code problem

[Steven-jiaqi]

I found this function get_aggregator(args) in tvgnet.py 82 lines. But there is no args.aggregation ''. And i found the args.features_dim=768 in the function get_encoder(args). So I assign the args.features_dim =768. But new error appeared: ValueError: optimizer got an empty parameter list

I can't solve this problem. I need your help. I'm sorry to bother you again. Looking forward to get your reply as soon as possible!

[ga1i13o]

Hello, I think I solved your issue, and it is actually my fault. The problem is at this line:

https://github.com/vandal-vpr/vg-transformers/blob/25537ed1689d4b5401113f6c4476fb8f0cf28158/tvg/models/tvg_net.py#L159

In the README I indicated that you should pass --arch timesformer, but the code actually expects you to pass --arch Official-timesf, as you can see in the line above. I will fix this as soon as I can and for now you can change your command line args to --arch Official-timesf

Thank you for noticing. I am closing your other issue

[Steven-jiaqi]

Thank you for timely response.And I have another question. Is the --aggregation 'none' or ''? It seems to be '' in the code. Please forgive me for having a bit too many questions! Best wishes to you!

[ga1i13o]

Don't worry, I should have done a better job documenting the code. Anyway for the TimeSformer you can pass either 'none' or leave the field empty, since in that case only the encoder will be called as you can see in the forward method (using timesformer sets fusion='early ' )

https://github.com/vandal-vpr/vg-transformers/blob/25537ed1689d4b5401113f6c4476fb8f0cf28158/tvg/models/tvg_net.py#L45-L49

[Steven-jiaqi]

I have solved the problem. I'm sorry to bother you.

[ga1i13o]

Yes the issue should have been that you have to specify the image resolution. I am working to push a commit to make this automatic, and update the readme