> All algorithm implementation can be found in algorithms.py, and the Mean_Augmentation_AugCA is the final implementation of our method.

>script example:

python3 -m domainbed.scripts.train_fed       --data_dir=./domainbed/data/CIFAR10/       --algorithm ERM       --dataset RotatedCIFAR10       --train_envs 10 --steps 50000 --output_dir output-cifar10/ERM --local_steps 50 --device 6 --save_model_every_checkpoint --seed 12345

>Before running VHL, you need to first run generative/generate.sh to generate virtual data. You may change the size of virtual data and the dataset name for different datasets.

This code is submitted for a sanity check, and we will re-arrange the official code soon.

