About the diffence of the augment --dim_feedforward 1024/2048 in pretrained model and eval/test model

[wi11Iam5-mpu]

Thanks again for your work! As mentioned in the question, I found that in the pre-trained crowdhuman_hybrid_branch.pth (for detection, I guess), the dim_feedforward parameter is 2048. And it is also mentioned in the instructions. However, in the weights file val/test.pth (for tracking), this parameter is 1024. I don't know why this is happening, if you have time, dealing with these, I would be very appreciative.

[wi11Iam5-mpu]

BTW, the weight of TransTrack (marked as 'Our repro.') is missing.

[Rookielgc]

I need to use the datasets folder in the main_track.py file, but I can't find it, how can I do it, is it convenient for you to provide it?

[Rookielgc]

I found the relevant dataset and weight files in the TransTrack repo