This is the implementation codes for Reliability exploration with self-ensemble learning for domain adaptive person re-identification

• numpy
• torch==1.8.1
• torchvision
• six
• h5py
• Pillow
• scipy
• scikit-learn
• metric-learn
• metric-learn
• faiss_gpu==1.6.3
• dropblock==0.3.0
• pyyaml
• yacs
• termcolor
• tabulate
• gdown

you should prepare dataset like ECN /JVTC and unzip each dataset and corresponding CamStyle under 'ECN/data/' as following:

.
+-- examples/data
|   +-- Market-1501-v15.09.15
|           +-- bounding_box_train
|           +-- query
|           +-- bounding_box_test
|           +-- bounding_box_train_camstyle
|   +-- DukeMTMC-reID
|           +-- bounding_box_train
|           +-- query
|           +-- bounding_box_test
|           +-- bounding_box_train_camstyle
|   +-- MSMT17_V1
|           +-- bounding_box_train
|           +-- bounding_box_test
|           +-- bounding_box_train_camstyle
|           +-- query
+ -- other files in this repo

You can download the dataset from: https://pan.baidu.com/s/1vwApZ0St6KBWMapMGq5ALw 提取码：o3mk

We run our models on one NVIDIA 3090 GPUs.

1. Pretrained on source domain. Using the fast-reid to train ReSL model on the source domain dataset

CUDA_VISIBLE_DEVICES=0 python tools/train_net.py --config-file configs/Unsupervised/sbs_R50_resl_duke_supervised.yml 

CUDA_VISIBLE_DEVICES=0 python tools/train_net.py --config-file configs/Unsupervised/sbs_R50_resl_market_supervised.yml 

2. Training on target dataset

CUDA_VISIBLE_DEVICES=0  python tools/train_net_unsupervised.py --config-file configs/Unsupervised/sbs_R50_resl_market.yml MODEL.WEIGHTS /path/to/checkpoint_file

CUDA_VISIBLE_DEVICES=0 python tools/train_net_unsupervised.py --config-file configs/Unsupervised/sbs_R50_resl_duke.yml  MODEL.WEIGHTS /path/to/checkpoint_file

CUDA_VISIBLE_DEVICES=0 python tools/train_net_unsupervised.py --config-file configs/Unsupervised/sbs_R50_resl_msmt17.yml  MODEL.WEIGHTS /path/to/checkpoint_file

CUDA_VISIBLE_DEVICES=0  python tools/train_net_unsupervised.py --config-file configs/Unsupervised/sbs_R50_resl_market.yml

CUDA_VISIBLE_DEVICES=0 python tools/train_net_unsupervised.py --config-file configs/Unsupervised/sbs_R50_resl_duke.yml 

CUDA_VISIBLE_DEVICES=0 python tools/train_net_unsupervised.py --config-file configs/Unsupervised/sbs_R50_resl_msmt17.yml