Code and data for our paper CLoG: Benchmarking Continual Learning of Image Generation Models
- [Jun. 7, 2024]: We launch the first version of code for label-conditioned CLoG. Our codebase is still in development, please stay tuned for the comprehensive version.
We advocates for shifting the research focus from classification-based continual learning (CL) to continual learning of generative models (CLoG). Our codebase adapts 12 existing CL methodologies of three types—replay-based, regularization-based, and parameter-isolation-based methods—to generative tasks and introduce 8 benchmarks for CLoG that feature great diversity and broad task coverage.
To run CLoG from source, follow these steps:
- Clone this repository locally
cdinto the repository.- Run
conda env create -f environment.ymlto created a conda environment namedCLoG. - Activate the environment with
conda activate CLoG.
Coming soon! For the time being, you can check scripts/cifar-naive.sh for running NCL on CIFAR-10.
We would love to hear from the CL community, broader machine learning, and generative AI communities, and we welcome any contributions, pull requests, or issues! To do so, please either file a new pull request or issue. We'll be sure to follow up shortly!
If you find our work helpful, please use the following citations.
@article{
zhang2024clog,
title={CLoG: Benchmarking Continual Learning of Image Generation Models},
author={Haotian Zhang and Junting Zhou and Haowei Lin and Hang Ye and Jianhua Zhu and Zihao Wang and Liangcai Gao and Yizhou Wang and Yitao Liang},
booktitle={arxiv},
year={2024}
}
MIT. Check LICENSE.md.