Curvature-aware correction improves stability in diffusion image generation
Geometry-Aware Diffusion Guidance via Curvature-Adaptive Tubular Correction
Computer Vision and Pattern Recognition
Summary
When computer programs create images by gradually refining noise, they follow clues called gradients that guide the process. However, if these clues are too strong, the program can get confused and produce worse images. The authors came up with a way to carefully adjust these clues by considering the shape of the data’s surface, making the guidance more stable and accurate. Their new method, called curvature-adaptive tubular correction, improved image quality in several test problems, including making sharper faces and clearer natural images.
What this means in practice
- •For image synthesis developers: Enhance the visual quality and stability of AI-generated images by integrating curvature-adaptive guidance correction into diffusion sampling workflows.
- •For medical imaging engineers: Improve reconstruction accuracy in inverse imaging problems, such as black hole imagery, by applying curvature-aware corrections during diffusion-based sampling.
Authors
Enze Jiang, Jinwei He, Zheng Ma
Abstract
Gradient-guided diffusion samplers provide flexible priors for inverse problems and conditional generation, but strong guidance can move the sampling trajectory into regions where the learned score is poorly supported. Existing tangent-projection strategies limit first-order departure from an iso-density surface, yet discard potentially useful normal motion and overlook the second-order departure induced by tangent motion on a curved surface. We introduce curvature-adaptive tubular correction (CAT), a training-free plugin that regulates both effects within a shared, noise-dependent geometric budget. CAT decomposes the guidance gradient into normal and tangent components, charges normal displacement at first order and tangent displacement according to directional curvature, and obtains their jointly optimal magnitudes from a one-dimensional dual equation. Armijo backtracking calibrates the resulting finite step against the actual guidance objective, while matrix-free directional derivatives avoid constructing the full score Jacobian. We establish local guarantees for the tubular approximation, uniqueness of the correction, and sufficient objective decrease. Across seven inverse problems on FFHQ and ImageNet, CAT improves the evaluated pixel- and latent-space host samplers, with particularly consistent gains in perceptual metrics. It also improves black hole reconstruction on InverseBench and yields the lowest FID among the compared methods at every tested classifier-free guidance scale, while maintaining stable saturation and contrast. These results support curvature-aware tubular control as a reusable mechanism for stabilizing diffusion guidance.