Research map / Image generation / Autoregressive and latent generation
DiffusionAD: Norm-Guided One-Step Denoising Diffusion for Anomaly Detection
Hui Zhang, Zheng Wang, Dan Zeng, Zuxuan Wu, Yu–Gang Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace auto-encoder reconstruction with diffusion noise-to-norm reconstruction (ablation) | DiffusionAD segmentation pipeline · VisA | Image AUROC: 90.6 -> 96.8 | Improved |
| replace auto-encoder reconstruction with diffusion noise-to-norm reconstruction (ablation) | DiffusionAD segmentation pipeline · VisA | Pixel PRO: 76.1 -> 92.5 | Improved |
| use one-step denoising instead of 400-step iterative denoising (ablation) | DiffusionAD · VisA | Image AUROC: 96.8 -> 98.0 | Improved |
| use one-step denoising instead of 400-step iterative denoising (ablation) | DiffusionAD · VisA | FPS: 0.09 FPS -> 26.7 FPS | Improved |
| use one-step denoising instead of DDIM-30 denoising (ablation) | DiffusionAD · VisA | Image AUROC: 96.6 -> 98.0 | Improved |
| add norm-guided denoising on top of one-step denoising (ablation) | DiffusionAD · VisA | Image AUROC: 98.0 -> 98.8 | Improved |
| add norm-guided denoising on top of one-step denoising (ablation) | DiffusionAD · VisA | Pixel PRO: 94.7 -> 96.0 | Improved |
| add norm-guided denoising on top of one-step denoising (ablation) | DiffusionAD · VisA | Pixel AP: 44.2 -> 45.8 | Improved |
| add norm-guided denoising on top of one-step denoising (ablation) | DiffusionAD · VisA | FPS: 26.7 -> 23.5 | Hurt |
| use iterative-400 inference after one-step training (train-inference mismatch) (ablation) | DiffusionAD · VisA | Image AUROC: 98.0 -> 96.8 | Hurt |
| train segmentation on iterative-400 denoised outputs and infer iteratively (ablation) | DiffusionAD · VisA | Image AUROC: 98.0 -> 98.3 | Improved |
| raise noise scale t to 500 or more for one-step denoising (ablation) | DiffusionAD · VisA | Image AUROC: decrease at t >= 500 | Hurt |
Closest papers by meaning
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.92)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.91)
- DZAD: Diffusion-based Zero-shot Anomaly Detection (AAAI 2025 · similarity 0.91)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.90)
- Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · similarity 0.90)
- FAST: Foreground‑aware Diffusion with Accelerated Sampling Trajectory for Segmentation‑oriented Anomaly Synthesis (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
