保存coco dataset注释为单一文件，并逐一显示所有图片的mask

客官°小女子只卖身不卖艺 2022-01-22 23:27 143阅读 0赞

大意：

官方的例子只显示 一张图片，我需要逐一显示，并且官方的那个JSON文件太大了，我把注释文件分开存储，每张图片一个注释文件，另行保存在一个叫coco的文件夹中，

#
    # windows version cocoapi
    # https://github.com/philferriere/cocoapi
    #
    #
    from pycocotools.coco import COCO
    import numpy as np
    import skimage.io as io
    import json
    import os
    
    import matplotlib as mpl
    mpl.use('TkAgg')
    import pylab
    import matplotlib.rcsetup as rcsetup
    pylab.rcParams['figure.figsize'] = (8.0, 10.0)
    
    #dataDir='..'
    #dataType='val2017'
    #dataDir='F:/BigData/msCoco2014'
    #dataType='val2014'
    
    dataDir='F:/BigData/msCoco2017'
    dataType='val2017'
    annFile='{}/annotations/instances_{}.json'.format(dataDir,dataType)
    
    # initialize COCO api for instance annotations
    coco=COCO(annFile)
    
    # display COCO categories and supercategories
    catIds = coco.getCatIds()
    cats = coco.loadCats(catIds)
    #print the names out
    nms=[cat['name'] for cat in cats] 
    print('COCO categories: \n{}\n'.format(' '.join(nms)))
    #print the supercat out
    nms = set([cat['supercategory'] for cat in cats])
    print('COCO supercategories: \n{}'.format(' '.join(nms)))
    
    # recursively display all images and its masks
    imgIds = coco.getImgIds()
    for id in imgIds: 
        annIds = coco.getAnnIds([id], catIds=catIds, iscrowd=None)
        anns = coco.loadAnns(annIds)
        imgIds = coco.getImgIds(imgIds = [id])
        img = coco.loadImgs(imgIds[0])[0]
        file_name_ext=img['file_name']
        (filename,extension) = os.path.splitext(file_name_ext)
        file_path = "coco/" + filename + ".json"
        data = {"annotations":anns}
        with open(file_path, 'w') as result_file:
            json.dump(data, result_file)
        I = io.imread('%s/%s/%s'%(dataDir,dataType,img['file_name']))
        mpl.pyplot.imshow(I)
        mpl.pyplot.axis('off')
        coco.showAnns(anns)

顺带再提一下coco数据集中各个参数的解释吧，一般的参数望文即可生义，只需要注意的是iscrowd，这个值为0 即表示polygon,注意，单个的对象（iscrowd=0)可能需要多个polygon来表示，比如某个对象在图像中被挡住了一部分。而iscrowd=1时，segmentation使用的就是RLE格式。

具体样式和解释参数官方文档：[https://cocodataset.org/\#format-data][https_cocodataset.org_format-data]

如果是自己定义的数据集，采用coco数据格式的话，各个id到底有什么用也是需要注意的地方。

{
      "type": "instances",
      "images": [
        {
          "file_name": "0.jpg",
          "height": 600,
          "width": 800,
          "id": 0                    ----> image_id
        }
      ],
      "categories": [
        {
          "supercategory": "none",   ----> supercategory can be anything
          "name": "date",
          "id": 0                    ----> category_id
        },
        {
          "supercategory": "none",
          "name": "hazelnut",
          "id": 2
        },
        {
          "supercategory": "none",
          "name": "fig",
          "id": 1
        }
      ],
      "annotations": [
        {
          "id": 1,                  ----> annotation id (each annotation has a unique id)
          "bbox": [
            100,
            116,
            140,
            170
          ],
          "image_id": 0,
          "segmentation": [],
          "ignore": 0,
          "area": 23800,
          "iscrowd": 0,
          "category_id": 0
        },
        {
          "id": 2,
          "bbox": [
            321,
            320,
            142,
            102
          ],
          "image_id": 0,
          "segmentation": [],
          "ignore": 0,
          "area": 14484,
          "iscrowd": 0,
          "category_id": 0
        }
      ]
    }

[https_cocodataset.org_format-data]: https://cocodataset.org/#format-data