DGAM-Weakly-Supervised-Action-Localization icon indicating copy to clipboard operation
DGAM-Weakly-Supervised-Action-Localization copied to clipboard

Published 20 hours ago • verified publisher icon bfshi

NUM_INPUT_FRAMES

Open simpler66 opened this issue 4 years ago • 1 comments

Hi, thank you very much for sharing your code! Quick question about I3D features.

I notice the config.DATASET.NUM_INPUT_FRAMES is set to16. When I borrow the released codes in https://github.com/piergiaj/pytorch-i3d for extracting I3D features, the I3D network takes every 8-frame chuncks as input for both two streams. Thus, the final features are sampled by 8. I wonder if there is any process to get the feature sampled by 16.

simpler66 avatar Oct 09 '20 13:10 simpler66

Hi, thanks for the interest.

The I3D feature we use are extracted every 16 frames. The official repo is https://github.com/deepmind/kinetics-i3d

But if you are using 8-frame features, then I think you can safely change the config.DATASET.NUM_INPUT_FRAMES to 8.

bfshi avatar Oct 09 '20 15:10 bfshi