pinglmlcv / BA3US

code for our ECCV 2020 paper "A Balanced and Uncertainty-aware Approach for Partial Domain Adaptation"

Code for our ECCV (2020) paper A Balanced and Uncertainty-aware Approach for Partial Domain Adaptation.

Prerequisites:

• python == 3.6.8
• pytorch ==1.1.0
• torchvision == 0.3.0
• numpy, scipy, PIL, argparse, tqdm

Dataset:

• Please manually download the datasets Office, Office-Home, ImageNet-Caltech from the official websites, and modify the path of images in each '.txt' under the folder './data/'.
• We adopt the same data protocol as PADA.

Training:

1. Partial Domain Adaptation (PDA) on the Office-Home dataset [Art(s=0) -> Clipart(t=1)]

   python run_partial.py --s 0 --t 1 --dset office_home --net ResNet50 --cot_weight 1. --output run1 --gpu_id 0

2. Partial Domain Adaptation (PDA) on the Office dataset [Amazon(s=0) -> DSLR(t=1)]

   python run_partial.py --s 0 --t 1 --dset office --net ResNet50 --cot_weight 5. --output run1 --gpu_id 0
   python run_partial.py --s 0 --t 1 --dset office --net VGG16 --cot_weight 5. --output run1 --gpu_id 0

3. Partial Domain Adaptation (PDA) on the ImageNet-Caltech dataset [ImageNet(s=0) -> Caltech(t=1)]

   python run_partial.py --s 0 --t 1 --dset imagenet_caltech --net ResNet50 --cot_weight 5. --output run1 --gpu_id 0

Citation

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

@inproceedings{liang2020baus,
    title={A Balanced and Uncertainty-aware Approach for Partial Domain Adaptation},
    author={Liang, Jian, and Wang, Yunbo, and Hu, Dapeng, and He, Ran and Feng, Jiashi},
    booktitle={European Conference on Computer Vision (ECCV)},
    pages={xx-xx},
    month = {August},
    year={2020}
}

Acknowledgement

Some parts of this project are built based on the following open-source implementation

• CDAN https://github.com/thuml/CDAN/tree/master/pytorch
• COT https://github.com/henry8527/COT

Contact

• [e-mail]
• [e-mail]
• [e-mail]