Research map / Label-efficient and robust learning / OOD and anomaly detection
A Hierarchical Transformation-Discriminating Generative Model for Few Shot Anomaly Detection
Shelly Sheynin, Sagie Benaim, Lior Wolf
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 the generative component (fake class) to the discriminator (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 60.7 -> 64.9 (1-shot); 64.9 -> 67.5 (5-shot) | Improved |
| add transformation discrimination instead of real/fake only (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 59.1 -> 64.9 (1-shot); 60.0 -> 67.5 (5-shot) | Improved |
| discriminate transformations rather than use them as augmentation (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 59.7 -> 64.9 (1-shot); 63.4 -> 67.5 (5-shot) | Improved |
| add the multi-scale hierarchy over single scale s=100 (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 57.6 -> 64.9 (1-shot); 57.8 -> 67.5 (5-shot) | Improved |
| add the multi-scale hierarchy over single scale s=20 (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 57.3 -> 64.9 (1-shot); 63.8 -> 67.5 (5-shot) | Improved |
| use full model instead of no components (MSE baseline) (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 47.7 -> 64.9 (1-shot); 48.0 -> 67.5 (5-shot) | Improved |
| train the model end to end instead of GEOM on generated samples (ablation) | hierarchical transformation-discriminating generative model · CIFAR10 (1-shot and 5-shot) | average AUC: 58.8 -> 64.9 (1-shot); 63.7 -> 67.5 (5-shot) | Improved |
| increase training images from one to five | full method · Paris | AUC (%): 66.6% to 79.8% | Improved |
| increase number of training samples to 50 and 80 shots | full method vs GEOM · CIFAR10 | AUC | Mixed |
| average over 5% vs 10% of lowest-scoring patches for defect detection (ablation) | full method · MVTec | average AUC: 5% best for one-shot, 10% best for five and ten shot | No clear effect |
Closest papers by meaning
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.89)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.88)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.88)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.87)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.87)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (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.
