Transformer-Based Multi-Grained Features for Unsupervised Person Re-Identification (TMGF)

Official implementation of paper Transformer-Based Multi-Grained Features for Unsupervised Person Re-Identification (WACV2023 workshop).

In this work, a dual-branch network based upon ViT is build to generate different granularities of part features from local tokens, learning together with global features for better discriminating capacity. Extensive experiments on three person Re-ID datasets are conducted and show that the proposed method achieves the state-of-the-art performance in unsupervised methods.

Prerequisites

Pretrained Weight

TMGF is fine-tuned on the pretrained weight, which can be found from TransReID-SSL. In our model, we use ViT-S/16+ICS. Download here.

Installation

Clone this repo and extract the files.

We recommand conda to create a virtual Python 3.7 environment and install all requirements in it. Extra packages are listed in requirements.txt and can be installed by pip:

conda create -n torch1.6 python=3.7
conda activate torch1.6

pip install -r requirements.txt

Training

Download the datasets and put them into the right place. Check and run the shell script train.sh:

CUDA_VISIBLE_DEVICES=0 ./train.sh # run on GPU 0

We use yacs config system for better parameter management. You may need to modify the dataset root directory DATASET.ROOT_DIR and the pretrained weight path MODEL.PRETRAIN_PATH. Check here to learn how to change configs as you like.

2023/8/18: We fix a typo of nn.DataParallel misusage in prepare_model.py, which may cause some bugs in forwarding.

Evaluation

You can run evaluation on any datasets with model weight provided.

CUDA_VISIBLE_DEVICES=0 python evaluate.py --weight /path/to/model/weight.pth --conf configs/TMGF_full.yml # run on GPU 0

Performance

Acknowledgement

We would like to sincerely thank TransReID, TransReID-SSL and O2CAP for their insightful ideas and outstanding works!

Citation

If you feel our work helpful in your research, please cite it like this:

@InProceedings{Li_2023_WACV,
    author    = {Li, Jiachen and Wang, Menglin and Gong, Xiaojin},
    title     = {Transformer Based Multi-Grained Features for Unsupervised Person Re-Identification},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops},
    month     = {January},
    year      = {2023},
    pages     = {42-50}
}