Hello TripoSR team,

I have been trying to install TripoSR on a Windows 11 system and have run into a persistent and blocking issue related to the torchmcubes dependency. After several days of extensive troubleshooting, there is no straightforward way to install TorchMCubes on a modern Windows environment.

My Environment:

OS: Windows 11
GPU: NVIDIA 40-series (CUDA architecture sm_89 detected)
Env Management: Freshly reinstalled Miniconda
Python: 3.9
Build Tools: Freshly reinstalled Visual Studio 2022 Build Tools with "Desktop development with C++" workload.
Problem Description:

The installation consistently fails because torchmcubes cannot be installed, which ultimately results in a ModuleNotFoundError when attempting to run gradio_app.py.

Summary of Troubleshooting Steps Taken:

I have gone through an exhaustive process to ensure my environment is spotless and all dependencies are installed correctly. The torchmcubes package is the single remaining point of failure.

Initial pip Failure: The standard pip install -r requirements.txt fails with the error 'ERROR: Failed building wheel for torchmcubes due to a CMake configuration failure.' This issue persisted even after correctly installing the VS 2022 Build Tools.

Clean Environment Rebuild: To rule out any environment corruption, I have performed a complete uninstall and clean reinstall of both Miniconda and all Visual Studio components.

Dependency Installation Strategy: In the new clean environment, I followed this robust procedure:

Created a new conda environment: conda create-- name triposr python=3.9 -y
Activated it: conda activate triposr
Successfully installed PyTorch with CUDA 12.1: conda install pytorch torchvision pytorch-cuda=12.1 -c pytorch -c nvidia -y. We tried both the latest PyTorch and v2.2.2.
Attempting to install torchmcubes:

Via pip from PyPI, 'pip install torchmcubes' fails with the error: 'ERROR: No matching distribution found for torchmcubes.' This indicates a working wheel is not available on PyPI for my platform.
Via conda-forge: conda install -c conda-forge torchmcubes fails with PackagesNotFoundError. The package is not available for Windows on this channel.
Via Building from Source: This was our final attempt.
We cloned the official tatsy/torchmcubes repository.
Navigated into the source directory.
Ran pip install ..
This failed with a deep CMake Error related to torch::nvtoolsext and CUDA::NvToolsExt not being found, even when attempting with different PyTorch versions (latest and 2.2.2).
Conclusion:

The torchmcubes package is fundamentally incompatible with the modern PyTorch, CUDA, and Windows build toolchain, making it impossible to install via standard methods. As this is a critical dependency, it currently prevents TripoSR from being used on Windows.

Additional Feedback:

As a side note, the requirements.txt file appears to be missing several key dependencies required to run gradio_app.py, which had to be installed manually:

gradio
onnxruntime-gpu (as a dependency of rembg)
Please provide a working installation method for torchmcubes on Windows, or consider replacing it with a more cross-platform compatible dependency.

Thank you for your time and this incredible project.