search

Arthur151 / ROMP

Monocular, One-stage, Regression of Multiple 3D People and their 3D positions & trajectories in camera & global coordinates. ROMP[ICCV21], BEV[CVPR22], TRACE[CVPR2023]
https://www.yusun.work/
Apache License 2.0
1.36k stars 231 forks source link

AGORA-PC results #473

Open fabienbaradel opened 1 year ago

fabienbaradel commented 1 year ago

Hi, Thanks for releasing the code. I have a question concerning your evaluation on AGORA for the AGORA-PC setting on the validation. You are mentioning that: "Each image has 5-15 people with frequent occlusions. AGORA-PC [27] is a high occlusion subset of the AGORA validation set. Each image has over 70% occlusion." But if we keep images where all the persons have at least 70% of occlusion there are only 19 images. Am I missing something? Thanks for your help, Best

Arthur151 commented 1 year ago

@fabienbaradel With AGORA's official evaluation code, you should be able to get the results on the subset which has over 70% occlusion. But sorry, we didn't count the number of samples here. Maybe you can refer to the results on test set if you seek more samples on occlusion cases. The official evaluation server of AGORA should also provide such result analysis every time.