Skip to content

ML-Security-Research-LAB/SSDA

Repository files navigation

SSDA-Secure-Source-Free-Domain-Adaptation

Official code repository of SSDA

  1. Download dataset following the instructions from here

  2. Train source model using the following command:

    python train_source.py --dset=office-home --max_epoch=100 --attack_type=badnet --device=cuda:0 --s=0
    
  3. Train target model without defense and save the pseudo labels for knowledge transfer using the following command

    python train_target.py --dset=office-home --max_epoch=15 --attack_type=badnet --save_knowledge --device=cuda:0 --s=0
    
  4. Finally, train target model using SSDA

    python train_target_defend_main.py --dset=office-home --max_epoch=15 --attack_type=badnet --clip --clip_val=1.0 --defend --device=cuda:0 --s=0
    

About

Official code repository of SSDA (ICCV 2023)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages