DAPRH : GAN-based Data Augmentation and Pseudo-Label Refinement with Holistic Features for Unsupervised Domain Adaptation Person Re-Identification
Official PyTorch implementation of GAN-based Data Augmentation and Pseudo-Label Refinement with Holistic Features for Unsupervised Domain Adaptation Person Re-Identification (Knowledge-Based Systems journal) (2023).
- [06/2024] Official code is released.
- [04/2024] Great news!!! Our paper is published and available in Sciencedirect. Please check in the citation section [below](## Citation)
- [05/2023] Push Initial rep on Github and Release unofficial code, results, pretrained models
- [02/2023] Citation, official code (when the paper is accepted)
git clone https://github.com/ewigspace1910/DAPRH.git
conda create --name DAPRH
conda activate DAPRH
pip install -r requirements.txt- Download the object re-ID datasets Market-1501, Duke-MTMC, MSMT17 from here, then move to /datasets. The directory should look like:
DAPRH/data
├── Market-1501-v15.09.15
├── DukeMTMC-reID
├── MSMT17
|── ....
├── 4Gan #(Must be created)
└── SystheImgs #(Must be created)
- To prepare data for GAN, we setup into /datasets/4Gan as following:
DAPRH/data/4Gan
├── duke2market
| ├──train
| | ├──dukemtmc
| | ├──market1501c0
| | ├──market1501c1
| | ├──market1501c2
| | ├──market1501c3
| | ├──market1501c4
| | ├──market1501c5
| ├──test
| | ├──dukemtmc
| | ├──market1501c0
| | ├──market1501c1
| | ├──market1501c2
| | ├──market1501c3
| | ├──market1501c4
| | ├──market1501c5
├── market2duke
| ├──...
├── ...
We utilize 1 Tesla T4 GPU 16G for training. We use 256x128 sized images for Market-1501, DukeMTMC and MSMT17 in both Training-GAN and Training-Reid
- For convenient, we utilize bash script to setup commands. You can reuse or modify them in './DAPRH/scripts'
- setup env
cd DAPRH/stargan
conda activate DAPRH- for duke-->market:
# Train StarGAN on custom datasets
LABEL_DIM=7
CROP_SIZE=128
IMG_SIZE=128
TRAIN_IMG_DIR="../../datasets/4GAN/duke2mark/train"
BATCHSIZE=16
Lidt=1
Lrec=10
Lgp=10
Lcls=1
python main.py --mode train --dataset RaFD --rafd_crop_size $CROP_SIZE --image_size $IMG_SIZE \
--c_dim $LABEL_DIM --rafd_image_dir $TRAIN_IMG_DIR --batch_size $BATCHSIZE\
--sample_dir ../../saves/Gan-duke2mark/samples \
--log_dir ../../saves/Gan-duke2mark/logs \
--model_save_dir ../../saves/Gan-duke2mark/models \
--result_dir ../../saves/Gan-duke2mark/results \
--lambda_idt $Lidt \
--lambda_rec $Lrec \
--lambda_gp $Lgp --lambda_cls $Lcls- for market-->duke:
# Train StarGAN on custom datasets
LABEL_DIM=7
CROP_SIZE=128
IMG_SIZE=128
TRAIN_IMG_DIR="../../datasets/4GAN/market2duke/train"
BATCHSIZE=16
Lidt=1
Lrec=10
Lgp=10
Lcls=1
python main.py --mode train --dataset RaFD --rafd_crop_size $CROP_SIZE --image_size $IMG_SIZE \
--c_dim $LABEL_DIM --rafd_image_dir $TRAIN_IMG_DIR --batch_size $BATCHSIZE\
--sample_dir ../../saves/Gan-mark2duke/samples \
--log_dir ../../saves/Gan-mark2duke/logs \
--model_save_dir ../../saves/Gan-mark2duke/models \
--result_dir ../../saves/Gan-mark2duke/results \
--lambda_idt $Lidt \
--lambda_rec $Lrec \
--lambda_gp $Lgp --lambda_cls $Lcls- After training, we can use trained-GAN models to gen synthetic datasets for reID :
# 4 duke2market
LABEL_DIM=7
CROP_SIZE=128
IMG_SIZE=128
TRAIN_IMG_DIR="../../datasets/4Gan/duke2mark/train/dukemtmc"
FAKEDIR="../../datasets/SyntheImgs/duke2mark" #!!!!
BATCHSIZE=1 #!!!!
ITER=200000
DOMAIN=0
python main.py --mode sample --dataset RaFD --rafd_crop_size $CROP_SIZE --image_size $IMG_SIZE \
--c_dim $LABEL_DIM --rafd_image_dir $TRAIN_IMG_DIR --batch_size $BATCHSIZE\
--sample_dir ../../saves/Gan-duke2mark/samples \
--log_dir ../../saves/Gan-duke2mark/logs \
--model_save_dir ../../saves/Gan-duke2mark/models \
--result_dir ../../saves/Gan-duke2mark/results \
--test_iters $ITER --except_domain=$DOMAIN \ #!!!!
--pattern "{ID}_{CX}_f{RANDOM}.jpg" \ #!!!!
--gen_dir $FAKEDIR #!!!!
#############################
# market4duke
LABEL_DIM=7
CROP_SIZE=128
IMG_SIZE=128
TRAIN_IMG_DIR="../../datasets/4Gan/mark2duke/train/dukemtmc"
FAKEDIR="../../datasets/SyntheImgs/mark2duke"
BATCHSIZE=1
ITER=200000
DOMAIN=6
python main.py --mode sample --dataset RaFD --rafd_crop_size $CROP_SIZE --image_size $IMG_SIZE \
--c_dim $LABEL_DIM --rafd_image_dir $TRAIN_IMG_DIR --batch_size $BATCHSIZE\
--sample_dir ../../saves/Gan-mark2duke/samples \
--log_dir ../../saves/Gan-mark2duke/logs \
--model_save_dir ../../saves/Gan-mark2duke/models \
--result_dir ../../saves/Gan-mark2duke/results \
--test_iters $ITER --except_domain=$DOMAIN \
--pattern "{ID}_{CX}_f{RANDOM}.jpg" \
--gen_dir $FAKEDIR
- activate env
cd DAPRH
conda activate DAPRH
- training on lable domain with both real and fake images + DIM:
#for example
python _source_pretrain.py \
-ds "dukemtmc" -dt "market1501" \
-a "resnet50" --iters 200 \
--num-instances 16 -b 128 --margin 0.3 \
--warmup-step 10 --lr 0.00035 --milestones 40 70 --epochs 80 --eval-step 1 \
--logs-dir "../saves/reid/duke2market/S1/R50Mix" \
--data-dir "../datasets" \
--fake-data-dir "../datasets/SystheImgs" \
--ratio 4 1 \
--dim --lamda 0.05
python _source_pretrain.py \
-dt "dukemtmc" -ds "market1501" \
-a "resnet50" --iters 200 \
--num-instances 16 -b 128 --margin 0.3 \
--warmup-step 128 --lr 0.00035 --milestones 40 70 --epochs 80 --eval-step 1 \
--logs-dir "../saves/reid/market2duke/S1/R50Mix" \
--data-dir "../datasets" \
--fake-data-dir "../datasets/SystheImgs" \
--ratio 4 1 \
--dim --lamda 0.05- Note: U can modify file
./modules/datasets/synimgs.pyin linendictto adapt stucture of path to Fake Image Folder.
- Modify the below script to excute:
### DUKE ---> MARKET ####
python _target_finetune.py \
-dt "market1501" -b 128 --num-instances 16 \
-a resnet50mulpart --epochs 26 --iters 400 --npart 2 \
--logs-dir "../saves/reid/duke2market/S2/finetune" \
--init "../saves/reid/duke2market/S1/R50Mix/model_best.pth.tar" \
--data-dir "../datasets" \
--pho 0 --uet-al 0.8 \
--ece 0.4 --etri 0.6 --ema
### MARKET ---> DUKE ####
python _target_finetune.py \
-dt "dukemtmc" -b 128 --num-instances 16 \
-a resnet50mulpart --epochs 26 --iters 400 --npart 2 \
--logs-dir "../saves/reid/market2duke/S2/finetune" \
--init "../saves/reid/market2duke/S1/R50Mix/model_best.pth.tar" \
--data-dir "../datasets" \
--gtri-weight 1 --gce-weight 1
--pho 0. --uet-al 0.5 \
--ece 0.4 --etri 0.6 --ema - for convenience, you can reuse directly scripts in
./DAPRH/scripts
python test_model.py \
-dt "market1501" --data-dir "../datasets" \
-a resnet50 --features 0 -b 128 \
--resume ".../model_best.pth.tar" \
--rerank #optional
- you can also download pretrained model from modelrepo
If you find this code useful for your research, please consider citing our paper
@article{PHAM2024111471,
title = {GAN-based data augmentation and pseudo-label refinement with holistic features for unsupervised domain adaptation person re-identification},
journal = {Knowledge-Based Systems},
pages = {111471},
year = {2024},
issn = {0950-7051},
doi = {https://doi.org/10.1016/j.knosys.2024.111471},
url = {https://www.sciencedirect.com/science/article/pii/S0950705124001060},
author = {Dang H. Pham and Anh D. Nguyen and Hoa N. Nguyen},
keywords = {Unsupervised person re-identification, Unsupervised domain adaptation, GAN-based data augmentation, Pseudo-label refinement}
}
@InProceedings{cDAUET-23,
author="Nguyen, Anh D.
and Pham, Dang H.
and Nguyen, Hoa N.",
title="GAN-Based Data Augmentation and Pseudo-label Refinement for Unsupervised Domain Adaptation Person Re-identification",
booktitle="Computational Collective Intelligence",
year="2023",
publisher="Springer Nature Switzerland",
address="Cham",
doi={10.1007/978-3-031-41456-5_45},
pages="591--605",
isbn="978-3-031-41456-5"
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent