cv2.fillPoly(): Convert the segmentation target labeled by labelme into a binarized segmentation image in png/jpg format

question

• Convert the segmentation target labeled by labelme into a binarized segmentation image in png/jpg format
• The file marked by labelme is a json file, which includes the coordinates of each point

    {
        
        
      "version": "5.0.1",
      "flags": {
        
        },
      "shapes": [
        {
        
        
          "label": "1",
          "points": [
            [
              256.90697674418607,
              175.8139534883721
            ],
            [
              251.32558139534888,
              195.34883720930233
            ],
            [
              243.4186046511628,
              219.06976744186048
            ],
            [
              229.93023255813955,
              248.8372093023256
            ],
            [
              204.34883720930233,
              313.48837209302326
            ],
            [
              128.06976744186048,
              474.8837209302326
            ],
            [
              145.27906976744185,
              476.27906976744185
            ],
            [
              219.2325581395349,
              310.69767441860466
            ],
            [
              264.34883720930236,
              266.51162790697674
            ],
            [
              270.8604651162791,
              266.51162790697674
            ],
            [
              280.6279069767442,
              249.30232558139537
            ],
            [
              277.37209302325584,
              241.86046511627907
            ],
            [
              274.58139534883725,
              215.8139534883721
            ],
            [
              266.2093023255814,
              232.09302325581396
            ],
            [
              270.8604651162791,
              241.86046511627907
            ],
            [
              257.8372093023256,
              260.4651162790698
            ],
            [
              229.46511627906978,
              291.6279069767442
            ],
            [
              255.04651162790697,
              250.2325581395349
            ],
            [
              262.95348837209303,
              238.13953488372093
            ],
            [
              277.8372093023256,
              204.6511627906977
            ],
            [
              278.30232558139534,
              191.16279069767444
            ],
            [
              288.0697674418605,
              176.27906976744185
            ],
            [
              280.6279069767442,
              171.62790697674419
            ],
            [
              268.5348837209302,
              188.8372093023256
            ],
            [
              268.06976744186045,
              207.4418604651163
            ],
            [
              249.46511627906978,
              244.18604651162792
            ],
            [
              216.4418604651163,
              300.93023255813955
            ],
            [
              239.2325581395349,
              244.6511627906977
            ],
            [
              258.30232558139534,
              198.13953488372093
            ],
            [
              256.90697674418607,
              191.62790697674419
            ],
            [
              263.8837209302326,
              174.4186046511628
            ]
          ],
          "group_id": null,
          "shape_type": "polygon",
          "flags": {
        
        }
        }
      ],
      "imagePath": "../pics/123.jpg",
      "imageData": "....",
      "imageHeight": 480,
      "imageWidth": 640
    }```
  
  
  

method

• Use the function in opencv-python: cv2.fillPoly()

  def fillPoly(img: Mat, pts, color, lineType=..., shift=..., offset=...) -> typing.Any:
  

application

• Use Python, and the opencv-python library

   import json  
   import numpy as np
   import cv2
   
   ## 获取分割点
   contours = []
   with open('a.json', 'r') as f_r:
       result = json.load(f_r)
       for shape in result['shapes']:
           ## 注意：数值的类型为 np.int32
           points = np.array(shape['points'], dtype=np.int32)
           ## 注意：分割点的格式
           points = points.reshape(len(points), 1, 2)
           contours.append(points)
           
   ## 转换为二值图像 
   img = np.zeros((480,640,3))
   for cnt in contours:
       fgmask = cv2.fillPoly(img, [cnt], color=(255, 255, 255))
   cv2.imwrite('a.jpg', fgmask) 
  

• Converted binary image