Research map / Graph learning / GNN architectures
Vision HGNN: An Image is More than a Graph of Nodes
Yan Han, Peihao Wang, Souvik Kundu, Ying Ding, Zhangyang Wang
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 |
|---|---|---|---|
| replace k-NN graph with Fuzzy C-Means hypergraph (ablation) | Isotropic ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 73.0 -> 74.9 (E=50, with HSL) | Improved |
| add hypergraph structure learning (HSL) module (ablation) | Isotropic ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 74.1 -> 74.9 | Improved |
| use Fuzzy C-Means instead of other clustering methods (ablation) | Isotropic ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 74.7 -> 74.9 (vs spectral) | Improved |
| use 2 update loops instead of 1 (ablation) | Isotropic ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 74.9 -> 75.0 | No clear effect |
| use 3 update loops instead of 1 (ablation) | Isotropic ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 74.9 -> 74.4 | Hurt |
| increase hyperedges from 25 to 50 (ablation) | Isotropic ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 73.5 -> 74.9 | Improved |
| use 100 uniform hyperedges instead of 50 in pyramid model (ablation) | Pyramid ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 78.9 -> 77.6 | Hurt |
| halve hyperedges at each stage to match patch count (ablation) | Pyramid ViHGNN-Ti · ImageNet-1k | top-1 accuracy: 78.9 -> 79.4 | Improved |
| replace ViG graph with ViHGNN hypergraph (isotropic) | ViHGNN · ImageNet-1k | top-1 accuracy: 73.9 -> 74.3 (Ti); 80.4 -> 81.5 (S); 82.3 -> 82.9 (B) | Improved |
| use ViHGNN-S backbone for detection instead of Pyramid ViG-S | Pyramid ViHGNN-S · COCO RetinaNet 1x | mAP: 41.8 -> 42.2 | Improved |
| add Fuzzy C-Means hypergraph update overhead | Isotropic ViHGNN-Ti · training time per sample (ms) | total time: 29.77 -> 30.02 | No clear effect |
Closest papers by meaning
- Hypergraph Vision Transformers: Images are More than Nodes, More than Edges (CVPR 2025 · similarity 0.91)
- DVHGNN: Multi-Scale Dilated Vision HGNN for Efficient Vision Recognition (CVPR 2025 · similarity 0.91)
- GreedyViG: Dynamic Axial Graph Construction for Efficient Vision GNNs (CVPR 2024 · similarity 0.89)
- Vision Graph Prompting via Semantic Low-Rank Decomposition (ICML 2025 · similarity 0.87)
- K-hop Hypergraph Neural Network: A Comprehensive Aggregation Approach (AAAI 2025 · similarity 0.87)
- Deep Hypergraph Neural Networks with Tight Framelets (AAAI 2025 · 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.
