use video-only on LRS2

[angiecao]

Hi,
I used the video-only model for 100 iteration training on the dataset LRS2, but it always predicted almost the same sentences. What is the cause of the problem?

[georgesterpu]

Hi @Dejavuee
This is a really great question. I assume that you are using a sequence to sequence neural network model for the video-to-text task, is this correct? Please send me an email (can be found in our articles), since the issue is not fully related to this github repository. Thanks.