two-stream-action-recognition-keras
two-stream-action-recognition-keras copied to clipboard
Temporal predicts with same probability for every video
When I try to train the temporal model with my custom dataset, the loss and accuracy values seems to oscillate a lot. Also when I use this to predict (currently only on two classes), it predicts the same class everytime for each video and and its every frame. I even tried a different architecture for the temporal stream. Even that didn't work.