L-YeZhu/D2M-GAN

Checkpoint

liangx66 opened this issue · 3 comments

Hi author, Nice Work!
I have tried the generation process on AIST++, it seems that the generated samples have the basic rhythm but don't sound real enough. I don't know what's the problem with it. Is it the reason for high level model setting? And I wonder if you could share the pre-trained models of low level for quick start?

Hi, thanks for your interest in our work.
It is true that sometimes the rhythms may come earlier than the melody part, leaving it with a little bit of longer training time should help with mitigating this problem. In the meanwhile, I can search for the low level checkpoints to share.

thank you for your reply

You can download the checkpoints for the low level here: Generator and Motion encoder.