Research map / Detection and segmentation / 3D detection and driving
SparseFusion: Fusing Multi-Modal Sparse Representations for Multi-Sensor 3D Object Detection
Yichen Xie, Chenfeng Xu, Marie-Julie Rakotosaona, Patrick Rim, Federico Tombari, Kurt Keutzer, Masayoshi Tomizuka, Wei Zhan
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 |
|---|---|---|---|
| use self-attention instead of MLP for sparse candidate fusion (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 64.5 -> 66.4 | Improved |
| use self-attention instead of cross-attention fusion (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 65.8 -> 66.4 | Improved |
| use self-attention instead of optimal transport fusion (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 65.6 -> 66.4 | Improved |
| add semantic transfer from camera to LiDAR (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 64.2 -> 66.4 | Improved |
| add geometric transfer from LiDAR to camera (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 65.7 -> 66.4 | Improved |
| add view transformation of camera candidates into LiDAR space (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 65.6 -> 66.4 | Improved |
| use parallel modality-specific detectors instead of sequential (ablation) | SparseFusion · nuScenes val (1/5 train) | mAP: 64.2 -> 66.4 | Improved |
| swap ResNet-50 for stronger Swin-T image backbone | SparseFusion · nuScenes val | NDS: 72.8 -> 73.1 | Improved |
| add semantic transfer to the LiDAR branch | TransFusion-L · nuScenes val | mAP: 65.1 -> 68.4 | Improved |
| add camera view transformation module to camera branch | camera branch · nuScenes val | mAP: 40.6 -> 41.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepInteraction — DeepInteraction: 3D Object Detection via Modality Interaction (NeurIPS 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SparseLIF: High-Performance Sparse LiDAR-Camera Fusion for 3D Object Detection (ECCV 2024 · similarity 0.93)
- DeepFusion: Lidar-Camera Deep Fusion for Multi-Modal 3D Object Detection (CVPR 2022 · similarity 0.92)
- ObjectFusion: Multi-modal 3D Object Detection with Object-Centric Fusion (ICCV 2023 · similarity 0.92)
- GAFusion: Adaptive Fusing LiDAR and Camera with Multiple Guidance for 3D Object Detection (CVPR 2024 · similarity 0.90)
- SAFDNet: A Simple and Effective Network for Fully Sparse 3D Object Detection (CVPR 2024 · similarity 0.89)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 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.
