Get in trouble with pytorch3d 's installing and tiny-cuda-nn 's compiling.

[tianhan-zhang]

Here is my cmd log. I rent a linux machine with rtx 3080 on AutoDL.com, because i only have a windows laptop with rtx 3070.

1. pytorch3d seems don't match the version of pytorch and python.

(easyvolcap) root@autodl-container-ecb6119b52-cd4c9d8e:~/4K4D# conda install pytorch3d -c pytorch3d
Channels:

• pytorch3d
• https://mirrors.ustc.edu.cn/anaconda/pkgs/main
• https://mirrors.ustc.edu.cn/anaconda/pkgs/free
• defaults
  Platform: linux-64
  Collecting package metadata (repodata.json): done
  Solving environment: failed

LibMambaUnsatisfiableError: Encountered problems while solving:

• nothing provides pytorch 1.5.1 needed by pytorch3d-0.2.5-py36_cu101_pyt151

Could not solve for environment specs
The following packages are incompatible
├─ pin-1 is installable and it requires
│ └─ python 3.12.* , which can be installed;
└─ pytorch3d is not installable because there are no viable options
├─ pytorch3d [0.2.5|0.3.0|0.4.0] would require
│ └─ python >=3.6,<3.7.0a0 , which conflicts with any installable versions previously reported;
├─ pytorch3d [0.2.5|0.3.0|0.4.0] would require
│ └─ pytorch 1.5.1 , which does not exist (perhaps a missing channel);
├─ pytorch3d [0.2.5|0.3.0|0.4.0] would require
│ └─ pytorch 1.6.0 , which does not exist (perhaps a missing channel);
├─ pytorch3d [0.2.5|0.3.0|0.4.0] would require
│ └─ python >=3.7,<3.8.0a0 , which conflicts with any installable versions previously reported;
├─ pytorch3d [0.2.5|0.3.0|0.4.0] would require
│ └─ python >=3.8,<3.9.0a0 , which conflicts with any installable versions previously reported;
├─ pytorch3d [0.3.0|0.4.0] would require
│ └─ pytorch 1.7.0 , which does not exist (perhaps a missing channel);
├─ pytorch3d 0.4.0 would require
│ └─ python >=3.9,<3.10.0a0 , which conflicts with any installable versions previously reported;
├─ pytorch3d [0.5.0|0.6.0|0.6.1] would require
│ └─ python_abi 3.6.* _cp36m, which does not exist (perhaps a missing channel);
├─ pytorch3d [0.5.0|0.6.0|...|0.7.4] would require
│ └─ cudatoolkit >=11.1,<11.2 , which does not exist (perhaps a missing channel);
├─ pytorch3d [0.5.0|0.6.0|...|0.7.1] would require
│ └─ python_abi 3.7. _cp37m, which does not exist (perhaps a missing channel);
├─ pytorch3d [0.5.0|0.6.0|...|0.7.5] would require
│ └─ python_abi 3.8. _cp38, which does not exist (perhaps a missing channel);
├─ pytorch3d [0.5.0|0.6.1|...|0.7.5] would require
│ └─ python_abi 3.9. _cp39, which does not exist (perhaps a missing channel);
├─ pytorch3d [0.6.2|0.7.0|...|0.7.5] would require
│ └─ python_abi 3.10. _cp310, which does not exist (perhaps a missing channel);
├─ pytorch3d [0.6.2|0.7.0|...|0.7.4] would require
│ └─ cudatoolkit >=11.5,<11.6 , which does not exist (perhaps a missing channel);
├─ pytorch3d [0.7.0|0.7.1|...|0.7.5] would require
│ └─ cudatoolkit >=11.6,<11.7 , which does not exist (perhaps a missing channel);
└─ pytorch3d 0.7.5 would require
└─ python_abi 3.11. *_cp311, which does not exist (perhaps a missing channel).

2.tiny-cuda-nn compiling failed , seems that it can't find <cusolverDn. h> which is a file in CUDA toolkit.

It‘s worth mentioning that, with python 3.9 +pytorch 1.12.1 +cuda11.6 , I can install pytorch3d , but can't compile tiny-cuda-nn.
If I just follow your "envirment.yml", conda will install python 3.12 +pytorch 2.2.1 +cuda12.1 , in this case , I can't install pytorch3d.

And i still don't get open3d installed.

By the way , I wonder if i can use GUI viewer in the rented linux machine（ubuntu 20.04） , it don't have GUI.
I have run nerfstudio and instantNGP in my windows laptop successfully. But when I try to install easyvolcap in windows , seems that some dependencies are not support windows system. Is there any way to get 4K4D run on my windows laptop? Although it only have a RTX 3070 with 8GB VRAM : )

Do you have some solutions or suggestions for my problem? Thanks a lot !

[dendenxu]

Hi @tianhan-zhang.

1. [Windows Laptop] As mentioned in the installation guide, it should be OK to just install the dependencies listed there to run the viewer on Windows. Maybe you could try installing with pip install -e . or just follow the readme of EasyVolcap itself here for a basic non-development install.
2. [Open3D & PyTorch3D] We recommend building PyTorch3D directly from source (indicated by the pytorch3d @ git+https://github.com/facebookresearch/pytorch3d line in requirements-dev.txt). For Open3D, indeed it only supports up to 3.11 for now if I'm correct. And in the requirement file I also listed some potential fixes for higher version. We'll update the requirement files to reflect that. Note that neither PyTorch3D and Open3D is required for the viewer.
3. [tiny-cuda-nn] This could be CUDA_HOME environment variable issue. Could you check the install.md documentation and see whether it can fix the compilation?

[dendenxu]

We've also recently updated the installation documentation & procedure. Maybe you can check out this doc: https://github.com/zju3dv/EasyVolcap?tab=readme-ov-file#installation for a smoother experience.
Feel free to drop in a comment if you encounter any problems during the installation of packages.