Input of CNN_LSTM model

[MrDongdongLin]

Hi Nika,

I am also interested in using CNN-LSTM architecture to detect deepfake videos. Thank you for your awesome work. I get a lot of benefits from it. Here I would like to ask you a question about training CNN_LSTM model.

In the forward function of class CNN_LSTM, the shape of the input images returns five variables, which is different from the input of class FaceRecognitionCNN. What is the difference between 'num_channels' and 'depth'? It seems we have to change the function of dataloader to fit the input of CNN_LSTM, otherwise, it raises errors with not enough values to unpack.

Could you explain what are the variables of 'num_channels' and 'depth' means in this line? How should I change the dataloader to pack the train data to fit the dimension of the input of CNN_LSTM?

I am looking forward to your answer. Thank you very much.

Best

[MrDongdongLin]

Hi Nika,

I found codes about training on CNN_LSTM model in one of your branches... Now I know how to do with it. Forget my question... hahaha... Thank you very much. XD

Best

[MrDongdongLin]

Hi Nika,

I run the code on branch lstm, but I got the same error as I described before... What is the dimension of the input of CNN_LSTM?

[MrDongdongLin]

I checked the codes again, now I know what is 'depth'. I guess it is the window size, right?