Research map / Label-efficient and robust learning / OOD and anomaly detection
AnoStyler: Text-Driven Localized Anomaly Generation via Lightweight Style Transfer
Yulim So, Seokho Kang
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 modified global directional loss (Lgdir) (ablation) | AnoStyler · MVTec-AD | I-AUC (%): 88.2 -> 95.2 | Improved |
| add modified patch-wise directional loss (Lpdir) (ablation) | AnoStyler · MVTec-AD | I-AUC (%): 95.2 -> 96.7 | Improved |
| add masked CLIP loss (Lmclip) (ablation) | AnoStyler · MVTec-AD | I-AUC (%): 96.7 -> 98.0 | Improved |
| add masked CLIP loss (Lmclip) (ablation) | AnoStyler · MVTec-AD | IS: 1.96 -> 2.04 | Improved |
| use baseline CLIPstyler objective without mask-aware losses (ablation) | AnoStyler · MVTec-AD | P-AUC (%): 94.4 -> 85.7 | Hurt |
| use text-driven style transfer instead of AnomalyAny diffusion | AnoStyler · MVTec-AD | IC-L: 0.33 -> 0.32 | No clear effect |
| use AnoStyler anomalies instead of AnomalyAny for detection | AnoStyler · MVTec-AD | P-AUC (%): 89.0 -> 94.4 | Improved |
| use AnoStyler anomalies instead of RealNet for detection | AnoStyler · VisA | I-AUC (%): 92.6 -> 93.9 | Improved |
| use AnoStyler instead of RealNet on VisA pixel AP | AnoStyler · VisA | P-AP (%): 33.3 -> 31.4 | Hurt |
| zero-shot AnoStyler vs few-shot AnoDiff | AnoStyler · MVTec-AD | I-AUC (%): 99.2 -> 98.0 | Hurt |
| lightweight U-Net style transfer instead of diffusion | AnoStyler · per-image cost | TFLOPs: 22.8 -> 9.5 | Improved |
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.
- RealNet — RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024) · better on 11 of 12 shared results
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
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.91)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.90)
- AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 2025 · similarity 0.88)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.87)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.87)
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.
