Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Multi-Cue Zero-Shot Learning With Strong Supervision
Check 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 multiple visual parts (VP=19) during training instead of one (ablation) | Multi-cue joint embedding with attributes · CUB zero-shot classification, test VP=1 | Top-1 average per-class accuracy (%): 43.3 -> 47.0 | Helped |
| Use multiple visual parts (VP=19) at test time as well as training (ablation) | Multi-cue joint embedding with attributes · CUB zero-shot classification, train VP=19 | Top-1 average per-class accuracy (%): 47.0 -> 56.5 | Helped |
| Use multiple visual parts (VP=19) during training and testing (ablation) | Multi-cue joint embedding with word2vec · CUB zero-shot classification, unseen classes | Top-1 average per-class accuracy (%): 25.0 -> 32.1 | Helped |
| Use multiple visual parts (VP=19) during training and testing (ablation) | Multi-cue joint embedding with BoW · CUB zero-shot classification, unseen classes | Top-1 average per-class accuracy (%): 21.8 -> 26.0 | Helped |
| Use 50 NAD2 language parts instead of one word2vec language part (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=19, test VP=19 | Top-1 average per-class accuracy (%): 32.1 -> 33.9 | Helped |
| Use multiple language parts (NAD2 LP=50) with only one visual part at test time (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=19, test VP=1 | Top-1 average per-class accuracy (%): 26.8 -> 30.5 | Helped |
| Use NAD2 with VP=19 at test time instead of VP=1 (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=19 | Top-1 average per-class accuracy (%): 30.5 -> 33.9 | Helped |
| Use multiple language parts (NAD2 LP=50) with only one visual part during training and testing (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=1, test VP=1 | Top-1 average per-class accuracy (%): 25.0 -> 23.6 | Hurt |
| Combine word2vec and BoW language cues (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=19, test VP=1 | Top-1 average per-class accuracy (%): 26.8 -> 33.2 | Helped |
| Combine word2vec and BoW language cues (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=19, test VP=19 | Top-1 average per-class accuracy (%): 32.1 -> 34.7 | Helped |
| Combine NAD2 and BoW language cues (ablation) | Multi-cue joint embedding · CUB zero-shot classification, train VP=19, test VP=19 | Top-1 average per-class accuracy (%): 33.9 -> 34.3 | Helped |
| Use NAD3, selecting each class's associated attributes, instead of NAD2's top-n neighbors (ablation) | Multi-cue joint embedding · CUB zero-shot classification | Top-1 average per-class accuracy (%) | Hurt |
Closest papers by meaning
- VGSE: Visually-Grounded Semantic Embeddings for Zero-Shot Learning (CVPR 2022 · similarity 0.91)
- Epsilon: Exploring Comprehensive Visual-Semantic Projection for Multi-Label Zero-Shot Learning (AAAI 2025 · similarity 0.89)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.88)
- Class Semantic Attribute Perception Guided Zero-Shot Learning (AAAI 2025 · similarity 0.88)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.88)
- Data-Free Generalized Zero-Shot Learning (AAAI 2024 · similarity 0.88)
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.