SHOT: A Python repository from abutaufique

Code for our ICML-2020 paper Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation.

Framework:

Prerequisites:

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

Dataset:

Please manually download the datasets Office, Office-Home, VisDA-C, Office-Caltech from the official websites, and modify the path of images in each '.txt' under the folder './object/data/'.
Concerning the Digits dsatasets, the code will automatically download three digit datasets (i.e., MNIST, USPS, and SVHN) in './digit/data/'.

Results:

Note that we update the code and further consider the standard learning rate scheduler like DANN and report new results in the final camera ready version. Please refer results.md for the detailed results on various datasets.

Training:

Unsupervised Closed-set Domain Adaptation (UDA) on the Digits dataset

MNIST -> USPS (m2u) SHOT (cls_par = 0.1) and SHOT-IM (cls_par = 0.0)

 cd digit/
 python uda_digit.py --dset m2u --gpu_id 0 --output ckps_digits --cls_par 0.0
 python uda_digit.py --dset m2u --gpu_id 0 --output ckps_digits --cls_par 0.1

Unsupervised Closed-set Domain Adaptation (UDA) on the Office/ Office-Home dataset

Train model on the source domain A (s = 0)

cd object/
python image_source.py --trte val --da uda --output ckps/source/ --gpu_id 0 --dset office --max_epoch 100 --s 0

Adaptation to other target domains D and W, respectively

python image_target.py --cls_par 0.3 --da uda --output_src ckps/source/ --output ckps/target/ --gpu_id 0 --dset office --s 0

Unsupervised Closed-set Domain Adaptation (UDA) on the VisDA-C dataset

Synthetic-to-real

cd object/
 python image_source.py --trte val --output ckps/source/ --da uda --gpu_id 0 --dset VISDA-C --net resnet101 --lr 1e-3 --max_epoch 10 --s 0
 python image_target.py --cls_par 0.3 --da uda --dset VISDA-C --gpu_id 0 --s 0 --output_src ckps/source/ --output ckps/target/ --net resnet101 --lr 1e-3

Unsupervised Partial-set Domain Adaptation (PDA) on the Office-Home dataset

Train model on the source domain A (s = 0)

 cd object/
 python image_source.py --trte val --da pda --output ckps/source/ --gpu_id 0 --dset office-home --max_epoch 50 --s 0

Adaptation to other target domains C and P and R, respectively

 python image_target.py --cls_par 0.3 --threshold 10 --da pda --dset office-home --gpu_id 0 --s 0 --output_src ckps/source/ --output ckps/target/

Unsupervised Open-set Domain Adaptation (ODA) on the Office-Home dataset

Train model on the source domain A (s = 0)

 cd object/
 python image_source.py --trte val --da oda --output ckps/source/ --gpu_id 0 --dset office-home --max_epoch 50 --s 0

Adaptation to other target domains C and P and R, respectively

 python image_target_oda.py --cls_par 0.3 --da oda --dset office-home --gpu_id 0 --s 0 --output_src ckps/source/ --output ckps/target/

Unsupervised Multi-source Domain Adaptation (MSDA) on the Office-Caltech dataset

Train model on the source domains A (s = 0), C (s = 1), D (s = 2), respectively

 cd object/
 python image_source.py --trte val --da uda --output ckps/source/ --gpu_id 0 --dset office-caltech --max_epoch 100 --s 0
 python image_source.py --trte val --da uda --output ckps/source/ --gpu_id 0 --dset office-caltech --max_epoch 100 --s 1
 python image_source.py --trte val --da uda --output ckps/source/ --gpu_id 0 --dset office-caltech --max_epoch 100 --s 2

Adaptation to the target domain W (t = 3)

 python image_target.py --cls_par 0.3 --da uda --output_src ckps/source/ --output ckps/target/ --gpu_id 0 --dset office --s 0
 python image_target.py --cls_par 0.3 --da uda --output_src ckps/source/ --output ckps/target/ --gpu_id 0 --dset office --s 1
 python image_target.py --cls_par 0.3 --da uda --output_src ckps/source/ --output ckps/target/ --gpu_id 0 --dset office --s 0
 python image_multisource.py --cls_par 0.0 --da uda --dset office-caltech --gpu_id 0 --t 3 --output_src ckps/source/ --output ckps/target/

Unsupervised Multi-target Domain Adaptation (MTDA) on the Office-Caltech dataset

Train model on the source domain A (s = 0)

 cd object/
 python image_source.py --trte val --da uda --output ckps/source/ --gpu_id 0 --dset office-caltech --max_epoch 100 --s 0

Adaptation to multiple target domains C and P and R at the same time

 python image_multitarget.py --cls_par 0.3 --da uda --dset office-caltech --gpu_id 0 --s 0 --output_src ckps/source/ --output ckps/target/

Unsupervised Partial Domain Adaptation (PDA) on the ImageNet-Caltech dataset without source training by ourselves (using the downloaded Pytorch ResNet50 model directly)
- ImageNet -> Caltech (84 classes) [following the protocol in PADA]
```
 cd object/
 python image_pretrained.py --gpu_id 0 --output ckps/target/ --cls_par 0.3
```

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

Citation

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

@inproceedings{liang2020shot,
title={Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation},
author={Liang, Jian and Hu, Dapeng and Feng, Jiashi},
booktitle={International Conference on Machine Learning (ICML)},
pages={xx-xx},
month = {July},
year={2020}
}

abutaufique/SHOT