Research map / Graph learning / Knowledge and temporal graphs
Representative Graph Neural Network
Changqian Yu, Yifan Liu, Changxin Gao, Chunhua Shen, Nong Sang
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 non-local block with RepGraph layer (S=9) | R50 baseline · ADE20K val (ablation, R50) | mIoU: 40.97 -> 43.12 | Improved |
| use a single representative node per position (S=1) (ablation) | RepGraph layer · ADE20K val (ablation, R50) | mIoU: 36.48 -> 42.53 | Improved |
| increase representative nodes from 9 to 27 (ablation) | RepGraph layer · ADE20K val (ablation, R50) | mIoU: 43.12 -> 43.06 | No clear effect |
| use bottleneck RepGraph instead of simple RepGraph (concat) (ablation) | RepGraph layer · ADE20K val (instantiation, R50) | mIoU: 42.98 -> 43.12 | Improved |
| use concatenation instead of summation fusion (ablation) | Bottleneck RepGraph · ADE20K val (instantiation, R50) | mIoU: 42.67 -> 43.12 | Improved |
| use concatenation instead of summation fusion (ablation) | Non-local · ADE20K val (instantiation, R50) | mIoU: 40.3 -> 40.97 | Improved |
| insert RepGraph layer at res3 vs res2/res4/res5 (ablation) | Bottleneck RepGraph · ADE20K val (insertion stage, R50) | mIoU: 41.25 -> 41.59 | No clear effect |
| add RepGraph layers (1 to 3 layers) (ablation) | Bottleneck RepGraph · ADE20K val (number of layers, R50) | mIoU: 41.59 -> 41.86 | Improved |
| add RepGraph layers (3 to 5 layers) (ablation) | Bottleneck RepGraph · ADE20K val (number of layers, R50) | mIoU: 41.86 -> 41.76 | No clear effect |
| group pixels spatially into nodes (Grid RepGraph g_s=65) (ablation) | RepGraph layer · ADE20K val (Grid RepGraph, R50) | mIoU: 43.12 -> 41.23 | Hurt |
| group channels (Group RepGraph g_c=32) (ablation) | RepGraph layer · ADE20K val (Group RepGraph, R50) | mIoU: 43.12 -> 42.38 | Hurt |
| group channels (g_c=8) (ablation) | RepGraph layer · ADE20K val (Group RepGraph, R50) | mIoU: 43.12 -> 43.01 | No clear effect |
Closest papers by meaning
- A Fair Comparison of Graph Neural Networks for Graph Classification (ICLR 2020 · similarity 0.83)
- RepSGG: Novel Representations of Entities and Relationships for Scene Graph Generation (TPAMI 2024 · similarity 0.83)
- Dynamic Graph Message Passing Networks for Visual Recognition (TPAMI 2022 · similarity 0.83)
- Learning 3D Semantic Scene Graphs with Instance Embeddings (IJCV 2022 · similarity 0.83)
- GPEN: Global Position Encoding Network for Enhanced Subgraph Representation Learning (ICML 2025 · similarity 0.82)
- RepAn: Enhanced Annealing through Re-parameterization (CVPR 2024 · similarity 0.82)
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.
