Research map / Label-efficient and robust learning / OOD and anomaly detection
ReContrast: Domain-Specific Anomaly Detection via Contrastive Reconstruction
Jia Guo, shuai lu, LIze JIa, Weihang Zhang, Huiqi Li
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 |
|---|---|---|---|
| use global cosine distance instead of regional cosine distance (ablation) | RD4AD-style reconstruction · MVTec AD | I-AUROC (%): 95.31 -> 98.86 | Improved |
| jointly optimize encoder and decoder without safeguards (ablation) | feature reconstruction · MVTec AD | I-AUROC (%): 98.86 -> 91.54 | Hurt |
| add stop-gradient when optimizing encoder (ablation) | Config.C · MVTec AD | I-AUROC (%): 91.54 -> 94.64 | Improved |
| add contrastive pairs (frozen plus trained encoder views) (ablation) | Config.C · MVTec AD | I-AUROC (%): 91.54 -> 97.59 | Improved |
| combine stop-gradient and contrastive pairs (ablation) | Config.C · MVTec AD | I-AUROC (%): 91.54 -> 99.13 | Improved |
| add hard-normal mining (L_global-hm) (ablation) | Config.E · MVTec AD | I-AUROC (%): 99.13 -> 99.45 | Improved |
| add hard-normal mining to frozen baseline (ablation) | Config.B · MVTec AD | I-AUROC (%): 98.86 -> 99.00 | Improved |
| remove global cosine loss from full pipeline (ablation) | Config.E · APTOS | I-AUROC (%): 96.39 -> 97.32 | Improved |
| add global cosine loss to Config.E (ablation) | Config.E · MVTec AD | I-AUROC (%): 98.93 -> 99.13 | Improved |
| raise hard-mining threshold alpha from 1 to 2 (ablation) | ReContrast · MVTec AD | I-AUROC (%): 99.45 -> 98.85 | Hurt |
| use two encoders at inference (frozen plus trained) instead of trained only (ablation) | ReContrast · MVTec AD | I-AUROC (%): 99.38 -> 99.45 | No clear effect |
| use larger backbone WideResNet50 versus ResNet18 (ablation) | ReContrast · MVTec AD | I-AUROC (%): 98.7 -> 99.5 | Improved |
Later papers that compared with this method
6 later accepted papers report ReContrast as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- ArcAD: Anomaly-Rectified Calibration for Cold-Start Supervised Anomaly Detection (ECCV 2026 · better on 4 of 4 shared results)
- AnomalyMoE: Towards a Language-free Generalist Model for Unified Visual Anomaly Detection (AAAI 2026 · better on 0 of 1 shared result)
- UniNet: A Contrastive Learning-guided Unified Framework with Feature Selection for Anomaly Detection (CVPR 2025 · better on 5 of 5 shared results)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · better on 2 of 2 shared results)
- Scale-Aware Contrastive Reverse Distillation for Unsupervised Medical Anomaly Detection (ICLR 2025 · better on 1 of 1 shared result)
- GeneralAD: Anomaly Detection Across Domains by Attending to Distorted Features (ECCV 2024 · better on 1 of 4 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
- A Comprehensive Augmentation Framework for Anomaly Detection (AAAI 2024 · similarity 0.86)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.86)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.86)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · similarity 0.86)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.85)
- Unlocking the Potential of Reverse Distillation for Anomaly Detection (AAAI 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.
