A Reproducible Extraction of Training Images from Diffusion Models
Paper • 2305.08694 • Published
Datasets:
Dataset Viewer
The dataset viewer is not available for this dataset.
Unexpected token '<', "<html>
<h"... is not valid JSON
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Webster Memorized GT Images
Ground truth training images for memorization evaluation in diffusion models. Extracted from Stable Diffusion v1.4 via the one-step extraction attack by Webster (2023).
Source
- Original repo: ryanwebster90/onestep-extraction
- Paper: A Reproducible Extraction of Training Images from Diffusion Models (Webster, 2023)
- Original HF: fraisdufour/templates-verbs
Contents
- 351 memorized GT images (
00000.jpg~00499.jpg, 149 missing due to dead LAION URLs) prompt_to_ref.csv: prompt_idx → ref_file mapping (500 rows, MISSING for unavailable)
License
Images originate from the LAION-2B dataset. Used here for research/evaluation purposes only. Original copyright holders retain all rights.
Citation
@article{webster2023reproducible,
title={A Reproducible Extraction of Training Images from Diffusion Models},
author={Webster, Ryan},
journal={arXiv preprint arXiv:2305.08694},
year={2023}
}
- Downloads last month
- 314