Our work focuses on addressing sample deficiency from low-density regions of data manifold in common image datasets. We leverage diffusion process based generative models to synthesize novel images from low-density regions. We observe that uniform sampling from diffusion models predominantly samples from high-density regions of the data manifold. Therefore, we modify the sampling process to guide it towards low-density regions while simultaneously maintaining the fidelity of synthetic data. We rigorously demonstrate that our process successfully generates novel high fidelity samples from low-density regions. We further examine generated samples and show that the model does not memorize low-density data and indeed learns to generate novel samples from low-density regions.
@article{arxiv.2203.17260,
title = {Generating High Fidelity Data from Low-density Regions using Diffusion Models},
author = {Vikash Sehwag and Caner Hazirbas and Albert Gordo and Firat Ozgenel and Cristian Canton Ferrer},
journal= {arXiv preprint arXiv:2203.17260},
year = {2022}
}
Comments
CVPR 2022 (fixed some discrepancies in notation - v2)