[Bug] pip install -e . fails in mmtracking 0.13.0 due to ModuleNotFoundError: No module named 'torch'

[Bosung-Baek]

Title:

[Bug] pip install -e . fails in mmtracking 0.13.0 due to ModuleNotFoundError: No module named 'torch'

---

Description

While setting up mmtracking (version 0.13.0) following the recommended installation steps, the installation fails at pip install -e . with ModuleNotFoundError: No module named 'torch'.

Even though PyTorch is installed and works when imported manually (python -c "import torch; print(torch.__version__)"), the setup process does not detect it, causing an error.

---

Environment

• OS: Ubuntu 22.04
• Python Version: 3.9
• Conda Version: 22.9.0
• Pip Version: 25.0.1
• mmtracking Version: 0.13.0
• mmcv-full Version: 1.5.3
• CUDA Version: 11.7
• PyTorch Version: 1.13.1, 1.13.0(tried both)
• NumPy Version: 2.21.5

---

Steps to Reproduce

1. Create Conda environment and install PyTorch

   conda create -n deepsvc python=3.9 -y
   conda activate deepsvc
   conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia

2. Install mmcv-full

   pip install mmcv-full==1.5.3 -f https://download.openmmlab.com/mmcv/dist/cu117/torch1.13/index.html

   • Issue: The installation gets stuck at building wheel for mmcv-full.

3. Workaround for mmcv-full

   • Clone mmcv repository (1.5.3 version)
   • Install manually:

     git clone -b v1.5.3 https://github.com/open-mmlab/mmcv.git
     cd mmcv
     MMCV_WITH_OPS=1 pip install -e .

   • Issue: Error occurs due to missing pyproject.toml.

4. Fix: Add pyproject.toml with the following content and install using PEP 517:

   [build-system]
   requires = ["setuptools>=64", "wheel"]
   build-backend = "setuptools.build_meta"

   pip install -e . --use-pep517

   • Success: mmcv-full installs successfully.

5. Download and install mmtracking

   git clone -b v0.13.0 https://github.com/open-mmlab/mmtracking.git
   cd mmtracking
   pip install -v -e .

6. Error occurs:

   ModuleNotFoundError: No module named 'torch'
   error: subprocess-exited-with-error
   

---

Observed Behavior

Even though PyTorch is correctly installed and works when imported manually (python -c "import torch; print(torch.__version__)"), the pip install -e . process does not recognize it and fails.

---

What I Have Tried

1. Checked PyTorch installation

   python -c "import torch; print(torch.__version__)"

   • Output: 1.13.1 (PyTorch is installed correctly)

2. Reinstalled dependencies

   pip install --upgrade pip setuptools wheel
   pip install --force-reinstall torch torchvision torchaudio

   • Issue persists.

3. Checked Python environment

   conda list | grep torch

   • PyTorch is installed in the correct environment.

4. Tried a fresh Conda environment and reinstalled everything

   conda remove --name deepsvc --all
   conda create -n deepsvc python=3.9 -y
   conda activate deepsvc
   conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
   pip install mmcv-full==1.5.3 -f https://download.openmmlab.com/mmcv/dist/cu117/torch1.13/index.html
   git clone -b v0.13.0 https://github.com/open-mmlab/mmtracking.git
   cd mmtracking
   pip install -v -e .

   • Same error occurs.

---

Expected Behavior

pip install -e . should complete successfully without errors.

---

Additional Information

• It seems like setuptools does not detect the installed torch package during the editable installation process.
• Other users facing a similar issue with mmtracking may need to check whether pyproject.toml is required for installation.
• Any insights on whether this is a setuptools issue or specific to mmtracking would be helpful.

Would appreciate any guidance on resolving this! 🚀