Awesome

Class-Incremental Learning with CLIP: Adaptive Representation Adjustment and Parameter Fusion (ECCV24)

This is the official code for our paper: <a href='https://arxiv.org/pdf/2407.14143'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a>

Getting Started

Environment

create enviroment using Miniconda (or Anaconda)

conda create -n continual_clip python=3.8
conda activate continual_clip

install dependencies:

bash setup_environment.sh

Running scripts

We provide the scripts for imagenet100. Please run:

python main.py \
    --config-path configs/class \
    --config-name imagenet100_10-10.yaml \
    dataset_root="[imagenet1k_path]" \
    class_order="class_orders/imagenet100.yaml"

The dataset_root folder should contain the train and val folders.

imagenet1k_path
├── train
│   ├── n01440764 
│   └── ···
├── val
│   ├── n01440764 
│   └── ···

imagenet-r_path
├── train
│   ├── n01443537 
│   └── ···
├── val
│   ├── n01443537 
│   └── ···

The command to run the other two datasets is similar, in run_experiment.sh

datasets

Cifar100 will download automatically. Imagenet-R is randomly splited. You can also use our splited list in RAPF/imgr_split/imgr_train_test_split.txt.

The format of imgr_train_test_split.txt:

train
n02051845/art_0.jpg
...
test
n02051845/tattoo_4.jpg
...

Acknowledgement

Our method implementation is based on the Continual-CLIP.

Citation

If you find our repo useful for your research, please consider citing our paper:

@misc{huang2024rapf,
      title={Class-Incremental Learning with CLIP: Adaptive Representation Adjustment and Parameter Fusion}, 
      author={Linlan Huang and Xusheng Cao and Haori Lu and Xialei Liu},
      year={2024},
      eprint={2407.14143},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2407.14143}, 
}

License

This code is licensed under the Creative Commons Attribution-NonCommercial 4.0 International for non-commercial use only. Please note that any commercial use of this code requires formal permission prior to use.

Contact

For technical questions, please contact <a href="huanglinlan@mail.nankai.edu.cn">huanglinlan@mail.nankai.edu.cn</a>