Convert annotations RLE format (.json)

[scholltan]

hi all, I want to know how to convert annotations of my own data into the RLE format (.json). thx

[vchiley]

For one instance of an annotation:

from pycocotools import mask as maskUtils
import json

segmentation = maskUtils.encode(msk)
mskDic = {"image_id": int(image_id),
                  "category_id": category_id,
                  "segmentation": segmentation,
                  "score": float(score)}

with open('ouputFile.json', 'w') as json_file:
        json.dump(mskDic, json_file)

for more loop through the annotations and create a list of annotations then do the dump

is that what you were asking?

[meetps]

The value corresponding to the key 'segmentation' in the minival json file is a list of floats, but the maskUtils.encode(msk) returns a dictionary with keys 'counts' and 'size'. How to convert the dictionary into the segmentation list as in minival.json

[raninbowlalala]

@meetshah1995 Have you convert the segmentaion list as minival.json successfully?

[waspinator]

how would you encode masks to uncompressed RLE for use when iscrowd=1?

[ray-lee-94]

I meet the same problem. I can't dump the compressed RLE in the JSON file

[ZP-Guo]

Hi, @vchiley, Recently I use cocoapi to evaluate my results, but I do not how to get the "score" that comes from mskDic = {"image_id": int(image_id), "category_id": category_id, "segmentation": segmentation, "score": float(score)}. And my net does not output confidence or probability. I am looking forward to your reply.