Research map / Detection and segmentation / CNN features
SBNet: Sparse Blocks Network for Fast Inference
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 |
|---|---|---|---|
| Use road-map computation masks with SBNet | LiDAR detector · TOR4D test | AP: 75.70 -> 76.01 | Helped |
| Use road-map computation masks with SBNet | LiDAR detector · TOR4D test | Avg. speed-up: 1.0× -> 1.78× | Helped |
| Use PSPNet-predicted computation masks with SBNet | LiDAR detector · TOR4D test | AP: 73.28 -> 73.01 | Hurt |
| Use PSPNet-predicted computation masks with SBNet | LiDAR detector · TOR4D test | Avg. speed-up: 1.0× -> 2.66× | Helped |
| Apply sparse-block computation | LiDAR BEV detector · KITTI BEV 2017 | Avg. Time: 47.3 ms -> 17.9 ms | Helped |
| Apply sparse-block computation | LiDAR BEV detector · KITTI BEV 2017 Moderate | score: 77.05 -> 76.79 | Hurt |
| Apply sparse-block computation | LiDAR BEV detector · KITTI BEV 2017 Easy | score: 81.70 -> 81.90 | Helped |
| Apply sparse-block computation | LiDAR BEV detector · KITTI BEV 2017 Hard | score: 72.95 -> 71.40 | Hurt |
| Group multiple convolutions, batch normalizations, and ReLUs into one sparse block | SBNet residual units · Synthetic masks, 90% sparsity, conv-2 | speed-up: 3.39× -> 8.22× | Helped |
| Use PSPNet-predicted masks instead of road-map masks | SBNet residual units · conv-2 | speed-up: 3.05× -> 5.21× | Helped |
Closest papers by meaning
- 3D Semantic Segmentation With Submanifold Sparse Convolutional Networks (CVPR 2018 · similarity 0.89)
- Fast Sparse ConvNets (CVPR 2020 · similarity 0.89)
- Inducing and Exploiting Activation Sparsity for Fast Inference on Deep Neural Networks (ICML 2020 · similarity 0.89)
- Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling (ICLR 2023 · similarity 0.89)
- FSHNet: Fully Sparse Hybrid Network for 3D Object Detection (CVPR 2025 · similarity 0.88)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · similarity 0.88)
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.