Backbone-Equated Diffusion OOD via Sparse Internal Snapshots

2026-05-10

AI Summary

Existing diffusion models for out-of-distribution (OOD) detection lack a fair comparison benchmark due to disparities in backbone architectures, corruption modeling, and test-time overhead. This work proposes a Mutualized Backbone-Equated evaluation protocol and introduces Canonical Feature Snapshots (CFS), a method that efficiently detects OOD samples using only sparse internal activations from a frozen diffusion backbone at low noise levels—eliminating the need for full denoising or complex downstream heads. Theoretical analysis reveals that OOD signals concentrate in complementary encoder–decoder local states that remain stable under low noise. Experiments demonstrate that the CFS(1×2) single-forward variant achieves state-of-the-art performance on CIFAR benchmarks, while an extremely lightweight decoder variant remains competitive, confirming the method’s efficiency and effectiveness.

Abstract

Fair comparison between diffusion-based OOD detectors is challenging, as conclusions can vary with backbone choice, corruption parameterization, and test-time budget. We address this issue through a Mutualized Backbone-Equated (MBE) protocol that aligns canonical corruption levels and logical test-time cost across diffusion backbones. Within this setting, we introduce Canonical Feature Snapshots (CFS), a family of detectors that probes a frozen diffusion backbone using only a tiny number of native internal activations at canonical low-noise levels. On a controlled CIFAR-scale benchmark, the strongest one-forward CFS variant is CFS(1x2), while an even smaller decoder-only variant remains highly competitive. This shows that much of the relative-OOD signal exposed by frozen diffusion backbones is concentrated in a small number of sparse internal states, rather than requiring full denoising trajectories or high-capacity downstream heads. We further provide a local diagnostic theory explaining these observations through conditional encoder-decoder complementarity, diagonal-score separation, and low-noise corruption stability. The official implementation is available at https://github.com/RouzAY/cfs-diffusion-ood/.

Problem

• diffusion-based OOD detection
• fair comparison
• backbone selection
• test-time budget
• corruption parameterization

Innovation

• diffusion-based OOD detection
• Canonical Feature Snapshots
• Mutualized Backbone-Equated protocol
• sparse internal activations
• frozen backbone