Tushar-N/pytorch-resnet3d

10-crops I3D feature extraction

Closed this issue · 3 comments

DungVo1507 commented

Hi @Tushar-N
Can you show me in detail how you 10-crops argumentation from an input video? Something like converting the video to a frame, then 10-crops argumentation,...
Sorry, I'm new in this field so I'm still having a lot of trouble.
Thank you very much!

ro1406 commented

Hi @DungVo1507, did you manage to figure out how to do 10crop augmentation from the features extracted? If so, could you let me know too? Thanks :)

DungVo1507 commented

@ro1406, all you need is here

  • 👍3
wzy1845 commented

@ro1406, all you need is here

Thank you for your code! I ran the code very smoothly, but it did not achieve the expected effect after the run of feature extraction using this repo. The result of the SH dataset was only less than 70%. My reproduction steps are as follows:

1.Use your code to extract i3D features
2.Run Make_list_SH.py in RTFM
3.Train and test RTFM
Are there any steps I miss? thank you for your time