Home

Awesome

Python >=3.5 PyTorch >=1.0

[ECCV2022] Unstructured Feature Decoupling for Vehicle Re-Identification (UFDN)

Unstructured Feature Decoupling for Vehicle Re-Identification pdf

News

Pipeline

framework

Requirements

Installation

pip install -r requirements.txt
(we use /torch 1.7.1 /torchvision 0.8.2 /timm 0.3.2 /cuda 11.0 / 16G or 32G V100 for training and evaluation.)

Prepare Datasets

mkdir data

Download the vehicle datasets VehicleID, VeRi-776, VERIWILD. Then unzip them and rename them under the directory like

data
└── VeRi
    └── images ..
└── VehicleID
    └── images ..
└── VERI-WILD
    └── images ..

Datalist: VeRi-776

Prepare Res50 or Swin Pre-trained Models

You need to download the ImageNet pretrained transformer model : Res50, Swin-tiny.

Training

We utilize 1 GPU for training VeRi-776 Dataset

sh experiments/train_res50_UFDN_776.sh or train_swin_UFDN_776.sh

We utilize 1 GPU for training VehicleID Dataset

sh experiments/train_res50_UFDN_VehicleID.sh or train_swin_UFDN_VehicleID.sh

Trained Models and logs

We have reproduced the performance to verify the reproducibility. The reproduced results may have a gap of about 0.1-0.2% with the numbers in the paper.

Experiments
methodbackbonedatasetResultlogmodel
UFDNRes50VeRi-77681.5%/96.5%logmodel
UFDNSwin-tinyVeRi-77680.8%/96.5%logmodel

Acknowledgement

Codebase from reid-strong-baseline , pytorch-image-models, TransReID

Contact

If you have any question, please feel free to contact us. E-mail: qianwen2018@ia.ac.cn , haoluocsc@zju.edu.cn

Citation

If you find this code useful for your research, please cite our paper

@InProceedings{Qian_2022_ECCV,
    author    = {Qian, Wen and Luo, Hao and Peng, Silong and Wang, Fan and Chen, Chen and Li, Hao},
    title     = {Unstructured Feature Decoupling for Vehicle Re-Identification},
    booktitle = { European Conference on Computer Vision (ECCV)},
    month     = {October},
    year      = {2022},
}