Research map / Detection and segmentation / Open-set and few-shot detection
TransFusion -- A Transparency-Based Diffusion Model for Anomaly Detection
Matic Fučka, Vitjan Zavrtanik, Danijel Skočaj
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 normal appearance loss L_n (ablation) | TransFusion · VisA | image AUROC: -31.8 when removed | Improved |
| feed simulated previous mask estimate as input (ablation) | TransFusion · VisA localization | AUPRO: -8.6 when removed | Improved |
| add 2D positional encoding input (ablation) | TransFusion · VisA | image AUROC: -1.4 when removed | Improved |
| add mask, anomaly appearance and consistency losses (ablation) | TransFusion · VisA and MVTec AD | image AUROC: about -1 each when removed | Improved |
| combine discriminative mask with reconstruction error mask for final mask (ablation) | TransFusion · VisA localization | AUPRO: -13.5 when only M_recon used | Improved |
| average masks over all diffusion steps instead of last mask (ablation) | TransFusion · VisA | image AUROC: -1.5 when only last mask used | Improved |
| reduce diffusion steps from 20 to 5 (ablation) | TransFusion · VisA | image AUROC: -1.0 | No clear effect |
| use linear instead of quadratic transparency schedule (ablation) | TransFusion · VisA | image AUROC: -2.0 AUROC but +1.2 AUPRO for quadratic | Mixed |
| replace two-stage reconstruct-then-discriminate with single-stage transparency diffusion | TransFusion vs DiffAD · VisA | image AUROC: 89.5 -> 98.5 | Improved |
| iterative diffusion inference instead of single-pass models | TransFusion · A100 inference time | seconds per sample: 0.05 (DRAEM) -> 0.34 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DRAEM — DRAEM - A Discriminatively Trained Reconstruction Embedding for Surface Anomaly Detection (ICCV 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.89)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.89)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.88)
- Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · similarity 0.88)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · similarity 0.86)
- A Unified Latent Schrodinger Bridge Diffusion Model for Unsupervised Anomaly Detection and Localization (CVPR 2025 · 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.
