使用OpenCV polyLines绘制YOLO分割掩码时报错(-215:Assertion failed)

尝试绘制YOLO分割掩码时，使用cv2.polylines触发断言错误，相关信息如下：

标注行示例 #

36 0.6158357764423077 0.814453125 0.6158357764423077 0.8095703125 0.6070381225961539 0.8095703125 0.6041055721153846 0.8115234375 0.5894428149038462 0.8154296875 0.5747800576923077 0.8125 0.5513196490384615 0.8134765625 0.5483870961538462 0.81640625 0.5923753653846154 0.818359375 0.6158357764423077 0.814453125

原代码 #

image_height, image_width, c = img.shape
isClosed = True
color = (255, 0, 0)
thickness = 2

with open(annotation_file) as f:
    for line in f:
        split_line = line.split()
        class_id = split_line[0]
        mask_shape = [float(numeric_string) for numeric_string in split_line[1:len(split_line)]]
        mask_points = []
        for i in range(0,len(mask_shape),2):
            x,y = mask_shape[i:i+2]
            mask_points.append((x * image_width, y * image_height))
        points = np.array([mask_points])
        image = cv2.polylines(img, points,
                              isClosed, color, thickness)
        break

错误信息（翻译后） #

OpenCV(4.7.0) /Users/xperience/GHA-OCV-Python/_work/opencv-python/opencv-python/opencv/modules/imgproc/src/drawing.cpp:2434: 错误: (-215:断言失败) p.checkVector(2, CV_32S) >= 0 在函数 'polylines' 中

错误原因 #

cv2.polylines对输入点集有两个硬性要求：

• 坐标必须是32位整数（CV_32S），不能是浮点数
• 点集数组形状需符合(点数, 1, 2)格式

原代码问题：

1. 计算出的坐标为浮点数（如x*image_width结果为小数）
2. 数组形状为(1, N, 2)，不符合要求格式

修正后的代码 #

import cv2
import numpy as np

# 假设img已正确读取
image_height, image_width, c = img.shape
isClosed = True
color = (255, 0, 0)
thickness = 2

with open(annotation_file) as f:
    for line in f:
        split_line = line.split()
        class_id = split_line[0]
        mask_shape = [float(numeric_string) for numeric_string in split_line[1:]]
        mask_points = []
        for i in range(0, len(mask_shape), 2):
            x, y = mask_shape[i:i+2]
            # 转换为整数像素坐标
            px = int(x * image_width)
            py = int(y * image_height)
            mask_points.append([px, py])
        # 调整数组格式和类型，适配polylines要求
        points = np.array(mask_points, np.int32).reshape((-1, 1, 2))
        # 绘制多边形
        image = cv2.polylines(img, [points], isClosed, color, thickness)
        break

关键修改点 #

• 坐标转整数：用int()将浮点坐标转为像素级整数（也可根据需求用np.round()后转int）
• 数组格式调整：通过reshape((-1,1,2))将点集转为(N,1,2)形状，同时指定类型为np.int32
• 点集传递：用[points]包裹，适配polylines接受多多边形点集列表的参数要求

内容的提问来源于stack exchange，提问作者219CID