Image Interpolation with Score-Based Riemannian Metrics of Diffusion Models
Abstract
Diffusion models excel in content generation by implicitly learning the data manifold, yet they lack a practical method to leverage this manifold---unlike other deep generative models equipped with latent spaces. This paper introduces a novel framework that treats the data space of pre-trained diffusion models as a Riemannian manifold, with a metric derived from score function. Experiments with MNIST and Stable Diffusion show that this geometry-aware approach yields smoother interpolations than linear or spherical linear interpolation and other methods, demonstrating its potential for improved content generation and editing.
Cite
Text
Saito and Matsubara. "Image Interpolation with Score-Based Riemannian Metrics of Diffusion Models." ICLR 2025 Workshops: DeLTa, 2025.Markdown
[Saito and Matsubara. "Image Interpolation with Score-Based Riemannian Metrics of Diffusion Models." ICLR 2025 Workshops: DeLTa, 2025.](https://mlanthology.org/iclrw/2025/saito2025iclrw-image/)BibTeX
@inproceedings{saito2025iclrw-image,
title = {{Image Interpolation with Score-Based Riemannian Metrics of Diffusion Models}},
author = {Saito, Shinnosuke and Matsubara, Takashi},
booktitle = {ICLR 2025 Workshops: DeLTa},
year = {2025},
url = {https://mlanthology.org/iclrw/2025/saito2025iclrw-image/}
}