search

kahnchana / svt

Official repository for "Self-Supervised Video Transformer" (CVPR'22)
https://kahnchana.github.io/svt
MIT License
104 stars 22 forks source link

Data preparation #12

Closed weirukai closed 2 years ago

weirukai commented 2 years ago

Thanks for sharing your codes. I have some problems loading the training data of UCF101. 1) Where can I find the ucf101_train(test)_split_1_videos.txt? 2) Should I extract RGB frames for each dataset before training?

kahnchana commented 2 years ago

Thank you for your interest in our work!

We use MMAction2 repo for data preparation. We run section 1, 2, and 4 (skip 3 - extract RGB frames / flow).

Running section 4 will generate those files.

You do not need to extract RGB frames. The dataloader directly reads from videos.