README.md

LoRACL - Low-Rank Adversarial Contrastive Learning for Unlabeled Domain Generalization

how to start

1. installing environment from environment.yml
2. prepare datasets
3. bash ./script/simclr_vit_art.sh

datasets preparation

file structure

base_folder/  
|–– LoRACL/  
|–– datasets/  
|   |–– /PACS
|   |   |–– /art_painting  
|   |   |–– /cartoon
|   |   |–– /photo
|   |   |–– /sketch  
|   |–– /DomainNet
|   |   |–– /clipart
|   |   |–– /clipart_test
|   |   |–– /infograph  
|   |   |–– /infograph_test
|   |   |–– ...

link to datasets

PACS DomainNet

Things to notice

DO rememeber preprocess DomainNet before using!

• train-test split
• remove unused categories according to Table 18 in DARLING