Research map / Detection and segmentation / 3D detection and driving
Focal Sparse Convolutional Networks for 3D Object Detection
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 plain sparse convolutions with Focals Conv | PV-RCNN · KITTI val | AP3D R40, Car moderate: 84.36 -> 85.19 | Helped |
| Replace plain sparse convolutions with fusion Focals Conv | Voxel R-CNN · KITTI test | AP3D R40, Car moderate: 81.62 -> 82.28 | Helped |
| Use Focals Conv in the backbone | CenterPoint · nuScenes test | mAP: 58.0 -> 63.8 | Helped |
| Use fusion Focals Conv | CenterPoint · nuScenes 1/4 | mAP: 59.0 -> 61.7 | Helped |
| Use dynamic output shapes rather than fixed regular dilation (ablation) | Focals Conv on PV-RCNN · KITTI val, Car moderate | AP3D R40: 84.02 -> 85.19 | Helped |
| Use importance-guided rather than random input selection (ablation) | Focals Conv on PV-RCNN · KITTI val, Car moderate | AP3D R40: 82.77 -> 85.19 | Helped |
| Add objective-loss supervision alongside attention supervision (ablation) | Focals Conv on PV-RCNN · KITTI val, Car moderate | AP3D R40: 84.49 -> 85.19 | Helped |
| Add attention multiplication alongside objective-loss supervision (ablation) | Focals Conv on PV-RCNN · KITTI val, Car moderate | AP3D R40: 85.05 -> 85.19 | Helped |
| Extend Focals Conv use from stages 1–3 to stage 4 (ablation) | PV-RCNN · KITTI val, Car moderate | AP3D R40: 85.19 -> 84.42 | Hurt |
| Vary the importance threshold from 0.1 to 0.9 (ablation) | Focals Conv on PV-RCNN · KITTI val, Car | AP3D R40: 84.97; 85.09; 85.19; 84.96; 84.68 | No clear effect |
| Fuse RGB features only to important outputs rather than all outputs (ablation) | Focals Conv-F on CenterPoint · nuScenes 1/4 | mAP: 60.9 -> 61.7 | Helped |
| Move fusion Focals Conv from stage 1 to stage 3 (ablation) | CenterPoint · nuScenes 1/4 | mAP: 61.7 -> 55.0 | Hurt |
Closest papers by meaning
- Spatial Pruned Sparse Convolution for Efficient 3D Object Detection (NeurIPS 2022 · similarity 0.92)
- FSHNet: Fully Sparse Hybrid Network for 3D Object Detection (CVPR 2025 · similarity 0.87)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Text-guided Sparse Voxel Pruning for Efficient 3D Visual Grounding (CVPR 2025 · similarity 0.86)
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · similarity 0.85)
- SparseFormer: Sparse Visual Recognition via Limited Latent Tokens (ICLR 2024 · similarity 0.85)
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.