Closed trainma closed 2 years ago
Hello it is the initial features, reshaped such that we can concatenate them with the rnn representations (line 160). this is fed to the final layer and it is called a skip connection.
It just like the residual connection in resnet. Thank you very much.
hello author.I have read your code and found a skip tensor(models.py line-120).I dont know what the function of this skip tensor in the forward function.Hoping for your reply.