hynnsk / PGE_DG

Pivot-Guided Embedding for Domain Generalization

This is the official implementation of "Pivot-Guided Embedding for Domain Generalization" in Pytorch.

The code heavily relies on 'In Search of Lost Domain Generalization'. ( Paper | Github )

1. Requirements

• numpy==1.20.3
• wilds==1.2.2
• imageio==2.9.0
• gdown==3.13.0
• torchvision==0.8.2
• torch==1.7.1
• tqdm==4.62.2
• backpack==0.1
• parameterized==0.8.1
• Pillow==8.3.2

2. Training & Evaluation

The data should be prepared on /DATA

To train the model, run the code as below:

python -m domainbed.scripts.sweep launch\
       --data_dir=/DATA\
       --expname pge01 \
       --output_dir experiments\
       --command_launcher local\
       --algorithms PGE\
       --dataset PACS\
       --single_test_envs\
       --n_hparams 20\
       --n_trials 3

For simplicity, we provide the training scripts. You can execute the shell file by the command below:

sh run.sh

To view the results of your sweep:

python -m domainbed.scripts.collect_results\
       --input_dir=./[output_dir]/[expname]

Acknowledgement

Again, this repository is built based on DomainBed repository. Thanks for the great work.