tntek / sclm

Semantic Consistency Learning on Manifold for Source Data-absent Unsupervised Domain Adaptation

SCLM

Code (pytorch) for Semantic Consistency Learning on Manifold for Source Data-absent Unsupervised Domain Adaptation on Office-31, Office-Home, VisDA-C. This article have been accepted by Neural Networks.

Framework

Preliminary

You need to download the Office-31, Office-Home, VisDA-C dataset, modify the path of images in each '.txt' under the folder './data/'.

The experiments are conducted on one GPU (NVIDIA RTX TITAN).

• python == 3.7.10
• pytorch ==1.6.0
• torchvision == 0.7.0

Training and evaluation

1. First training model on the source data, Office-31 dataset is shown here.

cd ./object
python SCLM_source.py --trte val --da uda --gpu_id 0  --output Office31/r0/src/ --dset office --max_epoch 100 --s 0 --seed 2020

2. Then adapting source model to target domain, with only the unlabeled target data.

python SCLM_target.py --da uda --gpu_id 0 --cls_par 0.3 --cls_snt 0.1 --s 0 --t 1  --output_src Office31/r0/src/ --output Office31/r0/sclm/  --dset office --lr 1e-2 --net resnet50 --seed 2020

Please refer to ./object/run.sh for all the settings for different methods and scenarios.

Results

All results of SCLM on three datasets is under the folder './results/'.

Citation

@article{tang2022semantic, title={Semantic consistency learning on manifold for source data-free unsupervised domain adaptation}, author={Tang, Song and Zou, Yan and Song, Zihao and Lyu, Jianzhi and Chen, Lijuan and Ye, Mao and Zhong, Shouming and Zhang, Jianwei}, journal={Neural Networks}, volume={152}, pages={467-478}, year={2022}, publisher={Elsevier} }

Acknowledgement

The code is based on DeepCluster(ECCV 2018) and SHOT (ICML 2020, also source-free).

Contact

• [email protected]