Research map / Detection and segmentation / CNN features
Heterogeneous Grid Convolution for Adaptive, Efficient, and Controllable Computation
Check 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 |
|---|---|---|---|
| Replace regular convolutions with HG-Conv | Semantic segmentation models · Semantic segmentation | FLOPs: more than 90% reduction | Helped |
| Use HG-Conv in ResNet34-DCN | PASCAL-context segmentation · PASCAL-context | performance: +0.7% | Helped |
| Add HG-Conv to DCN | Semantic segmentation · Semantic segmentation | FPS: 14.9 FPS -> 6.5 FPS | Hurt |
| Use importance-based or top-k plus random cluster-center sampling instead of random sampling (ablation) | HG-ResNet segmentation · Semantic segmentation | performance | Helped |
| Use an extremely small downsampling ratio (ablation) | HG-ResNet · Semantic segmentation | performance | Helped |
| Apply noise-canceling to group adjacency matrices (ablation) | HG-Conv · Semantic segmentation | performance | Helped |
| Keep only the strongest adjacency direction (ablation) | HG-Conv · Semantic segmentation | performance | Mixed |
| Convert the 3rd stage of ResNet101 to HG-Conv (ablation) | ResNet101 · Semantic segmentation | performance: 79.9% -> 78.1% | Hurt |
| Convert the 3rd stage of ResNet101 to HG-Conv (ablation) | ResNet101 · Semantic segmentation | computational cost reduction: 15.1 % -> 54.7 % | Helped |
| Replace ResNet-DCN with HG-ResNet-DCN | COCO object localization · COCO | accuracy | Hurt |
| Replace ResNet-DCN with HG-ResNet-DCN | COCO object localization · COCO | FLOPs: 30% reduction | Helped |
| Use HG-Conv in ResUNet18+ | DeepGlobe road extraction · DeepGlobe | IoU: +0.6% | Helped |
Closest papers by meaning
- DiffGCN: Graph Convolutional Networks via Differential Operators and Algebraic Multigrid Pooling (NeurIPS 2020 · similarity 0.87)
- Grid R-CNN (CVPR 2019 · similarity 0.86)
- Full-Resolution Residual Networks for Semantic Segmentation in Street Scenes (CVPR 2017 · similarity 0.86)
- Spatial Pyramid Based Graph Reasoning for Semantic Segmentation (CVPR 2020 · similarity 0.85)
- Revisiting the Integration of Convolution and Attention for Vision Backbone (NeurIPS 2024 · similarity 0.85)
- DHGCN: Dynamic Hop Graph Convolution Network for Self-Supervised Point Cloud Learning (AAAI 2024 · similarity 0.84)
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.