如何解决从Git仓库安装detectron2时依赖torch的安装报错问题？

尝试部署依赖detectron2的Streamlit应用，本地通过python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'安装detectron2时，必须预先安装torch，否则会直接报错。执行pip freeze > requirements.txt生成的依赖文件内容如下：

torch==1.12.1
torchaudio==0.12.1
torchvision==0.13.1
detectron2 @ git+https://github.com/facebookresearch/detectron2.git@717ab9f0aeca216a2f800e43d705766251ba3a55

但运行pip install -r requirements.txt时触发报错，核心错误是detectron2的setup.py在生成元数据阶段找不到torch模块，具体报错信息如下：

Collecting detectron2@ git+https://github.com/facebookresearch/detectron2.git@717ab9f0aeca216a2f800e43d705766251ba3a55
  Cloning https://github.com/facebookresearch/detectron2.git (to revision 717ab9f0aeca216a2f800e43d705766251ba3a55) to /tmp/pip-install-60odo0ml/detectron2_3c0ef1b35e7b42cd95cfd938e09dd4d5
  Running command git clone --filter=blob:none --quiet https://github.com/facebookresearch/detectron2.git /tmp/pip-install-60odo0ml/detectron2_3c0ef1b35e7b42cd95cfd938e09dd4d5
  Running command git rev-parse -q --verify 'sha^717ab9f0aeca216a2f800e43d705766251ba3a55'
  Running command git fetch -q https://github.com/facebookresearch/detectron2.git 717ab9f0aeca216a2f800e43d705766251ba3a55
  Resolved https://github.com/facebookresearch/detectron2.git to commit 717ab9f0aeca216a2f800e43d705766251ba3a55
  Preparing metadata (setup.py) ... error
  error: subprocess-exited-with-error
  
  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [6 lines of output]
      Traceback (most recent call last):
        File "<string>", line 36, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "/tmp/pip-install-60odo0ml/detectron2_3c0ef1b35e7b42cd95cfd938e09dd4d5/setup.py", line 10, in <module>
          import torch
      ModuleNotFoundError: No module named 'torch'
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

方法1：分阶段安装依赖 #

pip处理requirements.txt时不会严格按顺序安装，尤其是git源包会优先处理元数据生成，导致torch还未安装就先处理detectron2。可以拆分安装步骤：

1. 先安装torch相关依赖：

   pip install torch==1.12.1 torchaudio==0.12.1 torchvision==0.13.1
   

2. 再安装指定版本的detectron2：

   pip install 'git+https://github.com/facebookresearch/detectron2.git@717ab9f0aeca216a2f800e43d705766251ba3a55'
   

在Streamlit部署配置中，把这两个命令按顺序添加到部署脚本即可，替代一次性安装requirements.txt。

方法2：拆分requirements文件 #

将依赖分成两个文件，强制先安装torch再安装detectron2：

1. 创建torch-reqs.txt：

   torch==1.12.1
   torchaudio==0.12.1
   torchvision==0.13.1
   

2. 创建detectron2-reqs.txt：

   detectron2 @ git+https://github.com/facebookresearch/detectron2.git@717ab9f0aeca216a2f800e43d705766251ba3a55 --no-deps
   

然后执行：

pip install -r torch-reqs.txt && pip install -r detectron2-reqs.txt

添加--no-deps是避免detectron2的setup.py尝试处理依赖，确保torch已提前安装完成。

方法3：使用预编译wheel包（推荐部署使用） #

detectron2提供对应torch版本的预编译wheel包，无需源码编译，也就不会触发setup.py中的torch导入操作：

1. 确认你的torch版本和CUDA版本（比如torch 1.12.1+cu113），使用对应链接安装：

   pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.12/index.html
   

2. 修改requirements.txt，加入find-links指定wheel源：

   torch==1.12.1
   torchaudio==0.12.1
   torchvision==0.13.1
   detectron2
   --find-links https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.12/index.html
   

这样pip会先安装torch，再从指定链接拉取预编译的detectron2包，跳过源码编译步骤。

内容的提问来源于stack exchange，提问作者Deshwal