Code release for "Transferable Calibration with Lower Bias and Variance in Domain Adaptation"
Office-31 dataset can be found here.
Office-Home dataset can be found here.
DomainNet dataset can be found here.
VisDA-2017 dataset can be found here
ImageNet-Sketch dataset can be found here
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
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.
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}
}
If you have any problem with our code, feel free to contact