threequestions about video classification about uniformer

[Andy1621]

1. The videos are resized before the program runs. You can follow the scripts in MMAction2 to prepare the data. Since the videos were processed by others, I don't have the processing code. I can provide you with the processed videos if necessary. If you want it, please leave your email~~

2. Freeze BN only works when the batch size is too small, which is a popular operation in Object Detection and Semantic Segmentation. I have not used it for video classification. Actually, in previous works like TSN, they freeze BN for UCF101 and HMDB51. In my opinion, the operation only works when fine-tuning special models on a small dataset.

3. In my repo, I read videos for Kinetics via decord, which processes the raw videos. If you also process raw videos, I think it is okay, and you only need to prepare the correct data list as in my data.

[JieHarry-Hu]

Thank you so much for your detailed answer, it couldn't be better if you could provide the processed video, my email is [email protected]. Finally, thank you for your help.

[Andy1621]

I have sent it to you email, please check it~~

[Andy1621]

As there is no more activity, I am closing the issue, don't hesitate to reopen it if necessary.

[Auroralyxa]

I have sent it to you email, please check it~~
Hi,
Could you please provide me with the processed video, my email is [email protected], thank you in advance