search

Charles-Xie / CQL

Code for our paper "Category Query Learning for Human-Object Interaction Classification" (CVPR2023)
https://arxiv.org/abs/2303.14005
Apache License 2.0
33 stars 5 forks source link

VCOCO Result #6

Closed songguoatongmu closed 11 months ago

songguoatongmu commented 11 months ago

Thanks for your code realease! However, I've observed that when using the QPIC+CQL checkpoint on the VCOCO dataset, I couldn't achieve the results you reported (63.57, 65.70). I obtained results of 60.83 and 62.91. Did I do something wrong?

Charles-Xie commented 11 months ago

The evaluation on V-COCO is same as QPIC and many other methods (GEN-VLKT, etc.). This is answered in previous issues of QPIC. Please see this: https://github.com/hitachi-rd-cv/qpic/issues/18 and this https://github.com/hitachi-rd-cv/qpic/issues/7. I believe you will be able to reproduce the result following these links.