I found that the bounding box loss function uses GIOU loss and L1 loss, and the hyperparameter settings are the same as TransT. Have you tried DIOU or CIOU? Do you have any experience with hyperparameter settings? thanks a lot！

MCG-NJU / MixFormer

[CVPR 2022 Oral & TPAMI 2024] MixFormer: End-to-End Tracking with Iterative Mixed Attention

https://arxiv.org/abs/2203.11082

MIT License

450 stars 75 forks source link

I found that the bounding box loss function uses GIOU loss and L1 loss, and the hyperparameter settings are the same as TransT. Have you tried DIOU or CIOU? Do you have any experience with hyperparameter settings? thanks a lot！ #2

Closed wildwolff closed 2 years ago

yutaocui commented 2 years ago

We haven't try other loss in order to take a relatively fair comparison with STARK/TransT. For hyperparameter settings, I usually train different-setting models for a few epochs.