Closed zf223669 closed 1 year ago
I believe the purpose of this was to project the output from the RNN to the conditioning vector length which the conditional denoising diffusion expects... however if the conditional denoising diffusion module expects the vector size to the same as the hidden size of the RNN then yes this part is perhaps not needed, but with this i can control the conditional vector size... hopefully this helps?
Kashif, thank you for your answer :)
Hi, I could not understand the code "distr_args = self.distr_args(rnn_outputs=rnn_outputs)" in time_grad_network.py, Could you explain more detail about that, what`s that purpose? Thank you :)