How do I convert labelme json directory to one json file for use in MaskRCNN?

Question

I am attempting to train an object classifier on MaskRCNN and the tutorial I am following uses VGG label software in which converts the labelled data into one JSON file. I have used labelme for my data and need to prepare this for MaskRCNN.

Labelme gives a JSON file for each labelled image in this format:

{   "version": "4.6.0",   "flags": {},   "shapes": [
    {
      "label": "Green",
      "points": [
        [
          1385.6666666666665,
          2.121212121212121
        ],
        [
          1349.3030303030303,
          174.84848484848484
        ],
        [
          1400.8181818181818,
          296.06060606060606
        ],
        [
          1482.6363636363635,
          344.5454545454545
        ],
        [
          1619.0,
          338.48484848484844
        ],
        [
          1715.969696969697,
          244.54545454545453
        ],
        [
          1728.090909090909,
          120.30303030303028
        ],
        [
          1712.939393939394,
          71.81818181818181
        ],
        [
          1679.6060606060605,
          11.212121212121211
        ]
      ],
      "group_id": null,
      "shape_type": "polygon",
      "flags": {}
    },

I have a directory of images and corresponding JSON files, any help on what to do to combine. Can't get labelme_json_to_dataset to work and I believe this is the solution ?

Answer 1

you can use lableme2coco.py script in lableme repository under examples\instance_segmentation folder with this command:

python labelme2coco.py srcfiles destinationfolder --labels labels.txt

it will convert your annotation files in one json file with coco format srcfiles contains your labels with images in same folder and lables.txt contains your labels