Skip to content

Latest commit

 

History

History
119 lines (99 loc) · 4.19 KB

README.md

File metadata and controls

119 lines (99 loc) · 4.19 KB
Raw

Python 3.6 PyTorch 1.1

Mutual Mean-Teaching (MMT)

The official implementation for the Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification which is accepted by ICLR-2020.

framework

Installation

git clone https://github.com/yxgeee/MMT.git
cd MMT
python setup.py install

Prepare Datasets

cd examples && mkdir data

Download the raw datasets DukeMTMC-reID, Market-1501, MSMT17, and then unzip them under the directory like

MMT/examples/data
├── dukemtmc
│   └── DukeMTMC-reID
├── market1501
│   └── Market-1501-v15.09.15
└── msmt17
    └── MSMT17_V1

Prepare Pre-trained Models

When training with the backbone of IBN-ResNet-50, you need to download the ImageNet pre-trained model from this link and save it under the path of logs/pretrained/.

mkdir logs && cd logs
mkdir pretrained

The file tree should be

MMT/logs
└── pretrained
    └── resnet50_ibn_a.pth.tar

Example #1:

Transferring from DukeMTMC-reID to Market-1501 on the backbone of ResNet-50, i.e. Duke-to-Market (ResNet-50).

Train

We utilize 4 GTX-1080TI GPUs for training.

Stage I: Pre-training on the source domain

sh scripts/pretrain.sh dukemtmc market1501 resnet50 1
sh scripts/pretrain.sh dukemtmc market1501 resnet50 2

Stage II: End-to-end training with MMT-500

sh scripts/train.sh dukemtmc market1501 resnet50 500

Test

We utilize 1 GTX-1080TI GPU for testing. Test the trained model with best performance by

sh scripts/test.sh market1501 resnet50 logs/dukemtmcTOmarket1501/resnet50-MMT-500/model_best.pth.tar

Other Examples:

Duke-to-Market (IBN-ResNet-50)

# pre-training on the source domain
sh scripts/pretrain.sh dukemtmc market1501 resnet_ibn50a 1
sh scripts/pretrain.sh dukemtmc market1501 resnet_ibn50a 2
# end-to-end training with MMT-500
sh scripts/train.sh dukemtmc market1501 resnet_ibn50a 500
# or MMT-700
sh scripts/train.sh dukemtmc market1501 resnet_ibn50a 700
# testing the best model
sh scripts/test.sh market1501 resnet_ibn50a logs/dukemtmcTOmarket1501/resnet_ibn50a-MMT-500/model_best.pth.tar
sh scripts/test.sh market1501 resnet_ibn50a logs/dukemtmcTOmarket1501/resnet_ibn50a-MMT-700/model_best.pth.tar

Duke-to-MSMT (ResNet-50)

# pre-training on the source domain
sh scripts/pretrain.sh dukemtmc msmt17 resnet50 1
sh scripts/pretrain.sh dukemtmc msmt17 resnet50 2
# end-to-end training with MMT-500
sh scripts/train.sh dukemtmc msmt17 resnet50 500
# or MMT-1000
sh scripts/train.sh dukemtmc msmt17 resnet50 1000
# testing the best model
sh scripts/test.sh msmt17 resnet50 logs/dukemtmcTOmsmt17/resnet50-MMT-500/model_best.pth.tar
sh scripts/test.sh msmt17 resnet50 logs/dukemtmcTOmsmt17/resnet50-MMT-1000/model_best.pth.tar

Download Trained Models

Source-domain pre-trained models and our MMT models can be downloaded from the link. results

Citation

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

@inproceedings{
  ge2020mutual,
  title={Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification},
  author={Yixiao Ge and Dapeng Chen and Hongsheng Li},
  booktitle={International Conference on Learning Representations},
  year={2020},
  url={https://openreview.net/forum?id=rJlnOhVYPS}
}