/TransCal

Primary LanguagePython

Code release for "Transferable Calibration with Lower Bias and Variance in Domain Adaptation"

Dataset

Office-31

Office-31 dataset can be found here.

Office-Home

Office-Home dataset can be found here.

DomainNet

DomainNet dataset can be found here.

VisDA-2017

VisDA-2017 dataset can be found here

ImageNet-Sketch

ImageNet-Sketch dataset can be found here

Calibration in DA:

Step 1: Train a domain adaptation model using the selected DA method

python 1_train_DA_models.py

Step 2: Fix the DA model and compute features for source train, source validation, and target, respectively

python 2_generate_features.py

Step 3: Transferable Calibration

python 3_TransCal.py

For a quick start, we provide the pre-trained features on Office-Home via CDAN+E here. You can directly skip the first two steps and run the third step to evaluate the performance of TransCal while calibrating this DA model on Office-Home.

Citation

If you find this code useful for your research, please consider citing:

@inproceedings{Wang20TransCal,
    title = {Transferable Calibration with Lower Bias and Variance in Domain Adaptation},
    author = {Wang, Ximei and Long, Mingsheng and Wang, Jianmin and Jordan, Michael I},
    booktitle = {Advances in Neural Information Processing Systems 33},
    year = {2020}
}

Contact

If you have any problem with our code, feel free to contact