DeltaDeno: Zero-Shot Anomaly Generation via Delta-Denoising Attribution

TL;DR

DeltaDeno introduces a zero-shot anomaly generation method using diffusion models that localizes and edits defects without real anomaly samples, improving detection and realism.

Contribution

The paper presents DeltaDeno, a training-free, zero-shot anomaly generation technique leveraging diffusion models and prompt refinement for improved defect localization and generation.

Findings

Achieves realistic and localized anomaly generation.

Enhances downstream anomaly detection performance.

Operates without training on anomalous data.

Abstract

Anomaly generation is often framed as few-shot fine-tuning with anomalous samples, which contradicts the scarcity that motivates generation and tends to overfit category priors. We tackle the setting where no real anomaly samples or training are available. We propose Delta-Denoising (DeltaDeno), a training-free zero-shot anomaly generation method that localizes and edits defects by contrasting two diffusion branches driven by a minimal prompt pair under a shared schedule. By accumulating per-step denoising deltas into an image-specific localization map, we obtain a mask to guide the latent inpainting during later diffusion steps and preserve the surrounding context while generating realistic local defects. To improve stability and control, DeltaDeno performs token-level prompt refinement that aligns shared content and strengthens anomaly tokens, and applies a spatial attention bias restricted to anomaly tokens in the predicted region. Experiments on public datasets show that DeltaDeno achieves great generation, realism and consistent gains in downstream detection performance. Code will be made publicly available.