Research map / Label-efficient and robust learning / OOD and anomaly detection
Memory-Distilled Selection for Noise-Robust Anomaly Detection
Sirojbek Safarov, Jaewoo Park, Yoon Gyo Jung, Kuan-Chuan Peng, Wonchul Kim, Seongdeok Bang, Octavia Camps
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 MeDS to Dinomaly | Dinomaly · MVTecAD (40% noise) | I-AUROC: 87.38 -> 99.16 | Improved |
| add MeDS to INP-Former | INP-Former · MVTecAD (40% noise) | I-AUROC: 85.85 -> 99.17 | Improved |
| add MeDS to HVQ | HVQ · MVTecAD (40% noise) | I-AUROC: 92.14 -> 94.26 | Improved |
| add MeDS to Dinomaly | Dinomaly · MVTecAD (clean, 0% noise) | I-AUROC: 99.64 -> 99.37 | Hurt |
| add MeDS to Dinomaly | Dinomaly · MVTecAD (40% noise) | P-AP: 53.00 -> 68.05 | Improved |
| add MeDS to Dinomaly | Dinomaly · VisA (10% noise) | I-AUROC: 93.56 -> 97.43 | Improved |
| add MeDS to Dinomaly | Dinomaly · Real-IAD (40% noise, single-class) | I-AUROC: 87.78 -> 90.99 | Improved |
| add bootstrapped memory ensemble to Dinomaly (ablation) | Dinomaly · MVTecAD (40% noise), Dinomaly ablation | I-AUROC: 90.45 -> 92.22 | Improved |
| add bootstrapped memory ensemble to Dinomaly (ablation) | Dinomaly · MVTecAD (40% noise), Dinomaly ablation | P-AP: 54.75 -> 53.04 | Hurt |
| add memory distillation on top of memory ensemble (ablation) | Dinomaly · MVTecAD (40% noise), Dinomaly ablation | I-AUROC: 92.22 -> 98.76 | Improved |
| add progressive-selection fine-tuning after distillation (random init, memory criteria) (ablation) | Dinomaly · MVTecAD (40% noise), Dinomaly ablation | P-AP: 56.88 -> 67.71 | Improved |
| initialize fine-tuning and select data with distilled model instead of random init (ablation) | MeDS · MVTecAD (40% noise), Dinomaly ablation | P-AP: 67.91 -> 68.05 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Dinomaly — Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025) · better on 5 of 6 shared results
- SoftPatch — SoftPatch: Unsupervised Anomaly Detection with Noisy Data (NeurIPS 2022) · better on 3 of 3 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
- Anomaly Detection using Score-based Perturbation Resilience (ICCV 2023 · similarity 0.88)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · similarity 0.87)
- An Evidence-Based Post-Hoc Adjustment Framework for Anomaly Detection Under Data Contamination (NeurIPS 2025 · similarity 0.87)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.86)
- Towards Training-free Anomaly Detection with Vision and Language Foundation Models (CVPR 2025 · similarity 0.86)
- OmiAD: One-Step Adaptive Masked Diffusion Model for Multi-class Anomaly Detection via Adversarial Distillation (ICML 2025 · similarity 0.85)
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.
