KLIP: localized distribution shift detection via KL-divergence with diffusion priors in Inverse Problems

2026-05-29

AI Summary

This work addresses the challenge of detecting subtle or localized out-of-distribution (OOD) shifts in inverse problems, where existing methods often rely on shift priors or require full-image inputs. The authors propose a novel OOD detection metric based on the Kullback–Leibler (KL) divergence between the prior and Bayesian posterior distributions of diffusion models. This approach enables both global and local OOD detection without requiring calibration data or prior knowledge of distributional shifts. Notably, it is the first method to achieve prior-free, localized OOD detection in inverse problems, effectively capturing semantically meaningful fine-grained anomalies. The framework demonstrates strong generalization across diverse diffusion models, datasets, and inverse problem settings—such as detecting tumors in liver CT scans—highlighting its robustness and broad applicability.

Abstract

Diffusion models have shown promising performance as data-driven priors for computational imaging, as well as some capacity to detect out-of-distribution (OOD) images. However, existing approaches to OOD detection often require some knowledge of the shifted distribution, fail to detect subtle or localized distribution shifts, and operate on full images, rather than the indirect measurements available in inverse problems. We propose an OOD detection metric based on the Kullback-Leibler divergence between the diffusion prior and the posterior distribution, that (i) does not require any calibration data or knowledge of the shifted distribution, and (ii) can detect whole images as OOD as well as localize OOD patches within an image. Experimentally, we show that this metric can detect subtle yet semantically meaningful distribution shifts, such as the shift from healthy liver CT scans to those with tumors, and generalizes across different types of diffusion models, datasets, and inverse problems. Our code can be found at https://github.com/voilalab/KLIP.

Problem

• out-of-distribution detection
• localized distribution shift
• inverse problems
• diffusion models
• KL-divergence

Innovation

• KL-divergence
• diffusion priors
• out-of-distribution detection
• localized distribution shift
• inverse problems