Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Inferring Prototypes for Multi-Label Few-Shot Image Classification with Word Vector Guided Attention
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 word-vector-guided local-feature attention to form visual prototypes | ML-FSIC · COCO | Macro-AP: 42.84 vs 35.30 | Helped |
| use word-vector-guided local-feature attention to form visual prototypes | ML-FSIC · PASCAL VOC | Macro-AP: 53.26 vs 47.29 | Helped |
| add the CMW-loss to align label and visual embeddings (ablation) | full model · COCO | Macro-AP: 36.17 -> 42.84 | Helped |
| use full local-feature attention rather than simple weighted global-feature attention (ablation) | full model · COCO | Macro-AP: 37.53 -> 42.84 | Helped |
| use local rather than global features in full attention (ablation) | full model · COCO | Macro-AP: 35.82 -> 42.84 | Helped |
| use full label-conditioned attention rather than simple attention on local features (ablation) | full model · COCO | Macro-AP: 31.21 -> 42.84 | Helped |
| use the proposed attention rather than low-rank bilinear pooling for local-feature weights (ablation) | full model · COCO | Macro-AP: 33.41 -> 42.84 | Helped |
| fine-tune parameters at test time using CMW-loss (ablation) | full model · COCO | Macro-AP: 42.84 -> 45.27 | Helped |
| fine-tune parameters at test time using CMW-loss (ablation) | full model · COCO | Micro-AP: 35.30 -> 35.70 | No clear effect |
| use GloVe-300 rather than the alternative reported embeddings (ablation) | full model · COCO | Macro-AP: FastText 35.67; Skip-Gram 38.67; GloVe-50 27.99; GloVe-100 33.48; GloVe-200 39.13; GloVe-300 42.84; BERT 36.46 | Helped |
| use 8 rather than fewer attention heads (ablation) | full model · COCO | Macro-AP: 1: 30.06; 2: 35.42; 4: 36.68; 6: 35.61; 8: 42.84 | Helped |
| increase the number of attention heads beyond 8 (ablation) | full model · COCO | Macro-AP: 8: 42.84; 10: 42.00; 12: 42.70; 14: 40.18; 16: 41.53 | No clear effect |
Closest papers by meaning
- Multi-Label Few-Shot Image Classification via Pairwise Feature Augmentation and Flexible Prompt Learning (AAAI 2025 · similarity 0.91)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · similarity 0.87)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · similarity 0.86)
- Epsilon: Exploring Comprehensive Visual-Semantic Projection for Multi-Label Zero-Shot Learning (AAAI 2025 · similarity 0.86)
- Multi-Label Prototype Visual Spatial Search for Weakly Supervised Semantic Segmentation (CVPR 2025 · similarity 0.86)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.86)
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.