Why some of the generate images by the official checkpoint of CelebA64 are NaN-value?

NVlabs / NVAE

The Official PyTorch Implementation of "NVAE: A Deep Hierarchical Variational Autoencoder" (NeurIPS 2020 spotlight paper)

https://arxiv.org/abs/2007.03898

Other

1.03k stars 164 forks source link

Why some of the generate images by the official checkpoint of CelebA64 are NaN-value? #36

Open YangXiaojiang1874 opened 2 years ago

YangXiaojiang1874 commented 2 years ago

Great work! I have observed that some generated images by official checkpoint of CelebA64 are NaN . This might be caused by "Batch Norm Statistics Readjusted" introduced by the paper, becuase without BN statistics readjusted, none of the generated images are NaN (but the generated iamges quality have a sharp decline). Do you have this problem and how to deal with it? Thank you for the excelllent paper and the great repo!

YangXiaojiang1874 commented 2 years ago

The same problem also occurs on CelebA256.

black0017 commented 1 year ago

Same here on CIFAR10 (using --temp=0.6 --readjust_bn).

When I run the official eval script to generate images on CIFAR10 (official checkpoints) I get the same behavior...