joslefaure / HIT

Official Implementation of our WACV2023 paper: “Holistic Interaction Transformer Network for Action Detection”
https://arxiv.org/abs/2210.12686
48 stars 7 forks source link

the data of UCF101-24 #11

Closed yan-ctrl closed 3 months ago

yan-ctrl commented 1 year ago

Hello, can you share the data of UCF101-24? I want to try the training of AVA dataset after UCF101-24 and JHMDB training. I hope you can share it with me. Thank you very much!

yan-ctrl commented 1 year ago

The AVA data set is too large. It takes a long time to train once

joslefaure commented 1 year ago

I used 8 32GB Titan GPUs to train it. It took around 25 hours

On Wed, Feb 15, 2023, 02:18 yan-ctrl @.***> wrote:

The AVA data set is too large. It takes a long time to train once

— Reply to this email directly, view it on GitHub https://github.com/joslefaure/HIT/issues/11#issuecomment-1430178581, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALOMSBBGEGZO3KADHDVRBVLWXPEAJANCNFSM6AAAAAAU34RPIY . You are receiving this because you are subscribed to this thread.Message ID: @.***>

yan-ctrl commented 1 year ago

I don't have so many cards. The device is limited. The processing method of UCF101-24 is the same as that of JHMDB, right? Can you share with me what you still have?