Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Semantic Diversity Learning for Zero-Shot Multi-Label Classification
Avi Ben-Cohen, Nadav Zamir, Emanuel Ben-Baruch, Itamar Friedman, Lihi Zelnik-Manor
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 SDW (semantic diversity weight) to the baseline (ablation) | Fast0Tag-style baseline · NUS-WIDE ZSL | mAP: 20.6 -> 22.7 | Improved |
| add SDW to the full method with M=7 and Reg 0.3 (ablation) | Ours · NUS-WIDE ZSL | mAP: 25.8 -> 25.9 | Improved |
| add SDW to the full method with M=7 and Reg 0.3 (ablation) | Ours · NUS-WIDE GZSL | mAP: 11.8 -> 12.1 | Mixed |
| add SDW to samples with more than 6 labels (ablation) | Ours · NUS-WIDE ZSL, diverse images | F1 (top-10): 42.4 -> 42.9 | Improved |
| use multiple principal directions with our loss instead of baseline (ablation) | Baseline M=7 · NUS-WIDE ZSL, diverse images | F1 (top-10): 37.1 -> 42.4 | Improved |
| add multiple principal directions naively (M=7, no reg) (ablation) | Fast0Tag in our framework · NUS-WIDE ZSL | mAP: 21.6 -> 20.6 | Hurt |
| add regularization (0.1) with M=2 (ablation) | SDW M=2 · NUS-WIDE ZSL | mAP: 22.2 -> 23.8 | Improved |
| increase principal directions from M=2 to M=7 (ablation) | SDW + Reg 0.1 · NUS-WIDE ZSL | mAP: 23.8 -> 25.1 | Improved |
| increase regularization from 0.1 to 0.3 at M=7 (no SDW) (ablation) | Ours · NUS-WIDE ZSL | mAP: 25.1 -> 25.8 | Improved |
| use ResNet50 instead of VGG19 backbone (ablation) | Ours · NUS-WIDE ZSL | mAP: 24.2 -> 24.4 | Improved |
| use TResNet-M instead of ResNet50 backbone (ablation) | Ours · NUS-WIDE ZSL | mAP: 24.4 -> 25.9 | Improved |
| use our loss instead of Fast0Tag loss with TResNet-M (ablation) | TResNet-M · NUS-WIDE ZSL | mAP: 21.6 -> 25.9 | Improved |
Closest papers by meaning
- A Shared Multi-Attention Framework for Multi-Label Zero-Shot Learning (CVPR 2020 · similarity 0.89)
- Multilabel Deep Visual-Semantic Embedding (TPAMI 2019 · similarity 0.89)
- Epsilon: Exploring Comprehensive Visual-Semantic Projection for Multi-Label Zero-Shot Learning (AAAI 2025 · similarity 0.88)
- Label-Embedding for Image Classification (TPAMI 2016 · similarity 0.88)
- Heterogeneous Semantic Transfer for Multi-label Recognition with Partial Labels (IJCV 2024 · similarity 0.88)
- Multi-Label Learning From Single Positive Labels (CVPR 2021 · similarity 0.87)
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.
