search

chaneyddtt / Coarse-to-fine-3D-Animal

MIT License
34 stars 9 forks source link

Quesion about stage 2 results #7

Closed MooreManor closed 1 year ago

MooreManor commented 1 year ago

@chaneyddtt

Do you still remember the results after stage2 which freezes the coarse network and only trains the subsequent network? Do the results get better before stage 3, or only after stage 3 the results get better?

chaneyddtt commented 1 year ago

Hi @MooreManor, the results in terms of keypoints PCK are slightly better after stage2, but not in terms of mask IOU. This is because we only apply keypoints supervision in stage1 and stage2. Both measures improve only after stage 3.

MooreManor commented 1 year ago

@chaneyddtt

Thanks! By the way, did you try to optimize the camera parameters in stage 2 and 3?

chaneyddtt commented 1 year ago

Not in stage 2 since the coarse estimation part is frozen, but yes in stage 3 since the whole network is optimized.

MooreManor commented 1 year ago

:) clear now! Thanks for your quick reply!!!