Backbone-Equated Diffusion OOD via Sparse Internal Snapshots
For researchers in OOD detection, this work provides a fair evaluation protocol and a computationally efficient detector, though the results are limited to CIFAR-scale datasets.
The paper introduces a Mutualized Backbone-Equated (MBE) protocol for fair comparison of diffusion-based OOD detectors and proposes Canonical Feature Snapshots (CFS), a detector that uses only a few internal activations from a frozen diffusion backbone. On CIFAR-scale benchmarks, the best variant CFS(1x2) achieves strong OOD detection performance, showing that OOD signal is concentrated in sparse internal states.
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/.