| # Acknowledgments |
|
|
| ## Original Work |
|
|
| This package is based on the **Erased Stable Diffusion (ESD)** framework: |
|
|
| - **Original Repository**: https://github.com/rohitgandikota/erasing-concepts-diffusion |
| - **Authors**: Rohit Gandikota, Joanna Materzynska, Jaden Fiotto-Kaufman, David Bau |
| - **License**: MIT License (see LICENSE file) |
| - **Paper**: Erasing Concepts from Diffusion Models (ICCV 2023) |
|
|
| ### Citation |
|
|
| If you use this code in your research, please cite the original paper: |
|
|
| ```bibtex |
| @inproceedings{gandikota2023erasing, |
| title={Erasing Concepts from Diffusion Models}, |
| author={Rohit Gandikota and Joanna Materzy\'nska and Jaden Fiotto-Kaufman and David Bau}, |
| booktitle={Proceedings of the 2023 IEEE International Conference on Computer Vision}, |
| year={2023} |
| } |
| ``` |
|
|
| ## Modifications |
|
|
| This package adaptation includes: |
|
|
| 1. **Packaged for pip installation**: Adapted for standard Python packaging (pyproject.toml, setuptools) |
| 2. **Simplified API**: Wrapped the training logic in an `ESDPipeline` class for easier programmatic use |
| 3. **Framework integration**: Added compatibility with the eval-learn benchmarking framework |
| 4. **Flexible training methods**: Support for multiple fine-tuning approaches (xattn, full, selfattn, noxattn) |
|
|
| All modifications maintain the original algorithm and research contributions unchanged. |
|
|
| ## License |
|
|
| This package retains the original MIT License. See LICENSE file for full terms. |
|
|
