Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Multi-modal Cycle-consistent Generalized Zero-Shot Learning
Rafael Felix, Vijay B. G. Kumar, Ian Reid, Gustavo Carneiro
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 multi-modal cycle-consistency loss (cycle-WGAN) | f-CLSWGAN baseline · CUB GZSL | harmonic mean H: 50.8 -> 52.2 | Improved |
| add multi-modal cycle-consistency loss (cycle-WGAN) | f-CLSWGAN baseline · FLO GZSL | harmonic mean H: 63.9 -> 64.5 | Improved |
| add multi-modal cycle-consistency loss (cycle-WGAN) | f-CLSWGAN baseline · SUN GZSL | harmonic mean H: 38.7 -> 39.2 | Improved |
| add multi-modal cycle-consistency loss (cycle-WGAN) | f-CLSWGAN baseline · AWA GZSL | harmonic mean H: 59.2 -> 59.7 | Improved |
| add cycle loss on top of classification loss (cycle-CLSWGAN) | f-CLSWGAN baseline · SUN GZSL | harmonic mean H: 38.7 -> 40.0 | Improved |
| also use synthetic features of unseen classes in cycle loss (cycle-(U)WGAN) instead of seen only (ablation) | cycle-WGAN · CUB GZSL | harmonic mean H: 52.2 -> 53.0 | Improved |
| also use synthetic features of unseen classes in cycle loss (cycle-(U)WGAN) instead of seen only (ablation) | cycle-WGAN · SUN GZSL | harmonic mean H: 39.2 -> 39.4 (unseen acc 48.3 -> 47.2) | Mixed |
| also use synthetic features of unseen classes in cycle loss (cycle-(U)WGAN) instead of seen only (ablation) | cycle-WGAN · AWA GZSL | unseen accuracy: 56.4 -> 59.6 | Improved |
| add cycle-consistency loss with unseen-class synthesis (cycle-(U)WGAN) | f-CLSWGAN baseline · FLO ZSL | top-1 accuracy: 66.8 -> 70.3 | Improved |
| add cycle-consistency loss with unseen-class synthesis (cycle-(U)WGAN) | f-CLSWGAN baseline · CUB ZSL | top-1 accuracy: 57.7 -> 58.6 | Improved |
| add cycle-consistency loss with unseen-class synthesis (cycle-(U)WGAN) | f-CLSWGAN baseline · AWA ZSL | top-1 accuracy: 64.1 -> 66.8 | Improved |
| add cycle-consistency loss with unseen-class synthesis (cycle-(U)WGAN) | f-CLSWGAN baseline · SUN ZSL | top-1 accuracy: 58.5 -> 59.9 | Improved |
Closest papers by meaning
- Dual Adversarial Semantics-Consistent Network for Generalized Zero-Shot Learning (NeurIPS 2019 · similarity 0.92)
- Distinguishing Unseen From Seen for Generalized Zero-Shot Learning (CVPR 2022 · similarity 0.91)
- Meta-Learning for Generalized Zero-Shot Learning (AAAI 2020 · similarity 0.91)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.91)
- Semantics Disentangling for Generalized Zero-Shot Learning (ICCV 2021 · similarity 0.90)
- Adaptive and Generative Zero-Shot Learning (ICLR 2021 · 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.
