Hi,
I apologize for possible confusion, but it looks like the ranker is not used at all, when getting topk keypoints?
At least, that is what I see from the source code - the keypoints are sampled according to the point score, and the ranking score is only sampled for existing ones (within the number of selected keypoints)
https://github.com/cvg/RaCo/blob/main/raco/raco.py#L587C1-L616C54
To put is blunt, with current code user have to use, say 4000kpts and then sort and crop them by ranking manually, otherwise ranker is useless?
Best, Dmytro
Hi,
I apologize for possible confusion, but it looks like the ranker is not used at all, when getting topk keypoints?
At least, that is what I see from the source code - the keypoints are sampled according to the point score, and the ranking score is only sampled for existing ones (within the number of selected keypoints)
https://github.com/cvg/RaCo/blob/main/raco/raco.py#L587C1-L616C54
To put is blunt, with current code user have to use, say 4000kpts and then sort and crop them by ranking manually, otherwise ranker is useless?
Best, Dmytro