Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
En-Compactness: Self-Distillation Embedding & Contrastive Generation for Generalized Zero-Shot Learning
Xia Kong, Zuodong Gao, Xiaofan Li, Ming Hong, Jun Liu, Chengjie Wang, Yuan Xie, Yanyun Qu
Check 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 representation distillation and normalized logits distillation together (ablation) | ICCE · AWA1, AWA2, CUB, FLO, APY | H: 67.7 -> 72.5; 66.6 -> 71.8; 64.0 -> 64.9; 70.3 -> 72.5; 27.1 -> 40.5 | Improved |
| Add representation distillation loss only (ablation) | ICCE · AWA1, AWA2, CUB, FLO, APY | H: 67.7 -> 70.4; 66.6 -> 69.3; 64.0 -> 64.5; 70.3 -> 71.7; 27.1 -> 24.1 | Mixed |
| Add normalized logits distillation loss only (ablation) | ICCE · AWA1, AWA2, CUB, FLO, APY | H: 67.7 -> 69.3; 66.6 -> 62.6; 64.0 -> 62.6; 70.3 -> 69.7; 27.1 -> 28.6 | Mixed |
| Add synthetic self-distillation loss to generator training (ablation) | ICCE with both embedding distillation losses · AWA1, AWA2, CUB, FLO, APY | H: 72.5 -> 73.1; 71.8 -> 72.1; 64.9 -> 65.9; 72.5 -> 74.5; 40.5 -> 41.5 | Improved |
| Add instance-wise semantic-visual contrastive loss (ablation) | ICCE with synthetic self-distillation · AWA1, AWA2, CUB, FLO, APY | H: 73.1 -> 73.6; 72.1 -> 72.8; 65.9 -> 66.4; 74.5 -> 74.9; 41.5 -> 45.7 | Improved |
| Use a momentum-updated teacher instead of constant random, student-copy, previous-epoch, or previous-iteration teachers (ablation) | SDE · AWA1 | H: 70.5; 71.8; 71.5; 71.5 -> 73.6 | Improved |
| Use a momentum-updated teacher instead of constant random, student-copy, previous-epoch, or previous-iteration teachers (ablation) | SDE · APY | H: 26.8; 40.9; 20.7; 30.5 -> 45.7 | Improved |
| Increase the number of generated unseen-class samples per class (ablation) | ICCE · AWA1, AWA2, CUB, FLO, APY | unseen-class accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- FREE — FREE: Feature Refinement for Generalized Zero-Shot Learning (ICCV 2021) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Contrastive Embedding for Generalized Zero-Shot Learning (CVPR 2021 · similarity 0.93)
- Semantic Contrastive Embedding for Generalized Zero-Shot Learning (IJCV 2022 · similarity 0.93)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.91)
- Data Distribution Distilled Generative Model for Generalized Zero-Shot Recognition (AAAI 2024 · similarity 0.90)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.89)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 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.
