X-AnyLabeling系列：掩码转多边形坐标点

里清外明

于 2024-08-21 19:53:58 发布

阅读量183

点赞数 6

分类专栏： x-anylabeling 文章标签： opencv 人工智能计算机视觉

本文链接：https://blog.csdn.net/familytaijun/article/details/141401587

版权

x-anylabeling 专栏收录该内容

2 篇文章 0 订阅

订阅专栏

掩码转多边形坐标点

代码参考: X-AnyLabeling中的.\X-AnyLabeling\ tools\polygon_mask_conversion.py。这里主要拆分出掩码转成多边形坐标点存储在json文件的部分，同时增加了参数和支持带有中文名的路径。

代码示例

import os
import cv2
import json
import numpy as np


def get_image_size(image_file):
    image = cv2.imdecode(
      np.fromfile(image_file, dtype=np.uint8), 
      0
    )
    height, width = image.shape[:2]
    return width, height


def reset(version):
    custom_data = dict(
        version=version,
        flags={},
        shapes=[],
        imagePath="",
        imageData=None,
        imageHeight=-1,
        imageWidth=-1,
    )

    return custom_data

def mask_to_polygon(
  img_file, mask_file, json_file,
  version,
  epsilon_factor=0.001, 
  shape_dict=None,
  approx_len_threshold=5
):
    custom_data = reset(version)
    # binary_mask = cv2.imread(mask_file, cv2.IMREAD_GRAYSCALE)
    binary_mask = cv2.imdecode(
        np.fromfile(mask_file, dtype=np.uint8), 
        0
    )
    contours, _ = cv2.findContours(
        binary_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
    )
    image_name = os.path.basename(img_file)
    for contour in contours:
        epsilon = epsilon_factor * cv2.arcLength(contour, True)
        approx = cv2.approxPolyDP(contour, epsilon, True)
        if len(approx) < approx_len_threshold:
            print(
              f"{image_name}: \
                contour too small, len={len(approx)}"
            )
            continue
        for point in approx:
            x, y = point[0].tolist()
            shape_dict["points"].append([x, y])

        custom_data["shapes"].append(shape_dict)

    image_width, image_height = get_image_size(img_file)
    custom_data["imagePath"] = image_name
    custom_data["imageHeight"] = image_height
    custom_data["imageWidth"] = image_width

    with open(json_file, "w", encoding="utf-8") as f:
        json.dump(custom_data, f, indent=2, ensure_ascii=False)


if __name__ == "__main__":
    image_path = "./images/test.jpg"
    mask_path = "./masks/test.png"
    json_path = "./jsons/test.json"

    shape_dict = {
        "label": "test",
        "text": "test text",
        "points": [],
        "group_id": None,
        "shape_type": "polygon",
        "flags": {},
    }


    mask_to_polygon(
      image_path, 
      mask_path, 
      json_path, 
      epsilon_factor=0.001,
      version="2.4.0",
      shape_dict=shape_dict,
      approx_len_threshold=5
    )