leonardodora

> You can do it using the Flow module: > > https://github.com/RameenAbdal/StyleFlow/blob/153eff627b6cb14a433026d472f7c7a3adc35d2e/main.py#L103 > > Will include it in the next update. thanks for your answer!

> Yes, cuda11.2 also works. The requirements have been listed in the Readme file. I think the main problem is compiling the op provided by [StyleGAN2](https://github.com/rosinality/stylegan2-pytorch) or installing the [pytorch3D](https://github.com/facebookresearch/pytorch3d)....

> Maybe you can refer to this [Issue](https://github.com/facebookresearch/pytorch3d/issues/1076).

**here are the log of installing the op. there are many warnings but no error** python setup.py install running install /data/miniconda3/envs/fv/lib/python3.9/site-packages/setuptools/command/install.py:34: SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip...

> It seems that you didn't use ninja to compile the ops. ‘ninja: no work to do.’ above beause I have compiled before，so I delete the build files before compile...

> thanks for your help! I have fixed this error by not installing onnxruntime-gpu. but I got another error: maybe it's because the version of pytorch(1.10.0) and cuda(mine is 11.2...

I have solved this bug. It's just from pytorch1.10.0 installed by conda. And the time consuming per image is 40 minutes, is it expected?

> Yes, we use it directly. It adapts very quickly and the transformation is visible in about 500 steps. This is consistent with [pixart-sigma](https://arxiv.org/abs/2403.04692). Thanks for your reply! But if...

> We do not encounter this problem. One potential possibility is the half-precision training. You should use bf16 instead of fp16. Thanks, but bf16 is set in config file. Did...

> Our computing resource is limited and does not try training from scratch for a long time. Thanks for your reply! btw, the newest update is awsome!