After the core step, comes the crucial. This installation gave me a headache in the past because Open3D was there, and yet refusing to work with PyTorch. But this is a tutorial to keep you from it.
First, install Open3D and the dependencies of ML via conda:
- conda install -c open3d-admin open3d
- conda install -c conda-forge scikit-learn addict pandas plyfile tqdm
- conda install -c anaconda pyyaml
And check the installation:
- python3 -c "import open3d as o3d; print(o3d._build_config)"
- {'BUILD_TENSORFLOW_OPS': True, 'BUILD_PYTORCH_OPS': True, 'BUILD_CUDA_MODULE': True, 'BUILD_AZURE_KINECT': False, 'BUILD_LIBREALSENSE': False, 'BUILD_SHARED_LIBS': False, 'BUILD_GUI': True, 'BUILD_RPC_INTERFACE': True, 'ENABLE_HEADLESS_RENDERING': False, 'BUILD_JUPYTER_EXTENSION': True, 'BUNDLE_OPEN3D_ML': True, 'GLIBCXX_USE_CXX11_ABI': False, 'CMAKE_BUILD_TYPE': 'Release', 'CUDA_VERSION': '10.1', 'CUDA_GENCODES': '-gencode arch=compute_30,code=compute_30 -gencode arch=compute_75,code=compute_75', 'Tensorflow_VERSION': '2.3.1', 'Pytorch_VERSION': '1.6.0+cu101', 'WITH_OPENMP': True}
If the highlighted parts are empty or different in your output, then, something went wrong. It did not find CUDA and the ML (Machine Learning) module was disabled/not installed.
Note that it is not printing PyTorch 1.6 because we installed it first. There is no other version that works with Open3D-ML. You may already have PyTorch 1.4 or 1.8 installed before all this tutorial, but it doesn't matter, ML won't accept it.
Continuing the checks:
- python3 -c "import open3d.ml as o3d"
- Doesn't print anything, no errors.
- conda install -c anaconda matplotlib
- python3 -c "import open3d.ml.torch as o3d"
- Doesn't print anything, no errors.
No comments:
Post a Comment