Code (pytorch) for 'Progressive Source-Aware Transformer for Generalized Source-Free Domain Adaptation' on Digit, Office-31, Office-Home, VisDA-C. This paper has been accepted by IEEE Transactions on Multimedia (TMM). DOI: https://doi.org/10.1109/TMM.2023.3321421
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
- numpy, scipy, sklearn, PIL, argparse, randaugment, torchsample
We choose R50-ViT-B_16 as our encoder.
wget https://storage.googleapis.com/vit_models/imagenet21k/R50+ViT-B_16.npz
mkdir ./model/vit_checkpoint/imagenet21k
mv R50+ViT-B_16.npz ./model/vit_checkpoint/imagenet21k/R50+ViT-B_16.npz-
First training model on Office-31 dataset of single target is shown here.
~/anaconda3/envs/vit_kd/bin/python target_PSAT-GDA.py --cls_par 0.6 --da uda --dset VISDA-C --gpu_id 0 --s 0 --t 1 --output_src ckpskdr2/source_tt_08/ --output ckpskdr2/target_PSAT-GDA_08_0112/ --lr 1e-3
-
Second training model on Office-31 dataset of multiple targets is shown here.
~/anaconda3/envs/vit_kd/bin/python target_PSAT-GDA_cda_oc.py --cls_par 0.6 --da uda --dset office --gpu_id 0 --s 0 --t 1 --output ckpspdtcda/target_PSAT-GDAcda_0113/ --seed 2020;
The results of PKMSM is display under the folder './PSAT-GDA/results/'.
@ARTICLE{tang2023psat, author={Tang, Song and Shi, Yuji and Song, Zihao and Ye, Mao and Zhang, Changshui and Zhang, Jianwei}, journal={IEEE Transactions on Multimedia}, title={Progressive Source-Aware Transformer for Generalized Source-Free Domain Adaptation}, year={2023}, volume={}, number={}, pages={1-14}, doi={10.1109/TMM.2023.3321421}}
The codes are based on SHOT (ICML 2020, also source-free), VIT