search

tianyu0207 / RTFM

Official code for 'Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning' [ICCV 2021]
324 stars 77 forks source link

Getting low auc while training on UCF-Crime Data using I3D features(10-crop) #13

Closed Shashank56 closed 3 years ago

Shashank56 commented 3 years ago

We have used the following repo for extracting 10-crop I3D features. https://github.com/Tushar-N/pytorch-resnet3d

But while training we are not having any improvement in the auc after 400 iterations. I have attached the graph shown in visdom, so is there anything that we are doing wrong? And when will you publish the I3D features for UCF-Crime dataset? graph

Shashank56 commented 3 years ago

Can you please post the exact I3D code that you have used? We have used the above repo as mentioned and made some changes to incorporate 10crop and rgb frames. Here is the link for the code. https://github.com/GowthamGottimukkala/I3D_Feature_Extraction_resnet But we couldn't get the results that we have expected. It would be helpful if you send your exact code.

tianyu0207 commented 3 years ago

Can you please post the exact I3D code that you have used? We have used the above repo as mentioned and made some changes to incorporate 10crop and rgb frames. Here is the link for the code. https://github.com/GowthamGottimukkala/I3D_Feature_Extraction_resnet But we couldn't get the results that we have expected. It would be helpful if you send your exact code.

I have uploaded the UCF features and checkpoint. Please change the video level labels accordingly in the dataloader when using the UCF.

Cheers

nathanb97 commented 3 years ago

Hey, I took a look at your code. Do you end up getting good results? If yes, did you do a normalisation ?