Datasets:
Segmentation masks
Hello, I downloaded the dataset, and the annotations have the following structure:
{'id': 667968,
'category_id': 1,
'iscrowd': 0,
'segmentation': {'size': array([1170, 1654]),
'counts': ....,
'image_id': 9807,
'area': 123056,
'bbox': array([ 6, 214, 434, 339])}
Where I believe the counts string is the segmentation...But what format is this? how can I convert to points(polygon) which is more easy to use?
@abcp4
Just a quick note for parsing the masks:
• The dataset is in standard COCO format (RLE masks).
• There’s a helper script bundled with the repo—parse_rle_manga.py—that shows exactly how to decode the RLE into binary masks.
• The original script has a small path error (json/ARMS.json vs jsons/ARMS.json).
A corrected version and explanation are in Discussion #4:
https://huggingface.co/datasets/MS92/MangaSegmentation/discussions/4