search

antoine77340 / S3D_HowTo100M

S3D Text-Video model trained on HowTo100M using MIL-NCE
Apache License 2.0
191 stars 21 forks source link

Video preprocessing steps? #5

Closed yushuinanrong closed 4 years ago

yushuinanrong commented 4 years ago

Hi @antoine77340,

Thanks for sharing this codebase. I'd like to evaluate your pretrained model on several custom videos but I don't see any code/instruction on how to preprocess the videos for inference. Could you share some insight?

Thanks,

antoine77340 commented 4 years ago

Hi,

You can take a look at our end-to-end evaluation script on YouCook2: https://github.com/antoine77340/MIL-NCE_HowTo100M/blob/master/eval_youcook.py

Or HMDB: https://github.com/antoine77340/MIL-NCE_HowTo100M/blob/master/eval_hmdb.py

yushuinanrong commented 4 years ago

@antoine77340 Thanks for your reply. Based on this line: https://github.com/antoine77340/MIL-NCE_HowTo100M/blob/dd0ff324439fed58de6a4ffc50f3d582a22cc2a4/eval_youcook.py#L63 it seems preprocessing only need to rescale the pixel value to [0, 1], is it right?

antoine77340 commented 4 years ago

That's right!

yushuinanrong commented 4 years ago

@antoine77340 Could you help to confirm the pretrained S3D model in this codebase is same as the one you generating the public available features for Howto100M (https://www.di.ens.fr/willow/research/howto100m/)?