Research map / Detection and segmentation / CNN features
DeepGCNs: Making GCNs Go as Deep as CNNs
Guohao Li, Matthias F. Mueller, Guocheng Qian, Itzel C. Delgadillo, Abdulellah Abualshour, 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) | PlainGCN-28 · 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 dynamic k-NN edges (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 48.38 -> 49.64 | Improved |
| use dilation instead of downsampling (FPS) (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 47.98 -> 52.49 | Improved |
| use dilation instead of downsampling (random) (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 44.30 -> 52.49 | Improved |
| add stochastic dilation (epsilon 0.2 vs 0.0) (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 51.98 -> 52.49 | No clear effect |
| residual connection every layer instead of every two layers (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 50.86 -> 52.49 | Improved |
| increase depth from 7 to 56 layers with residual connections (ablation) | ResGCN · S3DIS area 5 | mIoU: 48.95 -> 53.64 | Improved |
| increase depth from 56 to 112 layers (ablation) | ResGCN · S3DIS area 5 | mIoU: 53.64 -> 51.97 | Mixed |
| increase depth without residual connections (ablation) | PlainGCN · S3DIS area 5 | mIoU: 49.23 -> 40.47 | Hurt |
| increase number of neighbors from 4 to 16 (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 49.22 -> 52.49 | Improved |
| increase filters from 64 to 128 (ablation) | ResGCN-28 · S3DIS area 5 | mIoU: 52.49 -> 53.78 | Improved |
Closest papers by meaning
- DeepGCNs: Can GCNs Go As Deep As CNNs? (ICCV 2019 · similarity 0.96)
- Simple and Deep Graph Convolutional Networks (ICML 2020 · similarity 0.90)
- Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective (ICLR 2022 · similarity 0.90)
- On Provable Benefits of Depth in Training Graph Convolutional Networks (NeurIPS 2021 · similarity 0.89)
- pathGCN: Learning General Graph Spatial Operators from Paths (ICML 2022 · similarity 0.89)
- Deeper Insights Into Deep Graph Convolutional Networks: Stability and Generalization (TPAMI 2025 · similarity 0.89)
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.
