tntek/n2dcx

Official implementation for [N2DCX] Nearest Neighborhood-Based Deep Clustering for Source Data-absent Unsupervised Domain Adaptation

Official implementation for [N2DCX]([2107.12585] Nearest Neighborhood-Based Deep Clustering for Source Data-absent Unsupervised Domain Adaptation (arxiv.org))

Code (pytorch) for ['Nearest Neighborhood-Based Deep Clustering for Source Data-absent Unsupervised Domain Adaptation']([2107.12585] Nearest Neighborhood-Based Deep Clustering for Source Data-absent Unsupervised Domain Adaptation (arxiv.org)) on Office-31, Office-Home, VisDA-C. This article is under review.

Framework

Datasets and Prerequisites

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 == 3.7.3
• pytorch ==1.6.0
• torchvision == 0.7.0

Training and evaluation

1. First training model on the source data, VisDA-C dataset is shown here.

cd ./object
~/anaconda3/bin/python N2DC_source.py --trte val --output ckps2020r0/source/ --da uda --gpu_id 0 --dset VISDA-C --net resnet101 --lr 1e-3 --max_epoch 10 --s 0

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

# train the target domain by N2DC
~/anaconda3/bin/python N2DC_target.py --cls_par 0.2 --da uda --dset VISDA-C --gpu_id 0 --s 0 --t 1 --output_src ckps2020r0/source/ --output ckps2020r0/target_n2dc/ --net resnet101 --lr 1e-3 --seed 2020

# train the target domain by N2DC-EX
~/anaconda3/bin/python N2DCEX_target.py --cls_par 0.2 --da uda --dset VISDA-C --gpu_id 0 --s 0 --t 1 --output_src ckps2020r0/source/ --output ckps2020r0/target_n2dcex/ --net resnet101 --lr 1e-3 --seed 2020

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

Results

The results of N2DCX is display under the folder './object/results/'.

Citation

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

@article{tang2021n2dcx, title={Nearest Neighborhood-Based Deep Clustering for Source Data-absent Unsupervised Domain Adaptation}, author={Song Tang, Yan Yang, Zhiyuan Ma, Norman Hendrich, Fanyu Zeng, Shuzhi Sam Ge, Changshui Zhang, Jianwei Zhang}, year={2021}, journal={arXiv:2107.12585}, url= {https://arxiv.org/abs/2107.12585} }

Acknowledgement

DeepCluster(ECCV 2018)'s work.

SHOT (ICML 2020, also source-free)'s work.

Contact

• [email protected]