Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Visual Structure Constraint for Transductive Zero-Shot Learning in the Wild
Ziyu Wan, Dongdong Chen, Jing Liao
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 visual structure constraint with Wasserstein distance | VCL baseline · AwA2 (standard split, ResNet) | MCA (%): 82.5 -> 96.7 | Improved |
| add visual structure constraint with Chamfer distance | VCL baseline · CUB (standard split, ResNet) | MCA (%): 60.1 -> 74.2 | Improved |
| add structure constraint (Wasserstein) on proposed split | VCL baseline · AwA2 (proposed split) | MCA (%): 61.5 -> 87.3 | Improved |
| add structure constraint (Wasserstein) in generalized ZSL | VCL baseline · AwA2 (generalized ZSL) | harmonic mean H (%): 34.6 -> 81.8 | Improved |
| add structure constraint (Wasserstein) in generalized ZSL | VCL baseline · CUB (generalized ZSL) | harmonic mean H (%): 26.5 -> 57.5 | Improved |
| replace hard one-to-one bipartite matching with soft Wasserstein coupling | BMVSc · AwA2 (proposed split) | MCA (%): 81.7 -> 87.3 | Improved |
| replace Chamfer distance with bipartite matching | CDVSc · CUB (generalized ZSL) | harmonic mean H (%): 51.4 -> 47.9 | Hurt |
| apply structure constraint directly on test set mixed with unrelated images (ablation) | WDVSc · AwA2 (standard split, with 8K unrelated images) | MCA (%): 82.5 -> 81.3 | Hurt |
| add filtering of unrelated images (S+) before the constraint (ablation) | WDVSc · AwA2 (standard split, with 8K unrelated images) | MCA (%): 81.3 -> 92.4 | Improved |
| add filtering of unrelated images (S+) before the constraint (ablation) | WDVSc · AwA2 (proposed split, with 8K unrelated images) | MCA (%): 60.8 -> 78.3 | Improved |
| increase number of K-means clusters from 3 to 10 (ablation) | WDVSc · AwA2 (proposed split) | MCA (%): 63.4 -> 87.3 | Improved |
| increase number of K-means clusters from 3 to 10 (ablation) | BMVSc · AwA2 (proposed split) | MCA (%): 62.3 -> 81.7 | Improved |
Closest papers by meaning
- Transductive Zero-Shot Learning with Visual Structure Constraint (NeurIPS 2019 · similarity 0.98)
- Domain-Invariant Projection Learning for Zero-Shot Recognition (NeurIPS 2018 · similarity 0.90)
- Semantics Disentangling for Generalized Zero-Shot Learning (ICCV 2021 · similarity 0.90)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.89)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · 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.
