Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Modeling Inter and Intra-Class Relations in the Triplet Loss for Zero-Shot Learning
Yannick Le Cacheux, Herve Le Borgne, Michel Crucianu
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 partial normalization of projected visual features (ablation) | triplet loss ZSL model (theta+phi) · CUB ZSL | per-class accuracy (%): 57.6 -> 63.8 | Improved |
| add flexible semantic margin (ablation) | triplet loss ZSL model (theta+phi) · CUB ZSL | per-class accuracy (%): 61.8 -> 63.8 | Improved |
| add relevance weighting of training samples (ablation) | triplet loss ZSL model (theta+phi) · CUB ZSL | per-class accuracy (%): 61.7 -> 63.8 | Improved |
| compute class distances with Mahalanobis distance (attribute correlations) rather than Euclidean (ablation) | flexible semantic margin · CUB ZSL | per-class accuracy (%): 61.7 -> 63.8 | Improved |
| add all three components (FSM, PN, RW) to a DEVISE-like triplet loss (ablation) | triplet loss ZSL model (theta+phi) · CUB ZSL | per-class accuracy (%): 56.6 -> 63.8 | Improved |
| add all three components (FSM, PN, RW) (ablation) | triplet loss ZSL model (theta only) · CUB ZSL | per-class accuracy (%): 55.0 -> 61.3 | Improved |
| learn projection of class prototypes too (theta+phi instead of theta only) (ablation) | full model · CUB ZSL | per-class accuracy (%): 61.3 -> 63.8 | Improved |
| apply calibration to penalize seen classes at prediction | full model · CUB GZSL | harmonic mean H (%): 41.2 -> 53.0 | Improved |
| apply calibration to penalize seen classes at prediction | full model · GZSL average over datasets | harmonic mean H (%): 32.3 -> 50.4 | Improved |
| combine relevance weighting and flexible margin (ablation) | triplet loss ZSL model · CUB ZSL | per-class accuracy (%): 56.6 -> 61.0 | Improved |
Closest papers by meaning
- Rethinking Zero-Shot Learning: A Conditional Visual Classification Perspective (ICCV 2019 · similarity 0.92)
- Adaptive and Generative Zero-Shot Learning (ICLR 2021 · similarity 0.92)
- Leveraging Seen and Unseen Semantic Relationships for Generative Zero-Shot Learning (ECCV 2020 · similarity 0.92)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.91)
- Class Semantic Attribute Perception Guided Zero-Shot Learning (AAAI 2025 · similarity 0.90)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.89)
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.
