search

zyong812 / STIP

Code for CVPR22 paper: Exploring Structure-aware Transformer over Interaction Proposals for Human-Object Interaction Detection.
Other
44 stars 6 forks source link

pretrained models #2

Closed WXLL579 closed 2 years ago

WXLL579 commented 2 years ago

Thx for your excellent work! Could you please provide the best model for hico-det and vcoco datasets? thx!

WXLL579 commented 2 years ago

It takes large time to train, could you please provide the best model? thxxxx very much!!!

zyong812 commented 2 years ago

Hi @WXLL579 ,

How long does it take for your training? If you follow README instructions, ~1-day training will be enough, and it will be easy to reproduce the results.

Or, you can test with following pretrained models:

WXLL579 commented 2 years ago

thx!! The slow training is caused by our machine, we use one 1080 (ㄒoㄒ).

WXLL579 commented 2 years ago

Is there a pretrained model through Jointly fine-tune object detector & HOI detector on HICO-DET? As shown in Table 2. STIP (Ours) R50 A+S+L 32.22 28.15 33.43 35.29 31.43 36.45 Thanks!