fraunhoferhhi / gaussian_gan_decoder

Official implementation for: https://arxiv.org/abs/2404.10625
https://florian-barthel.github.io/gaussian_decoder/index.html
40 stars 1 forks source link

Failed building wheel for diff_gaussian_rasterization #11

Open ErisRolo opened 4 days ago

ErisRolo commented 4 days ago

Hi, I found that the gaussian_splatting directory in the open-source code does not contain a submodules directory, making it impossible to directly create an environment using the YML file. Therefore, I have attempted to create it manually. I've noticed that in the .gitmodules file, the URL for the diff-gaussian-rasterization module is specified as "https://github.com/slothfulxtx/diff-gaussian-rasterization.git", and the generated dependency name is "diff_gauss". However, in the provided source code, specifically in the gaussian_splatting/gaussianrenderer/__init_\.py file, the import statement still uses diff_gaussian_rasterization. Could you please explain if the submodule being used is indeed the same as the one in the original 3DGS project?

Florian-Barthel commented 3 days ago

Hi, thanks for pointing out this issue. The problem was that the gitignore file was removing the submodules. This is now fixed. If you clone the repository again using the --recursive flag, it should now install all submodules.

The "https://github.com/slothfulxtx/diff-gaussian-rasterization.git" was an artifact from prior experiments with depth. This is now replaced with the official submodule used in 3DGS. Let me know if it is working now

Florian-Barthel commented 3 days ago

After switching to the original 3DGS rasterization submodule I could not install it on my machine any longer due to CUDA_HOME issues. I changed it back to the diff-gaussian with depth information and fixed the code accordingly

ErisRolo commented 10 hours ago

Hi, thank you for your help, but there are still error messages.

Loading networks from "../PanoHead/models/easy-khair-180-gpc0.8-trans10-025000.pkl"... Reloading Modules! 0%| | 0/100001 [00:00<?, ?it/s]Setting up PyTorch plugin "bias_act_plugin"... Done. Setting up PyTorch plugin "upfirdn2d_plugin"... Done. 0%| | 0/100001 [00:01<?, ?it/s] Traceback (most recent call last): File "/home/ghx/3DV/gaussian_gan_decoder/main/train_pano2gaussian_decoder.py", line 322, in main() File "/home/ghx/.conda/envs/gs_decoder/lib/python3.10/site-packages/click/core.py", line 1157, in call return self.main(args, kwargs) File "/home/ghx/.conda/envs/gs_decoder/lib/python3.10/site-packages/click/core.py", line 1078, in main rv = self.invoke(ctx) File "/home/ghx/.conda/envs/gs_decoder/lib/python3.10/site-packages/click/core.py", line 1434, in invoke return ctx.invoke(self.callback, ctx.params) File "/home/ghx/.conda/envs/gs_decoder/lib/python3.10/site-packages/click/core.py", line 783, in invoke return __callback(args, *kwargs) File "/home/ghx/3DV/gaussian_gan_decoder/main/train_pano2gaussian_decoder.py", line 232, in main render_obj = render_simple(viewpoint, gaussians, bg_color=bg) File "/home/ghx/3DV/gaussian_gan_decoder/main/../gaussian_splatting/gaussian_renderer/init.py", line 167, in render_simple rendered_image, rendered_depth, rendered_alpha, radii = rasterizer( File "/home/ghx/.conda/envs/gs_decoder/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(args, **kwargs) TypeError: GaussianRasterizer.forward() got an unexpected keyword argument 'cov3D_precomp'

I will try the original 3DGS rasterization submodule, but I'm not sure if it works.

ErisRolo commented 9 hours ago

I switched to the original 3DGS rasterization submodule, and then it worked.

But can you fix the error mentioned above? Thank you.

ErisRolo commented 5 hours ago

Additionally, I noticed that although the modified .gitmodules file points the diff-gaussian-rasterization module to "https://github.com/slothfulxtx/diff-gaussian-rasterization.git", the uploaded submodules file points the module URL to "https://github.com/slothfulxtx/diff-gaussian-rasterization/tree/59f5f77e3ddbac3ed9db93ec2cfe99ed6c5d121d". The latter does not differ from the original version of the 3DGS module. Should the contents of the submodules be modified?