Lin Yihong

hello, is there any pretrained styleGAN2 with 256x256 px that provided, as the FFHQ dataset is too large to download and it may take a long times for training.

styleGAN 256x256

2

hello, is there any pretrained styleGAN2 with 256x256 px that provided, as the FFHQ dataset is too large to download and it may take a long times for training.

Train on multi-gpus

3

thanks for your great work! I have found that the project seems to support training on multi-gpus. How can I run it with multi-gpus?

Hi, thanks for your great work! As `smpl_numpy.py` is located in `third_parties` directory, I can not find where is this part of code comes from. Can you provide any information...

Hi, thanks for your great work and for providing this dataset. I noticed that in the training set, the SMPL-X parameters are provided, but in the test set only the...

support gender configuration

1

Hello, thank you for your great work! I’m trying to perform novel pose synthesis (animation), but I only have SMPL-X parameters for male and female models, not the neutral one....

Thanks for your great work. As mentioned in the paper, for the 2k2k dataset, 30 azimuthal views per model with uniform angular spacing are rendered. Could you please provide more...

@chingswy Sorry to bother. As mentioned in the [SMPL-X repo](https://github.com/vchoutas/smplx?tab=readme-ov-file#modifying-the-global-pose-of-the-model) and the [EasyMocap doc](https://github.com/zju3dv/EasyMocap/blob/master/doc/02_output.md#attention-for-smplsmpl-x-users), the canonical way to modify the global pose is: ```python x = global_orient @ (x -...

LPIPS turn slow during training

1

## 🐛 the append operation in update() cause training turing slow ### To Reproduce Here is a simple demo for reimplementation. When using torchmetrics==1.7.0, it works well, but with 1.8.1...