Closed HyunsooCha closed 1 year ago
Maybe the pytorch3d is built without CUDA support?
Thanks for your reply I solved this issue using this: ENV TORCH_CUDA_ARCH_LIST="8.6+PTX"
@HyunsooCha Could you share docker image that you used and a little bit of instructructions how did you run this project, please?
Hi
Thank you for your excellent work. To install pytorch3d in your repository, I've got this message:
$ git clone git@github.com:zhengyuf/pytorch3d.git Cloning into 'pytorch3d'... git@github.com: Permission denied (publickey). fatal: Could not read from remote repository.
Please make sure you have the correct access rights and the repository exists.
So, I installed your modified version of pytorch3d like this:
git clone https://github.com/zhengyuf/pytorch3d.git git checkout point-avatar pip install -e .
However, when I trained the code, I got this error message:
Traceback (most recent call last): File "/root/GitHub/PointAvatar/code/scripts/exp_runner.py", line 35, in
runner.run()
File "/root/GitHub/PointAvatar/code/../code/scripts/train.py", line 325, in run
model_outputs = self.model(model_input)
File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, *kwargs)
File "/root/GitHub/PointAvatar/code/../code/model/point_avatar_model.py", line 176, in forward
images = self._render(transformed_point_cloud, cameras)
File "/root/GitHub/PointAvatar/code/../code/model/point_avatar_model.py", line 92, in _render
fragments = rasterizer(point_cloud)
File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(input, *kwargs)
File "/root/GitHub/pytorch3d/pytorch3d/renderer/points/rasterizer.py", line 132, in forward
idx, zbuf, dists2 = rasterize_points(
File "/root/GitHub/pytorch3d/pytorch3d/renderer/points/rasterize_points.py", line 142, in rasterize_points
return _RasterizePoints.apply(
File "/root/GitHub/pytorch3d/pytorch3d/renderer/points/rasterize_points.py", line 223, in forward
idx, zbuf, dists = _C.rasterize_points(args)
RuntimeError: Not compiled with GPU support
I used Docker because my server's cuda version is 12.0. When I built Dockerfile, there was no error message. The build was successful. Also, I double-checked if there's something wrong related to Pytorch or Pytorch3D like this:
Python 3.10.9 (main, Jan 11 2023, 15:21:40) [GCC 11.2.0] on linux Type "help", "copyright", "credits" or "license" for more information.
Do you know how to solve this issue? Thank you for your reply in advance.