Research map / Graph learning / GNN architectures
DeepGCNs: Can GCNs Go As Deep As CNNs?
Guohao Li, Matthias Muller, Ali Thabet, Bernard Ghanem
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 residual graph connections (ablation) | 28-layer GCN · S3DIS Area 5 | mIoU: 40.31 -> 52.49 | Improved |
| add dilated graph convolutions (ablation) | ResGCN-28 · S3DIS Area 5 | mIoU: 49.64 -> 52.49 | Improved |
| add stochasticity to dilated k-NN (ablation) | ResGCN-28 · S3DIS Area 5 | mIoU: 51.98 -> 52.49 | Improved |
| add dynamic k-NN (update edges every layer) (ablation) | ResGCN-28 · S3DIS Area 5 | mIoU: 48.38 -> 52.49 | Improved |
| use dilation without residual connections (ablation) | 28-layer GCN · S3DIS Area 5 | mIoU: 43.43 -> 40.31 | No clear effect |
| use dense instead of residual connections (ablation) | 28-layer GCN · S3DIS Area 5 | mIoU: 52.49 -> 51.27 | No clear effect |
| increase neighbors from 8 to 16 (ablation) | ResGCN-28 · S3DIS Area 5 | mIoU: 49.98 -> 52.49 | Improved |
| increase neighbors from 4 to 16 (ablation) | ResGCN-28 · S3DIS Area 5 | mIoU: 49.22 -> 52.49 | Improved |
| increase depth from 28 to 56 layers with residuals (ablation) | ResGCN · S3DIS Area 5 | mIoU: 52.49 -> 53.64 | Improved |
| increase depth from 7 to 28 layers with residuals (ablation) | ResGCN · S3DIS Area 5 | mIoU: 48.95 -> 52.49 | Improved |
| increase depth from 7 to 28 layers without residuals (ablation) | PlainGCN · S3DIS Area 5 | mIoU: 47.92 -> 40.47 | Hurt |
| increase filters from 64 to 128 (8 NNs) (ablation) | ResGCN-28 · S3DIS Area 5 | mIoU: 52.49 -> 53.78 | Improved |
Closest papers by meaning
- DeepGCNs: Making GCNs Go as Deep as CNNs (TPAMI 2021 · similarity 0.96)
- Simple and Deep Graph Convolutional Networks (ICML 2020 · similarity 0.91)
- On Provable Benefits of Depth in Training Graph Convolutional Networks (NeurIPS 2021 · similarity 0.91)
- Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective (ICLR 2022 · similarity 0.91)
- Deeper Insights Into Deep Graph Convolutional Networks: Stability and Generalization (TPAMI 2025 · similarity 0.90)
- pathGCN: Learning General Graph Spatial Operators from Paths (ICML 2022 · similarity 0.90)
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.
