search

Roc-Ng / XDVioDet

Official implementation of "Not only Look, but also Listen: Learning Multimodal Violence Detection under Weak Supervision" ECCV2020
103 stars 17 forks source link

How did you do 5-crop augmentation? #4

Closed chengengliu closed 3 years ago

chengengliu commented 3 years ago

Can you show more details on it? Thanks.

DungVo1507 commented 3 years ago

Hello @chengengliu, I am also currently researching this article and I have the same question as you. Can you show me how to extract 5-crops I3D features? Thank you in advance!!

chengengliu commented 3 years ago

Hello @chengengliu, I am also currently researching this article and I have the same question as you. Can you show me how to extract 5-crops I3D features? Thank you in advance!!

Hi there, I'm pretty sure that they followed the paradigm from the standard pytorch 5-crop. If you can't write code as Ng mentioned in another issue#5 check it out and hope it works for you https://pytorch.org/vision/stable/transforms.html#torchvision.transforms.FiveCrop Also, for those who looking for extraction of features, Ng mentioned in the issue#6 they used the code from piergiaj/pytorch-i3d.