CNN-M pre-trained model

[cwzat]

I am using CNN-M model to train my own data，but the acc is low. Could you push your trained model? Thank you!

[stillbreeze]

Sorry, but I think the trained models were lost during some issue we had with our version control. Reproducing it shouldn't be a problem though. Maybe the difference from the paper is a result of you using CNN-M instead of VGG16. It's difficult to predict how the temporal stream will differ in performance with CNN-M.