Research map / Label-efficient and robust learning / OOD and anomaly detection
Multiresolution Knowledge Distillation for Anomaly Detection
Mohammadreza Salehi, Niousha Sadjadi, Soroosh Baselizadeh, Mohammad H. Rohban, Hamid R. Rabiee
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 additional intermediate max-pooling-layer hints to distillation (ablation) | Multiresolution Knowledge Distillation · MVTecAD | AUROC: 79.22 -> 83.02 -> 87.74 | Improved |
| Add additional intermediate max-pooling-layer hints to distillation (ablation) | Multiresolution Knowledge Distillation · MNIST | AUROC: 95.33 -> 97.74 -> 98.71 | Improved |
| Use a smaller cloner instead of a source-sized cloner (ablation) | Multiresolution Knowledge Distillation · MVTecAD | AUROC: 85.58 -> 87.74 | Improved |
| Use a smaller cloner instead of a source-sized cloner (ablation) | Multiresolution Knowledge Distillation · CIFAR-10 | AUROC: 86.87 -> 87.18 | Improved |
| Add directional loss to MSE loss (ablation) | Multiresolution Knowledge Distillation · CIFAR-10 | AUROC: 70.16 -> 87.2 | Improved |
| Add MSE loss to directional loss (ablation) | Multiresolution Knowledge Distillation · MVTecAD | AUROC: 83.74 -> 87.74 | Mixed |
| Add 20 degree rotation and scaling in range [0.9, 1.05] (ablation) | Multiresolution Knowledge Distillation · HeadCT | AUROC: 78.04 ± 0.225 -> 80.42 ± 0.006 | Improved |
| Add Gaussian filtering to gradient localization maps (ablation) | Multiresolution Knowledge Distillation · MVTecAD | pixel-wise AUROC: 86.16 -> 90.51 | Improved |
| Use SmoothGrad rather than simple gradients without Gaussian filtering (ablation) | Multiresolution Knowledge Distillation · MVTecAD | pixel-wise AUROC: 86.16 -> 86.97 | Improved |
| Use SmoothGrad rather than simple gradients with Gaussian filtering (ablation) | Multiresolution Knowledge Distillation · MVTecAD | pixel-wise AUROC: 90.51 -> 90.54 | No clear effect |
| Use GBP rather than simple gradients without Gaussian filtering (ablation) | Multiresolution Knowledge Distillation · MVTecAD | pixel-wise AUROC: 86.16 -> 84.38 | Hurt |
Closest papers by meaning
- Unlocking the Potential of Reverse Distillation for Anomaly Detection (AAAI 2025 · similarity 0.90)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (CVPR 2025 · similarity 0.88)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.88)
- Co-Progression Knowledge Distillation with Knowledge Prototype for Industrial Anomaly Detection (AAAI 2025 · similarity 0.87)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.87)
- Boosting Fine-Grained Visual Anomaly Detection with Coarse-Knowledge-Aware Adversarial Learning (AAAI 2025 · 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.
