cocodataset / panopticapi

COCO 2018 Panoptic Segmentation Task API (Beta version)
Other
418 stars 184 forks source link

Questions about evaluation.py #19

Open Mooonside opened 5 years ago

Mooonside commented 5 years ago

Hi, I was reading evaluation.py and found some lines confusing: https://github.com/cocodataset/panopticapi/blob/8a32c19cde88dccb7336dc011e8a4a91f9d68c01/evaluation.py#L141-L147 For 'is_crowd' segments, category_id and segment id are stored in a dictionary, so it implies that for each category, coco annotates at most one segment with is_crowd as true. In other words, suppose that there are two distant crowds of people, they are annotated with the same segment id. However, I can't find any support of this implication in the description of the panoptic dataset. Is this a feature of coco dataset?

natsirtguy commented 2 years ago

A little late but +1. I think it turns out to be true for COCO, but if you want to convert another dataset to the COCO format, you probably need to make sure all the crowd segments of a given category have the same id for this code to work.