Customized Torchreid for pyppbox

Torchreid is a library for deep-learning person re-identification using PyTorch, and pyppbox-torchreid is a customized Torchreid for pyppbox and:

• Ensures that Cython natively works on all OS platforms (Windows/Linux/macOS),
• Enables freedom of passing local model/weight files from anywhere,
• Disables some models which are not used in pyppbox.

All source credit and more info -> Original KaiyangZhou's repo.

Install

Use the pre-built wheel in releases or install from PyPI:

pip install pyppbox-torchreid

Or install from GitHub repo directly (Require C++ compiler):

pip install git+https://github.com/rathaumons/torchreid-for-pyppbox.git

To be able to run, you must install OpenCV and PyTorch; for example, with GPU support:

pip install opencv-contrib-python
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Note: PyTorch doesn't have GPU support for macOS.

Build from source

(Optional, auto install) Building wheels/source distribution requires these modules:

pip install "setuptools>=67.2.0"
pip install "Cython>=0.29.32"
pip install "numpy>=1.23.5"

This requires C++ compiler and is recommended using build:

git clone https://github.com/rathaumons/torchreid-for-pyppbox/
cd torchreid-for-pyppbox
python -m pip install --upgrade pip
pip install wheel build
python -m build --sdist
python -m build --wheel
cd dist

After you install pyppbox-torchred, OpenCV and PyTorch, you can check Cython rank_cy:

cd pyppbox_torchreid/metrics/rank_cylib
python test_cython.py