search

wangxiang1230 / OadTR

Code for our ICCV 2021 Paper "OadTR: Online Action Detection with Transformers".
MIT License
87 stars 12 forks source link

About the Pre-trained pth model？ #20

Open fearless77 opened 2 years ago

fearless77 commented 2 years ago

The pth model file is not found. Will you upload it later? Thank you very much !!!! ><

wangxiang1230 commented 2 years ago

The pth model file is not found. Will you upload it later? Thank you very much !!!! ><

Hi, sorry for the late reply. The checkpoint file is available at here.

Acho-177 commented 1 year ago

I test the pre-trained model but only get mAP 0.251. Are there any hyper arguments didn't list in README?

Here is the result:

[Epoch-0] [IDU-Anet2016_feature_v2] mAP: 0.2514

dec_mAP all together: | 0.25450710949370564 |. dec_mAP_pred | 0 : 0.2589848446097752 |. dec_mAP_pred | 1 : 0.2581012134777999 |. dec_mAP_pred | 2 : 0.25719438306977993 |. dec_mAP_pred | 3 : 0.25606081013959126 |. dec_mAP_pred | 4 : 0.2547937451546783 |. dec_mAP_pred | 5 : 0.25335031046260187 |. dec_mAP_pred | 6 : 0.2518111834013882 |. dec_mAP_pred | 7 : 0.2501947586777621 |. all decoder map: | 0.2551 |. BaseballPitch: 0.0095 BasketballDunk: 0.2961 Billiards: 0.1855 CleanAndJerk: 0.4945 CliffDiving: 0.3580 CricketBowling: 0.1529 CricketShot: 0.0531 Diving: 0.2813 FrisbeeCatch: 0.2578 GolfSwing: 0.3417 HammerThrow: 0.3438 HighJump: 0.1918 JavelinThrow: 0.1723 LongJump: 0.2305 PoleVault: 0.5406 Shotput: 0.3216 SoccerPenalty: 0.0680 TennisSwing: 0.2675 ThrowDiscus: 0.3286 VolleyballSpiking: 0.1328

update

Problem solved. Need to use pytorch==1.6.0. I was using a higher version, which caused the low performance.