search

tinyvision / SOLIDER-REID

MIT License
55 stars 12 forks source link

Training Details #7

Closed ForeUP closed 1 year ago

ForeUP commented 1 year ago

I try to train Swin-T with my device, but I can't get similar performance as you uploaded .pth file : 91.6/96.1. I use batch_size=64, train 180 epochs, other configurations such as learning rate are set by default in your configuration file.

The results as below:

2023-05-02 18:15:32 transreid.train INFO: Validation Results - Epoch: 180
2023-05-02 18:15:32 transreid.train INFO: mAP: 85.1%
2023-05-02 18:15:32 transreid.train INFO: CMC curve, Rank-1  :94.2%
2023-05-02 18:15:32 transreid.train INFO: CMC curve, Rank-5  :98.1%
2023-05-02 18:15:32 transreid.train INFO: CMC curve, Rank-10 :98.7%

Could you share more Training Details, Thanks a lot!

cwhgn commented 1 year ago

Thanks for your interests. For the convenience of your re-implementation, we upload our training logs into the repo, you can find them in the README.md.

ForeUP commented 1 year ago

Thanks a lot! I think I know the reason why the error occurred. I used the Swin-small pretrained model with modified init_weight code in #4 . Thanks for your response.