DGCL

Domain Generalizable Continual Learning

Requirements

1. torch 2.0.1
2. torchvision 0.15.2
3. timm 0.6.12
4. tqdm
5. numpy
6. scipy
7. easydict
8. matplotlib

Usage

Data Preparation

Download the dataset and then update utils/data.py with the path to your data directory.

Baseline Experiments

1. For implemented methods (listed in models folder), follow the bash scripts named run_*.sh to run the experiments.

2. For detailed configurations, please refer to the configs folder, baseline configs are located in configs/DGIL/[dataset_name], where [method_name].json runs the regular class incremental learning pipeline, and [method_name]_dgil.json runs the domain generalizable continual learning pipeline.

DoT Experiments

Here specifically, we provide the scripts to run the DoT experiments on DigitsDG and OfficeHome datasets with SLCA and L2P:

1. DoT-SLCA:

# DoT-SLCA on DigitsDG
python main.py --config ./configs/DGIL/digitsdg/dot_slca_dgil.json

# DoT-SLCA on OfficeHome
python main.py --config ./configs/DGIL/officehome/dot_slca_dgil.json

2. DoT-L2P:

# DoT-L2P on DigitsDG
python main.py --config ./configs/DGIL/digitsdg/dot_l2p_dgil.json

# DoT-L2P on OfficeHome
python main.py --config ./configs/DGIL/officehome/dot_l2p_dgil.json