Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Semantic Contrastive Embedding for Generalized Zero-Shot Learning
Zongyan Han, Zhenyong Fu, Shuo Chen, Jian Kang Yang
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 semantic embedding to the generator and classify in a new embedding space (hybrid) (ablation) | Gen · CUB (GZSL) | harmonic mean H (%): 56.6 -> 63.1 | Improved |
| simply combine generation with semantic embedding and classify in semantic space (ablation) | Gen · CUB (GZSL) | harmonic mean H (%): 56.6 -> 55.5 | Hurt |
| use new embedding space instead of semantic space (ablation) | Gen+SE · CUB (GZSL) | harmonic mean H (%): 55.5 -> 63.1 | Improved |
| replace semantic embedding with contrastive embedding in the embedding space (ablation) | Gen+SE · CUB (GZSL) | harmonic mean H (%): 63.1 -> 65.3 | Improved |
| replace semantic embedding with contrastive embedding in the embedding space (ablation) | Gen+SE · AWA1 (GZSL) | harmonic mean H (%): 66.9 -> 69.1 | Improved |
| replace semantic embedding with contrastive embedding in semantic space (ablation) | Gen+SE (basic) · CUB (GZSL) | harmonic mean H (%): 55.5 -> 64.3 | Improved |
| add class-level loss to instance-level contrastive loss (ablation) | CE-GZSL · CUB (GZSL) | harmonic mean H (%): 62.4 -> 65.3 | Improved |
| add instance-level loss to class-level contrastive loss (ablation) | CE-GZSL · CUB (GZSL) | harmonic mean H (%): 63.0 -> 65.3 | Improved |
| add class-level loss to instance-level contrastive loss (ablation) | CE-GZSL · AWA2 (GZSL) | harmonic mean H (%): 68.1 -> 70.0 | Improved |
| increase embedding dimension from 256 to 2,048 (ablation) | CE-GZSL · AWA1 (GZSL) | harmonic mean H (%): 53.3 -> 69.0 | Improved |
| increase embedding dimension from 2,048 to 4,096 (ablation) | CE-GZSL · CUB (GZSL) | harmonic mean H (%): 65.3 -> 65.2 | No clear effect |
| use large random batch of 4,096 instead of 1 positive and 50 negatives (ablation) | CE-GZSL · AWA1 (GZSL) | harmonic mean H (%): 64.9 -> 69.1 | Improved |
Closest papers by meaning
- Contrastive Embedding for Generalized Zero-Shot Learning (CVPR 2021 · similarity 0.99)
- En-Compactness: Self-Distillation Embedding & Contrastive Generation for Generalized Zero-Shot Learning (CVPR 2022 · similarity 0.93)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.92)
- Data Distribution Distilled Generative Model for Generalized Zero-Shot Recognition (AAAI 2024 · similarity 0.91)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.90)
- GenZSL: Generative Zero-Shot Learning Via Inductive Variational Autoencoder (ICML 2025 · similarity 0.90)
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.
