In my understanding after reading your codes, optical flows are calculated for the videos in 'trainVidID.txt' and 'testVidID.txt'. And for each video, optical flows are divided into blocks of 50 frames. And for each block, 10 frames are used as training data. And the last few frames that can't form a block are discarded.
So it means that at most 1/5 of the optical flow will be used for training. Will this be a waste of data?
Also, the optical flows are sequences. op[i:i+10] and op[i+1:i+11] could both be used as training data, right? Why didn't you take the latter into the training data?
In my understanding after reading your codes, optical flows are calculated for the videos in 'trainVidID.txt' and 'testVidID.txt'. And for each video, optical flows are divided into blocks of 50 frames. And for each block, 10 frames are used as training data. And the last few frames that can't form a block are discarded.
So it means that at most 1/5 of the optical flow will be used for training. Will this be a waste of data?
Also, the optical flows are sequences. op[i:i+10] and op[i+1:i+11] could both be used as training data, right? Why didn't you take the latter into the training data?
And there's no data augmentation, right?