Catalyst for Clustering-based Unsupervised Object Re-Identification: Feature Calibration [pdf]
The official repository for Catalyst for Clustering-based Unsupervised Object Re-Identification: Feature Calibration.
pip install -r requirements.txtWe recommend to use /Python=3.8 /torch=1.10.1 /torchvision=0.11.2 /timm=0.6.13 /cuda==11.3 /faiss-gpu=1.7.2/ 24G RTX 3090 or RTX 4090 for training and evaluation. If you find some packages are missing, please install them manually.
mkdir dataDownload the datasets:
- Market-1501
- MSMT17
- LUPerson.
- We don't have the copyright of the LUPerson dataset. Please contact authors of LUPerson to get this dataset.
- You can download the file list ordered by the CFS score for the LUPerson. [CFS_list.pkl]
Then unzip them and rename them under the directory like
data
├── market1501
│ └── bounding_box_train
│ └── bounding_box_test
│ └── ..
├── MSMT17
│ └── train
│ └── test
│ └── ..
└── DukeMTMC-reID
└── bounding_box_train
└── bounding_box_test
└── query
└── *.txt
| Model | Download |
|---|---|
| ViT-S/16 | link |
| ViT-S/16+ICS | link |
| ViT-B/16+ICS | link |
| Model | Download |
|---|---|
| Market-1501 | link |
| MSMT17 | link |
| DukeMTMC-reID | link |
Please download pre-trained models and put them into your custom file path.
sh train.sh
We have reproduced the performance to verify the reproducibility. The reproduced results may have a gap of about 0.5% with the numbers in the paper.
| Model | Image Size | mAP/Rank-1 | Download |
|---|---|---|---|
| ViT-S/16 | 256*128 | 88.6/94.9 | model |
| Model | Image Size | mAP/Rank-1 | Download |
|---|---|---|---|
| ViT-S/16 | 256*128 | 49.7/74.8 | model |
| Model | Image Size | mAP/Rank-1 | Download |
|---|---|---|---|
| ViT-S/16 | 256*128 | 71.9/83.8 | model |
Our implementation is mainly based on the following codebases. We gratefully thank the authors for their wonderful works.
LUPerson, DINO, TransReID, cluster-contrast-reid, TransReID-SSL
If you find this code useful for your research, please cite our paper
wating
If you have any question, please feel free to contact us. E-mail: qingsonghu08@gmail.com