No module named torch when installing GLIP

[Kenneth-Wong]

I want to ask a question that when I install GLIP, why there still exists an error: No module named torch. I have installed pytorch.

[huwem]

创建或修改 pyproject.toml 文件，添加构建系统要求：

在GLIP目录下创建 pyproject.toml 文件（如果不存在）：
[build-system]
requires = ["setuptools", "wheel", "torch>=1.8.0", "torchvision"]
build-backend = "setuptools.build_meta"
这样可以确保在构建过程中有正确的依赖项。
清理构建缓存
rm -rf build/
重新安装

Create or modify pyproject.toml file, add build system requirements:

Create pyproject.toml file in GLIP directory (if it doesn't exist):
[build-system]
requires = ["setuptools", "wheel", "torch>=1.8.0", "torchvision"]
build-backend = "setuptools.build_meta"
This ensures that the correct dependencies are available during the build process.
Clean build cache
rm -rf build/
Reinstall

[wwu928706-lab]

I encountered this issue as well. I tried creating a pyproject.toml file with the following content:

[build-system]
requires = ["setuptools", "wheel", "torch>=1.9.1", "torchvision"]
build-backend = "setuptools.build_meta"

After rerunning the installation, I got the following error:

/home/demo/sgnav/GLIP/maskrcnn_benchmark/csrc/cuda/ROIAlign_cuda.cu:5:10: fatal error: THC/THC.h: No such file or directory
5 | #include <THC/THC.h>
| ^~~~~~~~~~~
compilation terminated.
error: command '/usr/local/cuda-12.8/bin/nvcc' failed with exit code 1
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building editable for maskrcnn_benchmark
Failed to build maskrcnn_benchmark
error: failed-wheel-build-for-install