Research map / Detection and segmentation / CNN features
Beyond Grids: Learning Graph Representations for Visual Recognition
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 |
|---|---|---|---|
| add 4 GCUs to Dilated FCN | Dilated FCN ResNet-50 · ADE20K val | mIoU: 35.60 -> 42.60 | Helped |
| add GCU | Dilated FCN · ADE20K val | pixel accuracy | Hurt |
| add GCUs to Mask RCNN | Mask RCNN (Detectron) ResNet-50 · COCO minival | AP box: 37.7 -> 38.7 | Helped |
| add GCUs to Mask RCNN | Mask RCNN (Detectron) ResNet-101 · COCO minival | AP seg: 35.9 -> 36.9 | Helped |
| add a single GCU with 2 vertices (ablation) | Dilated FCN ResNet-50 · ADE20K val | mIoU: 35.6% -> 39.43% | Helped |
| increase number of GCUs from 2 to 4 (ablation) | GCU · ADE20K val | mIoU: 40.92% -> 42.60% | Helped |
| increase vertices in last GCU 32 -> 64 (ablation) | GCU · ADE20K val | mIoU: 42.60% -> 42.58% | No clear effect |
| add diversity regularization (ablation) | GCU · ADE20K val | mIoU: 41.34 -> 41.73% | Helped |
| add clustering-based initialization (ablation) | GCU · ADE20K val | mIoU: 41.73% -> 42.60% | Helped |
Closest papers by meaning
- Learning Convolutional Neural Networks for Graphs (ICML 2016 · similarity 0.88)
- Pixels to Graphs by Associative Embedding (NeurIPS 2017 · similarity 0.88)
- LatentGNN: Learning Efficient Non-local Relations for Visual Recognition (ICML 2019 · similarity 0.88)
- Classification by Attention: Scene Graph Classification with Prior Knowledge (AAAI 2021 · similarity 0.88)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.87)
- The Underappreciated Power of Vision Models for Graph Structural Understanding (NeurIPS 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.