PINA-DIL

This is the official implementation of our ECCV 2024 paper:
Non-Exemplar Domain Incremental Learning via Cross-Domain Concept Integration

Environment

conda create -n pina python=3.8
conda activate pina
pip install -r requirements.txt

Datasets

DomainNet

Please refer to DomainNet Project to download the dataset or run:

cd datasets
bash download_domainnet.sh

Then unzip the downloaded files, and confirm the file directory as shown below:

DomainNet
├── clipart
│   ├── aircraft_carrier
│   ├── airplane
│   ... ...
├── clipart_test.txt
├── clipart_train.txt
├── infograph
│   ├── aircraft_carrier
│   ├── airplane
│   ... ...
├── infograph_test.txt
├── infograph_train.txt
├── painting
│   ├── aircraft_carrier
│   ├── airplane
... ...

CDDB

Please refer to CDDB Project and download the dataset from CDDB Dataset.

Then unzip the downloaded files, and confirm the file directory as shown below:

CDDB
├── biggan
│   ├── train
│   └── val
├── gaugan
│   ├── train
│   └── val
├── san
│   ├── train
│   └── val
├── whichfaceisreal
│   ├── train
│   └── val
├── wild
│   ├── train
│   └── val
... ...

CORe50

Please refer to CORe50 Project and download the file shown below:

CORe50
├── core50_imgs.npz
├── labels.pkl
├── LUP.pkl
└── paths.pkl

Training and Inference

Please confirm the path of your datasets in the config files.

DomainNet

python main.py --config configs/domainnet_pina_vit.yaml --device 0
python main.py --config configs/domainnet_pina_clip.yaml --device 0

CDDB

python main.py --config configs/cddb_pina_vit.yaml --device 0
python main.py --config configs/cddb_pina_clip.yaml --device 0

CORe50

python main.py --config configs/core50_pina_vit.yaml --device 0
python main.py --config configs/core50_pina_clip.yaml --device 0

Acknowledgement

We thank PyCIL and S-Prompts for their wonderful framework and codes!
We also thank CLIP and CoOp for their helpful components.

Citation

If any part of our paper and code is helpful to your research, please give us a star and consider citing the following bib entry. Thanks!

@inproceedings{wang2024non,
  title={Non-exemplar domain incremental learning via cross-domain concept integration},
  author={Wang, Qiang and He, Yuhang and Dong, Songlin and Gao, Xinyuan and Wang, Shaokun and Gong, Yihong},
  booktitle={European Conference on Computer Vision},
  pages={144--162},
  year={2024},
  organization={Springer}
}