Anyone tried it by extracting features from i3d.

WaqasSultani / AnomalyDetectionCVPR2018

503 stars 183 forks source link

Anyone tried it by extracting features from i3d. #54

Closed sanchit2843 closed 4 years ago

arash9900 commented 4 years ago

did you find any answer? thanks

sanchit2843 commented 4 years ago

I tried it, the performance was not adequate. I think it was because I couldn't find pretrained weights of the sports-1m dataset. The pretrained weight I used was kinetics-400. Please tell me if you are able to find weights of any good video recognition model on sports-1m.