G2G

This is the Official PyTorch implemention of our ICASSP2024 paper "G2G: Generalized Learning by Cross-Domain Knowledge Transfer for Federated Domain Generalization".

G2G:Generalized Learning by Cross-Domain Knowledge Transfer for Federated Domain Generalization

Code to reproduce the experiments of G2G:Generalized Learning by Cross-Domain Knowledge Transfer for Federated Domain Generalization.

How to use it

• Clone or download the repository

Install the requirement

pip install -r requirements.txt

Download VLCS, PACS and Office-Home datasets

• Download VLCS from https://github.com/jindongwang/transferlearning/blob/master/data/dataset.md, extract, move it to datasets/VLCS/
• Download PACS from https://drive.google.com/uc?id=0B6x7gtvErXgfbF9CSk53UkRxVzg, move it to datasets/PACS/
• Download Office-Home from https://datasets.activeloop.ai/docs/ml/datasets/office-home-dataset/, move it to datasets/OfficeHome/

Download Pre-trained models

• Download pre-trained AlexNet from https://download.pytorch.org/models/alexnet-owt-4df8aa71.pth and move it to models/
• Download pre-trained Resnet50 from https://download.pytorch.org/models/resnet50-19c8e357.pth and move it to models/
• Download pre-trained VGG16 https://download.pytorch.org/models/vgg16-397923af.pth and move it to models/

Running ours on VLCS

python main_warm.py --node_num 3  --device cuda:0 --dataset vlcs --classes 5 --lr 0.0008 --global_model Alexnet --local_model Alexnet --algorithm fed_mutual --R 50 --E 7 --pretrained True --batch_size 64 --iteration 0 

Running ours on PACS

python main_warm.py --node_num 3  --device cuda:0 --dataset pacs --classes 7 --lr 0.0008 --global_model Alexnet --local_model Alexnet --algorithm fed_mutual --R 50 --E 7 --pretrained True --batch_size 32 --iteration 0 

Running ours on Office-Home

python main_warm.py --node_num 3 --device cuda:0 --dataset office-home --classes 65 --lr 0.0008 --global_model ResNet50 --local_model ResNet50 --algorithm fed_mutual --R 50 --E 7 --batch_size 32 --iteration 0