关于Open3D-ML导入open3d.ml.torch报错「Exception: Open3D was not built with PyTorch support!」的问题求助

Let’s break down the possible causes for this error based on your environment setup:

• Version Compatibility Mismatch
  Open3D 0.12.0 has specific compatibility requirements with PyTorch versions. PyTorch 1.6.0 might be too old for this Open3D release—Open3D 0.12.0 typically aligns with newer PyTorch versions (like 1.7+). This mismatch could prevent the Open3D build from recognizing your PyTorch installation during setup.

• Incorrect Installation Method
  If you installed Open3D via the standard pip install open3d, the pre-built Windows binaries might not include PyTorch support by default. Open3D-ML requires a build that explicitly enables PyTorch integration. You might need to either:

  • Install the Open3D-ML package alongside Open3D using pip install open3d-ml (make sure it’s compatible with your Open3D 0.12.0 version), or
  • Compile Open3D from source with the BUILD_PYTORCH flag enabled—this is especially critical for Windows, as pre-built packages sometimes omit this feature.

• CUDA Environment Misconfiguration
  Even though you have CUDA 10.1 installed, Open3D might not have detected it during installation. While PyTorch 1.6.0 does support CUDA 10.1, ensure your system’s CUDA paths are properly set in environment variables (like CUDA_PATH and PATH pointing to CUDA 10.1 binaries). If compiling from source, verify that the build system picks up your CUDA and PyTorch installations correctly.

• Virtual Environment Isolation
  Double-check that both Open3D and PyTorch are installed in the same virtual environment. It’s easy to accidentally install packages in separate environments, which would lead to Open3D being unable to locate your PyTorch installation.

内容的提问来源于stack exchange，提问作者Yoann Le Gal