-drag based rather than absolute jump-to-position. | result = gui.progress(current=30, max=100) |
button_image |
(image) |
对象 / 数值 | Creates a clickable button from an ImageBuffer. Returns True if clicked. | result = gui.button_image(image=img) |
tree_node |
(title, is_expanded) |
对象 / 数值 | Begins a tree node. Returns True if expanded. Call tree_pop when done with children. | result = gui.tree_node(title='Demo', is_expanded=True) |
tree_pop |
() |
对象 / 数值 | Ends the current tree node. | result = gui.tree_pop() |
property_int |
(name, value, min, max, step, inc_per_pixel=...) |
对象 / 数值 | Integer property control. Touch drag adjusts by horizontal delta instead of jumping to an absolute position. | result = gui.property_int(name='demo', value=0, min=0, max=100, step=0) |
property_float |
(name, value, min, max, step, inc_per_pixel=...) |
对象 / 数值 | Float property control. Touch drag adjusts by horizontal delta instead of jumping to an absolute position. | result = gui.property_float(name='demo', value=0, min=0, max=100, step=0) |
combo_begin |
(text, width, height) |
对象 / 数值 | Begins a combo box. Popup placement is touch-first and constrained within the parent window. | result = gui.combo_begin(text='hello visiong', width=640, height=360) |
combo_item |
(text) |
对象 / 数值 | Adds an item to the current combo. Returns True if selected. | result = gui.combo_item(text='hello visiong') |
combo_end |
() |
对象 / 数值 | Ends the combo box. | result = gui.combo_end() |
contextual_begin |
(width, height) |
对象 / 数值 | Begins a contextual menu. On touch devices this is typically opened via long press, with popup placement constrained to the parent window. | result = gui.contextual_begin(width=640, height=360) |
contextual_item |
(text) |
对象 / 数值 | Adds an item to the contextual menu. Returns True if clicked. | result = gui.contextual_item(text='hello visiong') |
contextual_end |
() |
对象 / 数值 | Ends the contextual menu. | result = gui.contextual_end() |
chart_begin |
(type, count, min_val, max_val) |
对象 / 数值 | Begins a chart section. Type can be 'lines' or 'columns'. | result = gui.chart_begin(type='lines', count=0, min_val=0, max_val=100) |
chart_push |
(value) |
对象 / 数值 | Pushes a new value to the active chart. | result = gui.chart_push(value=0) |
chart_end |
() |
对象 / 数值 | Ends the chart section. | result = gui.chart_end() |
menubar_begin |
() |
对象 / 数值 | Begins a menubar at the top of the current window. | result = gui.menubar_begin() |
menubar_end |
() |
对象 / 数值 | Ends the menubar section. | result = gui.menubar_end() |
menu_begin |
(label, width, height) |
对象 / 数值 | Begins a dropdown menu. | gui.menu_begin(label='object', width=640, height=360) |
input_is_pointer_down_in_rect |
(rect, primary_pointer=...) |
对象 / 数值 | Touch-first alias of input_is_mouse_down_in_rect. Returns True if the primary pointer is held inside the rect. | result = gui.input_is_pointer_down_in_rect(rect=(60, 40, 120, 90)) |
input_is_pointer_dragging_in_rect |
() |
对象 / 数值 | Touch-first alias of input_is_mouse_dragging_in_rect. Returns drag state and delta, including momentum when active. | result = gui.input_is_pointer_dragging_in_rect() |
is_title_bar_active |
() |
bool | Touch-first alias of is_title_bar_pressed. Returns True while the title bar is held or being dragged. | result = gui.is_title_bar_active() |
get_scroll_delta_y |
() |
对象 / 数值 | Touch-first alias of get_smart_scroll_dy. Returns current scroll delta, including momentum when active. | result = gui.get_scroll_delta_y() |
menu_item |
(label) |
对象 / 数值 | Adds a clickable item to a menu. | gui.menu_item(label='object') |
menu_end |
() |
对象 / 数值 | Ends the menu section. | gui.menu_end() |
tooltip |
(text) |
对象 / 数值 | Shows a tooltip for the previously declared widget. On touch devices it appears on long press. | result = gui.tooltip(text='hello visiong') |
get_canvas |
() |
对象 / 数值 | Returns the current window's Canvas for custom drawing (stroke_line, fill_rect, draw_text, etc.). | result = gui.get_canvas() |
widget_bounds |
(canvas) |
对象 / 数值 | Returns (x, y, w, h) of the last laid-out widget. Pass the canvas from get_canvas. | result = gui.widget_bounds(canvas=canvas) |
input_is_mouse_down_in_rect |
(rect, left_mouse=...) |
对象 / 数值 | Returns True if the primary pointer is held inside the given (x,y,w,h) rect. On touch devices this follows finger contact. | result = gui.input_is_mouse_down_in_rect(rect=(60, 40, 120, 90)) |
window_set_focus |
(name) |
对象 / 数值 | Sets focus to the window with the given name. | gui.window_set_focus(name='demo') |
window_drag_from_pos |
(canvas) |
对象 / 数值 | Direct-manipulation window drag. On touch devices this moves the current window while dragging its title bar. | gui.window_drag_from_pos(canvas=canvas) |
window_set_scroll |
(scroll_y) |
对象 / 数值 | Sets the current window's vertical scroll offset. | gui.window_set_scroll(scroll_y=0) |
input_is_mouse_dragging_in_rect |
() |
对象 / 数值 | Returns (is_dragging, scroll_dy, (x,y,w,h) content_rect). On touch devices scroll_dy includes locked-axis drag and fling momentum. | result = gui.input_is_mouse_dragging_in_rect() |
is_title_bar_pressed |
() |
bool | Returns True if the current window title bar is actively held or dragged. | result = gui.is_title_bar_pressed() |
get_content_height |
() |
对象 / 数值 | Returns the content area height of the current layout. | result = gui.get_content_height() |
push_style_vec2 |
(name, x, y) |
对象 / 数值 | Pushes a vec2 style. name: 'padding' or 'spacing'. Must be popped with pop_style. | gui.push_style_vec2(name='demo', x=10, y=10) |
pop_style |
() |
对象 / 数值 | Pops the last pushed style (e.g. vec2). | gui.pop_style() |
get_smart_scroll_dy |
() |
对象 / 数值 | Returns touch-first scroll delta for the current window, including momentum when active. | result = gui.get_smart_scroll_dy() |
stroke_line |
(canvas, x0, y0, x1, y1, thickness, color) |
对象 / 数值 | Draws a line on the canvas. color: (R,G,B,A). | result = gui.stroke_line(canvas=canvas, x0=10, y0=10, x1=120, y1=90, thickness=1, color=(0, 255, 0)) |
stroke_rect |
(canvas, x, y, w, h, rounding, thickness, color) |
对象 / 数值 | Draws a rectangle outline. color: (R,G,B,A). | result = gui.stroke_rect(canvas=canvas, x=10, y=10, w=120, h=90, rounding=4, thickness=1, color=(0, 255, 0)) |
fill_rect |
(canvas, x, y, w, h, rounding, color) |
对象 / 数值 | Fills a rectangle. color: (R,G,B,A). | result = gui.fill_rect(canvas=canvas, x=10, y=10, w=120, h=90, rounding=4, color=(0, 255, 0)) |
draw_text |
(canvas, x, y, text, color) |
None / 状态码 | Draws text at (x,y). color: (R,G,B,A). | gui.draw_text(canvas=canvas, x=10, y=10, text='hello visiong', color=(0, 255, 0)) |
set_style_color |
(property_name, color) |
None / 状态码 | Sets a theme color. property_name: 'text', 'header_bg', 'button_normal', 'button_hover', 'button_active'. color: (R,G,B,A). | gui.set_style_color(property_name='name', color=(0, 255, 0)) |
set_style_button_rounding |
(rounding) |
None / 状态码 | Sets button corner rounding radius. | gui.set_style_button_rounding(rounding=4) |
set_style_window_rounding |
(rounding) |
None / 状态码 | Sets window corner rounding radius. | gui.set_style_window_rounding(rounding=4) |
set_window_background_color |
(color) |
None / 状态码 | Sets the current window background color. color: (R,G,B,A). | gui.set_window_background_color(color=(0, 255, 0)) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
PinId
Resolved pin identifier (bank + pin index).
构造模板： pinId = visiong.PinId()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | - | text = repr(pinId) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
bank |
readonly_field | value = pinId.bank |
pin |
readonly_field | value = pinId.pin |
PinMuxRegisterInfo
Raw IOMUX register field information for one pin.
构造模板： pinMuxRegisterInfo = visiong.PinMuxRegisterInfo()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | PinMuxRegisterInfo(domain=' | text = repr(pinMuxRegisterInfo) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
domain |
readonly_field | value = pinMuxRegisterInfo.domain |
base_addr |
readonly_field | value = pinMuxRegisterInfo.base_addr |
reg_offset |
readonly_field | value = pinMuxRegisterInfo.reg_offset |
absolute_addr |
readonly_field | value = pinMuxRegisterInfo.absolute_addr |
bit |
readonly_field | value = pinMuxRegisterInfo.bit |
width |
readonly_field | value = pinMuxRegisterInfo.width |
mask |
readonly_field | value = pinMuxRegisterInfo.mask |
gpio_only |
readonly_field | value = pinMuxRegisterInfo.gpio_only |
PinAltFunction
Alternative function description for one pin.
构造模板： pinAltFunction = visiong.PinAltFunction()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
__repr__ |
() |
对象 / 数值 | PinAltFunction(function=' | text = repr(pinAltFunction) |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
function |
readonly_field | value = pinAltFunction.function |
group |
readonly_field | value = pinAltFunction.group |
mux |
readonly_field | value = pinAltFunction.mux |
PinRuntimeStatus
Runtime pin ownership status from debugfs pinctrl.
构造模板： pinRuntimeStatus = visiong.PinRuntimeStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
found |
readonly_field | value = pinRuntimeStatus.found |
bank |
readonly_field | value = pinRuntimeStatus.bank |
pin |
readonly_field | value = pinRuntimeStatus.pin |
mux_owner |
readonly_field | value = pinRuntimeStatus.mux_owner |
gpio_owner |
readonly_field | value = pinRuntimeStatus.gpio_owner |
function |
readonly_field | value = pinRuntimeStatus.function |
group |
readonly_field | value = pinRuntimeStatus.group |
PinConflictReport
Pin conflict detection report.
构造模板： pinConflictReport = visiong.PinConflictReport()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
conflict |
readonly_field | value = pinConflictReport.conflict |
reason |
readonly_field | value = pinConflictReport.reason |
runtime |
readonly_field | value = pinConflictReport.runtime |
FunctionInterfaceStatus
无类级描述。
构造模板： functionInterfaceStatus = visiong.FunctionInterfaceStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
request |
readonly_field | value = functionInterfaceStatus.request |
function |
readonly_field | value = functionInterfaceStatus.function |
group |
readonly_field | value = functionInterfaceStatus.group |
owner |
readonly_field | value = functionInterfaceStatus.owner |
owner_bound |
readonly_field | value = functionInterfaceStatus.owner_bound |
interfaces |
readonly_field | value = functionInterfaceStatus.interfaces |
note |
readonly_field | value = functionInterfaceStatus.note |
AdcChannelStatus
IIO ADC channel readout status.
构造模板： adcChannelStatus = visiong.AdcChannelStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
available |
readonly_field | value = adcChannelStatus.available |
channel |
readonly_field | value = adcChannelStatus.channel |
raw |
readonly_field | value = adcChannelStatus.raw |
scale |
readonly_field | value = adcChannelStatus.scale |
millivolts |
readonly_field | value = adcChannelStatus.millivolts |
device |
readonly_field | value = adcChannelStatus.device |
raw_path |
readonly_field | value = adcChannelStatus.raw_path |
scale_path |
readonly_field | value = adcChannelStatus.scale_path |
pin_hint |
readonly_field | value = adcChannelStatus.pin_hint |
note |
readonly_field | value = adcChannelStatus.note |
GpioLineConfig
GPIO line request options (linux gpio-v2).
构造模板： gpioLineConfig = visiong.GpioLineConfig()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
GpioLineStatus
Requested GPIO line runtime status.
构造模板： gpioLineStatus = visiong.GpioLineStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
requested |
readonly_field | value = gpioLineStatus.requested |
value |
readonly_field | value = gpioLineStatus.value |
bank |
readonly_field | value = gpioLineStatus.bank |
pin |
readonly_field | value = gpioLineStatus.pin |
gpiochip |
readonly_field | value = gpioLineStatus.gpiochip |
config |
readonly_field | value = gpioLineStatus.config |
note |
readonly_field | value = gpioLineStatus.note |
DriveStrengthStatus
RV1106 IOC drive strength register status.
构造模板： driveStrengthStatus = visiong.DriveStrengthStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
available |
readonly_field | value = driveStrengthStatus.available |
level |
readonly_field | value = driveStrengthStatus.level |
raw |
readonly_field | value = driveStrengthStatus.raw |
reg_offset |
readonly_field | value = driveStrengthStatus.reg_offset |
absolute_addr |
readonly_field | value = driveStrengthStatus.absolute_addr |
domain |
readonly_field | value = driveStrengthStatus.domain |
note |
readonly_field | value = driveStrengthStatus.note |
PullStatus
RV1106 IOC pull-up/down register status.
构造模板： pullStatus = visiong.PullStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
available |
readonly_field | value = pullStatus.available |
mode |
readonly_field | value = pullStatus.mode |
raw |
readonly_field | value = pullStatus.raw |
reg_offset |
readonly_field | value = pullStatus.reg_offset |
absolute_addr |
readonly_field | value = pullStatus.absolute_addr |
domain |
readonly_field | value = pullStatus.domain |
note |
readonly_field | value = pullStatus.note |
SchmittStatus
RV1106 IOC input schmitt register status.
构造模板： schmittStatus = visiong.SchmittStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
available |
readonly_field | value = schmittStatus.available |
enabled |
readonly_field | value = schmittStatus.enabled |
raw |
readonly_field | value = schmittStatus.raw |
reg_offset |
readonly_field | value = schmittStatus.reg_offset |
absolute_addr |
readonly_field | value = schmittStatus.absolute_addr |
domain |
readonly_field | value = schmittStatus.domain |
note |
readonly_field | value = schmittStatus.note |
bank |
readonly_field | value = schmittStatus.bank |
pin |
readonly_field | value = schmittStatus.pin |
drive_supported |
readonly_field | value = schmittStatus.drive_supported |
pull_supported |
readonly_field | value = schmittStatus.pull_supported |
schmitt_supported |
readonly_field | value = schmittStatus.schmitt_supported |
PinMux
无类级描述。
构造模板： pm = visiong.PinMux()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
is_open |
() |
bool | Returns True if memory mappings are active. | result = pm.is_open() |
close |
() |
None / 状态码 | Closes /dev/mem mappings. | pm.close() |
parse_pin |
(pin_name) |
对象 / 数值 | Parses a pin string like 'GPIO1_C4', 'gpio1-20', or '1:20'. | result = pm.parse_pin(pin_name='GPIO0_C3') |
get_mux ×2 |
(bank, pin) |
对象 / 数值 | Reads current mux value from register field. | result = pm.get_mux(bank=0, pin=0) |
set_mux ×2 |
(bank, pin, mux) |
None / 状态码 | Writes mux value using Rockchip write-mask semantics (no reboot required). | pm.set_mux(bank=0, pin=0, mux=0) |
get_register_info ×2 |
(bank, pin) |
对象 / 数值 | Returns register address/bitfield info used for this pin. | result = pm.get_register_info(bank=0, pin=0) |
list_functions ×2 |
(bank, pin) |
对象 / 数值 | Lists available alternate functions by parsing /proc/device-tree/pinctrl. | result = pm.list_functions(bank=0, pin=0) |
get_runtime_status ×2 |
(bank, pin) |
对象 / 数值 | Reads mux/gpio owner and current function/group from debugfs pinctrl. | result = pm.get_runtime_status(bank=0, pin=0) |
check_conflict ×2 |
(bank, pin, target_function_or_group=...) |
对象 / 数值 | Checks whether switching this pin may conflict with current mux/gpio owners. | result = pm.check_conflict(bank=0, pin=0) |
release_conflict ×2 |
(bank, pin) |
None / 状态码 | Attempts to unbind current mux owner device. Returns False if release is incomplete. | pm.release_conflict(bank=0, pin=0) |
get_interface_status |
(function_or_group) |
对象 / 数值 | Reports whether Linux has exposed usable interfaces (/dev/* or /sys/class/*) for the function. | result = pm.get_interface_status(function_or_group='uart4') |
ensure_interface |
(function_or_group) |
对象 / 数值 | Attempts to bind the inferred owner device and re-check userspace interface visibility. | result = pm.ensure_interface(function_or_group='uart4') |
list_overlays |
() |
对象 / 数值 | Lists currently active device-tree overlays from configfs. | result = pm.list_overlays() |
apply_overlay |
(dtbo_path, overlay_name=...) |
对象 / 数值 | Applies a DT overlay (.dtbo) through configfs and returns created overlay entry name. | result = pm.apply_overlay(dtbo_path='/path/to/overlay.dtbo') |
remove_overlay |
(overlay_name) |
对象 / 数值 | Removes an applied configfs overlay by name. | result = pm.remove_overlay(overlay_name='demo') |
list_adc_channels |
() |
对象 / 数值 | Lists available SARADC channels from IIO sysfs and reads current values. | result = pm.list_adc_channels() |
read_adc ×2 |
(channel) |
对象 / 数值 | Reads one ADC channel by numeric index. | result = pm.read_adc(channel=0) |
gpio_request_line ×2 |
(bank, pin, config=...) |
对象 / 数值 | Requests one GPIO line with direction/bias/drive options. | result = pm.gpio_request_line(bank=0, pin=0) |
gpio_release_line ×2 |
(bank, pin) |
对象 / 数值 | Releases a previously requested GPIO line. | result = pm.gpio_release_line(bank=0, pin=0) |
gpio_set_value ×2 |
(bank, pin, value) |
对象 / 数值 | Sets value on a requested GPIO output line. | result = pm.gpio_set_value(bank=0, pin=0, value=0) |
gpio_get_value ×2 |
(bank, pin) |
对象 / 数值 | Reads value from a requested GPIO line. | result = pm.gpio_get_value(bank=0, pin=0) |
gpio_get_status ×2 |
(bank, pin) |
对象 / 数值 | Returns runtime status of requested GPIO line. | result = pm.gpio_get_status(bank=0, pin=0) |
set_drive_strength ×2 |
(bank, pin, level) |
None / 状态码 | Sets RV1106 IOC drive strength level (0..7) for a pin. | pm.set_drive_strength(bank=0, pin=0, level=3) |
get_drive_strength ×2 |
(bank, pin) |
对象 / 数值 | Reads RV1106 IOC drive strength level/raw register for a pin. | result = pm.get_drive_strength(bank=0, pin=0) |
set_pull ×2 |
(bank, pin, mode) |
None / 状态码 | Sets pull mode (disable/pull_up/pull_down/bus_hold or 0..3). | pm.set_pull(bank=0, pin=0, mode='up') |
get_pull ×2 |
(bank, pin) |
对象 / 数值 | Reads pull mode/raw register for a pin. | result = pm.get_pull(bank=0, pin=0) |
set_input_schmitt ×2 |
(bank, pin, enable) |
None / 状态码 | Enables/disables input schmitt for a pin. | pm.set_input_schmitt(bank=0, pin=0, enable=True) |
get_input_schmitt ×2 |
(bank, pin) |
对象 / 数值 | Reads input schmitt state/raw register for a pin. | result = pm.get_input_schmitt(bank=0, pin=0) |
probe_electrical_capability ×2 |
(bank, pin, active_test=...) |
对象 / 数值 | Probes drive/pull/schmitt capability for one pin. active_test=True performs write-restore checks. | result = pm.probe_electrical_capability(bank=0, pin=0) |
probe_electrical_capabilities |
(active_test=...) |
对象 / 数值 | Probes drive/pull/schmitt capability for all pins. | result = pm.probe_electrical_capabilities() |
get_function_name ×2 |
(bank, pin) |
对象 / 数值 | Returns best-effort function name matching current mux. | result = pm.get_function_name(bank=0, pin=0) |
set_function ×2 |
(bank, pin, function_or_group) |
None / 状态码 | Sets mux by function name (e.g. 'uart4', 'pwm1') or group name (e.g. 'uart4m1-xfer'). | pm.set_function(bank=0, pin=0, function_or_group='uart4') |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
NpuClockStatus
NPU clock probe status.
构造模板： npuClockStatus = visiong.NpuClockStatus()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
npu_node_present |
readonly_field | value = npuClockStatus.npu_node_present |
debugfs_available |
readonly_field | value = npuClockStatus.debugfs_available |
overlay_configfs_available |
readonly_field | value = npuClockStatus.overlay_configfs_available |
assigned_rate_hz |
readonly_field | value = npuClockStatus.assigned_rate_hz |
current_rate_hz |
readonly_field | value = npuClockStatus.current_rate_hz |
npu_root_rate_hz |
readonly_field | value = npuClockStatus.npu_root_rate_hz |
clk500m_src_rate_hz |
readonly_field | value = npuClockStatus.clk500m_src_rate_hz |
note |
readonly_field | value = npuClockStatus.note |
NpuClockApplyResult
NPU clock apply result.
构造模板： npuClockApplyResult = visiong.NpuClockApplyResult()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
| - | ||||
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
ok |
readonly_field | value = npuClockApplyResult.ok |
rebind_attempted |
readonly_field | value = npuClockApplyResult.rebind_attempted |
rebind_ok |
readonly_field | value = npuClockApplyResult.rebind_ok |
reboot_required |
readonly_field | value = npuClockApplyResult.reboot_required |
requested_rate_hz |
readonly_field | value = npuClockApplyResult.requested_rate_hz |
assigned_rate_hz |
readonly_field | value = npuClockApplyResult.assigned_rate_hz |
current_rate_hz |
readonly_field | value = npuClockApplyResult.current_rate_hz |
npu_root_rate_hz |
readonly_field | value = npuClockApplyResult.npu_root_rate_hz |
clk500m_src_rate_hz |
readonly_field | value = npuClockApplyResult.clk500m_src_rate_hz |
overlay_name |
readonly_field | value = npuClockApplyResult.overlay_name |
message |
readonly_field | value = npuClockApplyResult.message |
NpuClock
RV1106 NPU clock helper via DT overlay and clock readback.
构造模板： npu_clock = visiong.NpuClock()
构造签名： ()
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
status |
() |
对象 / 数值 | Reads assigned/runtime NPU clock status. | result = npu_clock.status() |
supported_rates_hz |
() |
对象 / 数值 | Returns conservative validated NPU rates in Hz. | result = npu_clock.supported_rates_hz() |
supported_rates_mhz |
() |
对象 / 数值 | Returns conservative validated NPU rates in MHz. | result = npu_clock.supported_rates_mhz() |
list_overlays |
(prefix=...) |
对象 / 数值 | Lists active DT overlays with the given prefix. | result = npu_clock.list_overlays() |
remove_overlay |
(overlay_name) |
对象 / 数值 | Removes one DT overlay by name. | result = npu_clock.remove_overlay(overlay_name='demo') |
set_rate_hz |
(rate_hz, update_cru_clk500m_src=..., unbind_rebind_npu=..., allow_unsafe_rate=...) |
None / 状态码 | Applies NPU assigned-clock-rates in Hz. Can optionally update CRU CLK_500M_SRC and rebind NPU driver. | npu_clock.set_rate_hz(rate_hz=500000000) |
set_rate_mhz |
(rate_mhz, update_cru_clk500m_src=..., unbind_rebind_npu=..., allow_unsafe_rate=...) |
None / 状态码 | Applies NPU assigned-clock-rates in MHz. | npu_clock.set_rate_mhz(rate_mhz=500) |
request_reboot |
() |
对象 / 数值 | Requests immediate system reboot (sync + reboot). | result = npu_clock.request_reboot() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
DisplayHTTP
无类级描述。
构造模板： http = visiong.DisplayHTTP()
构造签名： (port=..., quality=..., mode=..., flv_path=..., flv_codec=..., flv_fps=..., flv_rc_mode=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
stop |
() |
None / 状态码 | Stops the HTTP server and disconnects all clients. This is automatically called when the object is garbage collected. | http.stop() |
display |
(img) |
None / 状态码 | Encodes the ImageBuffer to JPEG (VENC) and pushes the frame to all connected MJPEG clients. In mode='jpg', DisplayHTTP keeps a local JPEG lock and may convert color format or black-pad smaller frames before encoding. | http.display(img=img) |
is_running |
() |
bool | Returns True if the server is currently running. | result = http.is_running() |
set_fps |
(fps) |
None / 状态码 | Sets max FPS. 0 disables limiting. | http.set_fps(fps=30) |
get_fps |
() |
对象 / 数值 | Returns current max FPS. | result = http.get_fps() |
set_quality |
(quality) |
None / 状态码 | Sets JPEG quality (1-100). | http.set_quality(quality=75) |
get_quality |
() |
对象 / 数值 | Returns current JPEG quality (1-100). | result = http.get_quality() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
DisplayRTSP
An RTSP streamer using hardware VENC (H264/H265). Supports multiple concurrent clients, TCP interleaved and UDP unicast transport.
构造模板： rtsp = visiong.DisplayRTSP()
构造签名： (port=..., path=..., quality=..., codec=..., fps=..., logs=..., rc_mode=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
stop |
() |
None / 状态码 | Stops the RTSP server. This is automatically called when the object is garbage collected. | rtsp.stop() |
set_fps |
(fps) |
None / 状态码 | Sets max frames per second. 0 disables limiting. | rtsp.set_fps(fps=30) |
get_fps |
() |
对象 / 数值 | Returns current max frames per second. | result = rtsp.get_fps() |
set_quality |
(quality) |
None / 状态码 | Sets encoding quality (1-100). Takes effect on the next display() call. | rtsp.set_quality(quality=75) |
get_quality |
() |
对象 / 数值 | Returns current encoding quality (1-100). | result = rtsp.get_quality() |
set_rc_mode |
(rc_mode) |
None / 状态码 | Sets rate control mode: 'cbr' or 'vbr'. Takes effect on next display() call. | rtsp.set_rc_mode(rc_mode='cbr') |
get_rc_mode |
() |
对象 / 数值 | Returns current rate control mode as string ('cbr' or 'vbr'). | result = rtsp.get_rc_mode() |
set_logs |
(logs) |
None / 状态码 | Sets logs: 1 enable, 0 suppress. | rtsp.set_logs(logs=1) |
get_logs |
() |
对象 / 数值 | Returns 1 if logs are enabled, otherwise 0. | result = rtsp.get_logs() |
display |
(img) |
None / 状态码 | Encodes the ImageBuffer and pushes the frame to all connected RTSP clients. | rtsp.display(img=img) |
is_running |
() |
bool | Returns True if the server is currently running. | result = rtsp.is_running() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
VencRecorder
Hardware VENC recorder (Annex-B raw stream or MP4 mux).
构造模板： rec = visiong.VencRecorder(filepath='/path/to/demo.jpg')
构造签名： (filepath, codec=..., container=..., quality=..., rc_mode=..., fps=..., mp4_faststart=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
write |
(img) |
None / 状态码 | Encodes and writes one frame. | rec.write(img=img) |
close |
() |
None / 状态码 | Closes and finalizes the output file (required for MP4). | rec.close() |
is_open |
() |
bool | Returns True if recorder is open. | result = rec.is_open() |
path |
() |
对象 / 数值 | Returns output filepath. | result = rec.path() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
DisplayHTTPFLV
无类级描述。
构造模板： flv = visiong.DisplayHTTPFLV()
构造签名： (port=..., path=..., quality=..., codec=..., fps=..., rc_mode=...)
Methods（逐函数调用模板）
| 方法 | 参数签名 | 返回值 | 说明 | 调用模板 |
|---|---|---|---|---|
stop |
() |
None / 状态码 | Stops the server. | flv.stop() |
display |
(img) |
None / 状态码 | Encodes and pushes one frame to all connected viewers. | flv.display(img=img) |
is_running |
() |
bool | Returns True if server is running. | result = flv.is_running() |
set_fps |
(fps) |
None / 状态码 | Sets max FPS. 0 disables limiting. | flv.set_fps(fps=30) |
get_fps |
() |
对象 / 数值 | Returns current max FPS. | result = flv.get_fps() |
set_quality |
(quality) |
None / 状态码 | Sets encoding quality 1-100. | flv.set_quality(quality=75) |
get_quality |
() |
对象 / 数值 | Returns current encoding quality. | result = flv.get_quality() |
set_rc_mode |
(rc_mode) |
None / 状态码 | Sets rate control mode: 'cbr' or 'vbr'. Takes effect on next display() call. | flv.set_rc_mode(rc_mode='cbr') |
get_rc_mode |
() |
对象 / 数值 | Returns current rate control mode as string ('cbr' or 'vbr'). | result = flv.get_rc_mode() |
Properties / Fields（字段访问模板）
| 字段 | 类型 | 访问模板 |
|---|---|---|
| - | ||
8.3 枚举
ImageTypeIVE
U8C1S16C1U16C1U64C1YUV420SPYUV422SPU8C3_PACKAGE
SobelOutCtrl
HORVERBOTH
OrdStatFilterMode
MEDIANMAXMIN
SubMode
ABSSHIFT
LogicOp
ANDORXOR
ThreshMode
BINARYTRUNCTO_MINVAL
IntegOutCtrl
COMBINESUMSQSUM
CscMode
YUV2RGB_BT601_LIMITEDYUV2RGB_BT709_LIMITEDYUV2RGB_BT601_FULLYUV2RGB_BT709_FULLYUV2HSV_BT601_LIMITEDYUV2HSV_BT709_LIMITEDYUV2HSV_BT601_FULLYUV2HSV_BT709_FULLRGB2YUV_BT601_LIMITEDRGB2YUV_BT709_LIMITEDRGB2YUV_BT601_FULLRGB2YUV_BT709_FULLRGB2HSV_BT601_LIMITEDRGB2HSV_BT709_LIMITEDRGB2HSV_BT601_FULLRGB2HSV_BT709_FULL
Cast16to8Mode
S16_TO_S8S16_TO_U8_ABSS16_TO_U8_BIASU16_TO_U8
DmaMode
DIRECT_COPYINTERVAL_COPYSET_3BYTESET_8BYTE
LbpCmpMode
NORMALABS
SadMode
MB_4X4MB_8X8MB_16X16
ModelType
YOLOV5RETINAFACEFACENETYOLO11YOLO11_SEGYOLO11_POSELPRNET
9) 常见问题
import visiong 报错找不到共享库怎么办？
优先检查 visiong.py 与 _visiong.so 是否在同一 Python 搜索路径下。其次确认设备运行时依赖库完整。
为什么某些类不存在，比如 IVE/NPU/GUI？
该能力通常由编译选项控制。请确认构建时是否启用了对应模块。
录制 MP4 结果损坏或无法播放？
通常是没有正确关闭写入器。请确保 VencRecorder.close() 被调用，或调用 close_venc_recorder/close_all_venc_recorders。
NPU 推理结果为空或质量差？
先核对 model_type、模型输入尺寸、color order（rgb/bgr）和预处理流程是否与训练/导出一致。
NumPy 零拷贝导入失败？
from_numpy_zero_copy 要求 uint8、C 连续、且宽高为偶数。任一条件不满足都会抛异常。