search

chenhang98 / BPR

code for `Look Closer to Segment Better: Boundary Patch Refinement for Instance Segmentation`
Apache License 2.0
173 stars 23 forks source link

What exactly are the contents inside the segmentation.json file used in the inference? #26

Closed rahulvellora closed 2 years ago

rahulvellora commented 2 years ago

I can see image_id, bbox, score, category id etcc...what are those things under the key segmentation? what is size and count? Can you kindly give a clarification?

chenhang98 commented 2 years ago

If you mean the coarse instance segmentation results, e.g. mask_rcnn_r50.val.segm.json in the demo commands, the contents are COCO style results. The things under the key segmentation (i.e. size and count) represent the binary instance mask using RLE. It is the standard format of instance segmentation in COCO dataset. You can find detailed instructions here.