Research map / Graph learning / Knowledge and temporal graphs
Zero-Shot Recognition via Semantic Embeddings and Knowledge Graphs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use graph convolutions over category relationships alongside semantic embeddings | zero-shot recognition on NELL/NEIL · NELL/NEIL All Edges | Hit@10 (%): 21.4 -> 49.0 | Helped |
| use graph convolutions to predict visual classifiers | zero-shot recognition on ImageNet · ImageNet 2-hops, unseen classes | Hit@5 (%): 32.3 -> 50.1 | Helped |
| use graph convolutions to predict visual classifiers | zero-shot recognition on ImageNet · ImageNet 3-hops, unseen classes | Hit@5 (%): 10.7 -> 13.1 | Helped |
| increase the GCN depth from 2 to 6 layers (ablation) | Ours on NELL/NEIL All Edges · NELL/NEIL All Edges | Hit@10 (%): 24.3 -> 49.0 | Helped |
| increase the GCN depth from 4 to 6 layers (ablation) | Ours on NELL/NEIL All Edges · NELL/NEIL All Edges | Hit@10 (%): 41.8 -> 49.0 | Helped |
| replace the semantic knowledge graph with a star graph (ablation) | Ours on NELL/NEIL All Edges · NELL/NEIL All Edges | Hit@10 (%): 49.0 -> 9.7 | Hurt |
| replace the semantic knowledge graph with a random graph (ablation) | Ours on NELL/NEIL All Edges · NELL/NEIL All Edges | Hit@10 (%): 49.0 -> 11.3 | Hurt |
| increase graph size from High Value Edges to All Edges (ablation) | Ours on NELL/NEIL · NELL/NEIL, 88 unseen classes | Hit@10 (%): 47.9 -> 49.0 | Helped |
| drop 5% to 10% of graph edges (ablation) | Ours on NELL/NEIL All Edges · NELL/NEIL All Edges | accuracy | No clear effect |
| drop more than 30% of graph edges (ablation) | Ours on NELL/NEIL All Edges · NELL/NEIL All Edges | accuracy | Hurt |
| replace Inception-v1 features with ResNet-50 features (ablation) | Ours on ImageNet · ImageNet 2-hops, unseen classes | Hit@5 (%): 50.1 -> 53.2 | Helped |
| change category embeddings from GloVe to FastText (ablation) | Ours on ImageNet · ImageNet 2-hops, unseen classes | Hit@5 (%): 50.1 -> 49.6 | No clear effect |
Closest papers by meaning
- VGSE: Visually-Grounded Semantic Embeddings for Zero-Shot Learning (CVPR 2022 · similarity 0.91)
- Generalized Zero-Shot Recognition Based on Visually Semantic Embedding (CVPR 2019 · similarity 0.90)
- Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning (CVPR 2024 · similarity 0.90)
- ZeroMamba: Exploring Visual State Space Model for Zero-Shot Learning (AAAI 2025 · similarity 0.89)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.88)
- Visual and Semantic Prompt Collaboration for Generalized Zero-Shot Learning (CVPR 2025 · 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.