0安装 CUDA 12.1 版本的 PyTorch与系统 CUDA 一致pip install torch2.1.0 torchvision0.16.0 torchaudio2.1.0 --index-url https://download.pytorch.org/whl/cu121python -c import torch; print(torch.__version__); print(torch.cuda.is_available())1反复尝试pip install setuptools系统提示Requirement already satisfied但程序依然无法运行。参考这个博主【已解决】ModuleNotFoundError: No module named ‘pkg_resources‘ | Setuptools 82.0.0 版本断层深度复盘-CSDN博客Traceback (most recent call last):File ..., line 18, in moduleimport pkg_resourcesModuleNotFoundError: No module named pkg_resources# 强制重新安装 81.0.0 版本python -m pip install setuptools82.0.0 --force-reinstall2diff-gaussian-rasterization安装失败关于PyTorch 2.1.0 内置 pybind11 CUDA 12.1 GCC 12.4这一编译组合不适配的问题。手动设置了CPATH指向 GLM 路径git clone https://github.com/g-truc/glm.git。export CPATH/home/lfw/seasplat/glm:$CPATH直接在 Conda 环境内安装 GCC 11conda activate env conda install -y \ anaconda::gcc_linux-6411.2.0 \ anaconda::gxx_linux-6411.2.0查找实际路径ls $CONDA_PREFIX/bin/*conda-linux-gnu-gcc \ $CONDA_PREFIX/bin/*conda-linux-gnu-g 2/dev/null设置当前终端export CC$(compgen -G $CONDA_PREFIX/bin/*conda-linux-gnu-gcc | head -1) export CXX$(compgen -G $CONDA_PREFIX/bin/*conda-linux-gnu-g | head -1) export CUDAHOSTCXX$CXX echo CC$CC echo CXX$CXX $CC --version $CXX --version然后重新编译git clone https://github.com/dxyang/diff-gaussian-rasterization(如果项目拉取不下来) cd /home/use/env/submodules/diff-gaussian-rasterization export CUDA_HOME/usr/local/cuda-12.1 export PATH$CUDA_HOME/bin:$PATH export LD_LIBRARY_PATH$CUDA_HOME/lib64:${LD_LIBRARY_PATH:-} export TORCH_CUDA_ARCH_LIST8.6 export MAX_JOBS1 export NVCC_PREPEND_FLAGS-ccbin$CXX rm -rf build dist *.egg-info python -m pip install -v --no-build-isolation . \ 21 | tee /home/use/env/rasterizer-gcc11.logpython - PY import torch import diff_gaussian_rasterization from diff_gaussian_rasterization import GaussianRasterizer from simple_knn._C import distCUDA2 print(PyTorch:, torch.__version__) print(CUDA:, torch.version.cuda) print(Rasterizer path:, diff_gaussian_rasterization.__file__) print(diff_gaussian_rasterization: OK) print(simple_knn: OK) PY
0安装 CUDA 12.1 版本的 PyTorch与系统 CUDA 一致pip install torch2.1.0 torchvision0.16.0 torchaudio2.1.0 --index-url https://download.pytorch.org/whl/cu121python -c import torch; print(torch.__version__); print(torch.cuda.is_available())1反复尝试pip install setuptools系统提示Requirement already satisfied但程序依然无法运行。参考这个博主【已解决】ModuleNotFoundError: No module named ‘pkg_resources‘ | Setuptools 82.0.0 版本断层深度复盘-CSDN博客Traceback (most recent call last):File ..., line 18, in moduleimport pkg_resourcesModuleNotFoundError: No module named pkg_resources# 强制重新安装 81.0.0 版本python -m pip install setuptools82.0.0 --force-reinstall2diff-gaussian-rasterization安装失败关于PyTorch 2.1.0 内置 pybind11 CUDA 12.1 GCC 12.4这一编译组合不适配的问题。手动设置了CPATH指向 GLM 路径git clone https://github.com/g-truc/glm.git。export CPATH/home/lfw/seasplat/glm:$CPATH直接在 Conda 环境内安装 GCC 11conda activate env conda install -y \ anaconda::gcc_linux-6411.2.0 \ anaconda::gxx_linux-6411.2.0查找实际路径ls $CONDA_PREFIX/bin/*conda-linux-gnu-gcc \ $CONDA_PREFIX/bin/*conda-linux-gnu-g 2/dev/null设置当前终端export CC$(compgen -G $CONDA_PREFIX/bin/*conda-linux-gnu-gcc | head -1) export CXX$(compgen -G $CONDA_PREFIX/bin/*conda-linux-gnu-g | head -1) export CUDAHOSTCXX$CXX echo CC$CC echo CXX$CXX $CC --version $CXX --version然后重新编译git clone https://github.com/dxyang/diff-gaussian-rasterization(如果项目拉取不下来) cd /home/use/env/submodules/diff-gaussian-rasterization export CUDA_HOME/usr/local/cuda-12.1 export PATH$CUDA_HOME/bin:$PATH export LD_LIBRARY_PATH$CUDA_HOME/lib64:${LD_LIBRARY_PATH:-} export TORCH_CUDA_ARCH_LIST8.6 export MAX_JOBS1 export NVCC_PREPEND_FLAGS-ccbin$CXX rm -rf build dist *.egg-info python -m pip install -v --no-build-isolation . \ 21 | tee /home/use/env/rasterizer-gcc11.logpython - PY import torch import diff_gaussian_rasterization from diff_gaussian_rasterization import GaussianRasterizer from simple_knn._C import distCUDA2 print(PyTorch:, torch.__version__) print(CUDA:, torch.version.cuda) print(Rasterizer path:, diff_gaussian_rasterization.__file__) print(diff_gaussian_rasterization: OK) print(simple_knn: OK) PY
