Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Hyperbolic Defect Feature Synthesis for Few-Shot Defect Classification
Huimin Li, Boxuan Hu, Yulin Zhang, Xiuzhuang Zhou, Junlin Hu
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 AlphaCLIP mask weighting instead of equal mask weight (ablation) | AlphaCLIP ViT-L/14 · MTD 5-shot | mean accuracy: 51.9 -> 69.5 | Improved |
| use ViT backbone instead of ResNet (ablation) | VanillaClip · MTD 5-shot | mean accuracy: 24.3 (ResNet50) -> 52.7 (ViT-B/16) | Improved |
| use AlphaCLIP instead of vanilla CLIP (ablation) | ViT-L/14 · MTD 5-shot | mean accuracy: 52.5 -> 69.5 | Improved |
| set hyperbolic curvature c=0.1 instead of Euclidean c=0 (ablation) | HypDFS · MTD 5-shot | mean accuracy: 78.1 -> 79.1 | Improved |
| make curvature learnable (ablation) | HypDFS · MTD 5-shot | mean accuracy: 79.2 -> 78.9 | No clear effect |
| vary fixed curvature from 0.01 to 1.0 (ablation) | HypDFS · MTD 5-shot | mean accuracy: 78.9 to 79.2 | No clear effect |
| add residual adapter trained on real features (ablation) | HypDFS · MTD 5-shot | mean accuracy: 69.5 -> 78.1 | Improved |
| add synthetic hyperbolic defect features to adapter training (ablation) | HypDFS · MTD 5-shot | mean accuracy: 78.1 -> 86.4 | Improved |
| add hierarchical defect contrastive loss in hyperbolic space (ablation) | HypDFS · MTD 5-shot | mean accuracy: 79.2 -> 86.4 | Improved |
| use hyperbolic distance without synthetic features (ablation) | HypDFS · MTD 5-shot | mean accuracy: 78.1 -> 79.2 | Improved |
| replace Euclidean synthesis with hyperbolic synthesis | Zip-A-F · MTD 3-shot | mean accuracy: 67.6 -> 78.6 | Improved |
Closest papers by meaning
- MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context (AAAI 2025 · similarity 0.86)
- Hyperbolic Anomaly Detection (CVPR 2024 · similarity 0.85)
- DefectFill: Realistic Defect Generation with Inpainting Diffusion Model for Visual Inspection (CVPR 2025 · similarity 0.85)
- Wavelet and Prototype Augmented Query-based Transformer for Pixel-level Surface Defect Detection (CVPR 2025 · similarity 0.84)
- Exploring the Relationship Between Samples and Masks for Robust Defect Localization (AAAI 2025 · similarity 0.83)
- Towards Open-Vocabulary Industrial Defect Understanding with a Large-Scale Multimodal Dataset (CVPR 2026 · similarity 0.82)
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.
