Research map / Vision-language and multimodal / CLIP and zero-shot
DeltaDeno: Zero-Shot Anomaly Generation via Delta-Denoising Attribution
Chaoran Xu, Chengkan Lv, Qiyu Chen, Yunkang Cao, Feng Zhang, Zhengtao Zhang
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 |
|---|---|---|---|
| add latent-mask inpainting in late denoising (ablation) | DeltaDeno · MVTec AD downstream detection | image AUROC: 79.7 -> 84.7 | Improved |
| add latent-mask inpainting in late denoising (ablation) | DeltaDeno · MVTec AD downstream detection | pixel AP: 25.8 -> 30.5 | Improved |
| add spatial attention biasing (ablation) | DeltaDeno · MVTec AD downstream detection | image AUROC: 79.5 -> 84.7 | Improved |
| add token-level prompt refinement (ablation) | DeltaDeno · MVTec AD downstream detection | image AUROC: 78.8 -> 84.7 | Improved |
| add SAM foreground prior (ablation) | DeltaDeno · MVTec AD downstream detection | pixel AUROC: 82.1 -> 84.9 | Improved |
| switch backbone from SD1.5 to FLUX | DeltaDeno · MVTec AD downstream detection | image AUROC: 84.7 -> 89.7 | Improved |
| switch backbone from SD1.5 to FLUX | DeltaDeno · VisA downstream detection | pixel AUROC: 90.1 -> 90.2 | No clear effect |
| raise t_mid threshold from 0.6 to 0.8 (ablation) | DeltaDeno · MVTec AD downstream detection | pixel AUROC: 84.9 -> 71.4 | Hurt |
| raise last-step threshold from 0.35 to 0.60 (ablation) | DeltaDeno · MVTec AD downstream detection | pixel AUROC: 84.9 -> 69.1 | Hurt |
| lower t_mid threshold from 0.6 to 0.4 (ablation) | DeltaDeno · MVTec AD downstream detection | image AUROC: 84.7 -> 86.3; AP_p 30.5 -> 35.3 | Mixed |
| zero-shot DeltaDeno vs few-shot AnoGen | DeltaDeno (FLUX) · MVTec AD downstream detection | pixel AUROC: 76.7 -> 85.4 | Improved |
| zero-shot DeltaDeno vs few-shot AnomalyDiffusion | DeltaDeno (FLUX) · MVTec AD downstream detection | pixel AP: 58.4 -> 41.5 | Hurt |
Closest papers by meaning
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.87)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.87)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.86)
- DZAD: Diffusion-based Zero-shot Anomaly Detection (AAAI 2025 · similarity 0.86)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.86)
- Beyond Normal References: Discriminative Few-Shot Anomaly Detection (NeurIPS 2026 · similarity 0.86)
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.
