[ICPR 2024] Clustering-based Image-Text Graph Matching for Domain Generalization

Set Environment

conda create -n GDG python=3.8

conda activate GDG

pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113

# install dgl package
pip install dgl-1.0.1+cu113-cp38-cp38-manylinux1_x86_64.whl
pip install dglgo==0.0.2

pip install pycocotools tensorboard matplotlib wilds

CUB-DG

Dataset

You can download the CUB-DG dataset at here!

How to Train?

./train.sh [LOG_DIR_PATH] [arg_OPTIONS]

• If you want to train other algorithms, you should edit the train.sh file.

How to Evaluate?

./eval.sh [LOG_DIR_PATH] [env0_CHECKPOINT] [env1_CHECKPOINT] [env2_CHECKPOINT] [env3_CHECKPOINT]

• If you want to evaluate other algorithms, you should edit the eval.sh file.

Domainbed

We modified the DomainBed repo to utilize natural language.
You have to download Domainbed dataset first. For more information, please see the original repo.

To train the model on the domainbed, see DomainBed/PACS.sh. DomainBed/PACS.sh is a shell file to train the model on the PACS dataset.