Research map / Detection and segmentation / 3D detection and driving
TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection With Transformers
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 |
|---|---|---|---|
| Add soft-attention LiDAR-camera fusion to the LiDAR-only TransFusion-L | TransFusion-L · nuScenes test | mAP: 65.5 -> 68.9 | Helped |
| Add soft-attention LiDAR-camera fusion to the LiDAR-only TransFusion-L | TransFusion-L · Waymo validation, LEVEL_2 | mAPH | Mixed |
| Add image feature fusion (ablation) | TransFusion · nuScenes validation | mAP: 61.6 -> 65.6 | Helped |
| Add image-guided query initialization (ablation) | TransFusion · nuScenes validation | mAP: 64.8 -> 65.6 | Helped |
| Add category embeddings to object queries (ablation) | TransFusion-L · nuScenes validation, initial LiDAR predictions | NDS: 63.9 -> 66.8 | Helped |
| Use input-dependent query initialization instead of learned query positions (ablation) | TransFusion-L · nuScenes validation, initial LiDAR predictions | NDS: 33.8 -> 66.8 | Helped |
| Increase decoder layers from 1 to 3 with input-dependent query initialization (ablation) | TransFusion-L · nuScenes validation, initial LiDAR predictions | mAP: 60.0 -> 59.9 | No clear effect |
| Increase decoder layers from 1 to 3 without input-dependent query initialization (ablation) | TransFusion-L · nuScenes validation, initial LiDAR predictions | NDS: 33.8 -> 43.4 | Helped |
| Increase training from 12 to 36 epochs without input-dependent query initialization (ablation) | TransFusion-L · nuScenes validation, initial LiDAR predictions | NDS: 43.4 -> 57.8 | Helped |
| Use soft-attention fusion instead of point-wise concatenation | TransFusion-L-based detector · nuScenes validation, nighttime | mAP: 49.2 -> 55.2 | Helped |
| Use soft-attention fusion instead of hard-associated fusion | TransFusion-L-based detector · nuScenes validation, 6 dropped images | mAP: 39.5 -> 61.7; 47.0 -> 61.7 | Helped |
| Use soft-attention fusion under a 1m sensor translation error | TransFusion-L-based detector · nuScenes validation | mAP drop: 0.49% versus 2.33% and 2.85% | Helped |
Closest papers by meaning
- DeepFusion: Lidar-Camera Deep Fusion for Multi-Modal 3D Object Detection (CVPR 2022 · similarity 0.92)
- BEVFusion: A Simple and Robust LiDAR-Camera Fusion Framework (NeurIPS 2022 · similarity 0.91)
- GAFusion: Adaptive Fusing LiDAR and Camera with Multiple Guidance for 3D Object Detection (CVPR 2024 · similarity 0.89)
- Adaptive-Smooth LiDAR-Camera Knowledge Distillation with Heterogeneous Fusion for Multi-View 3D Object Detection (AAAI 2026 · similarity 0.88)
- Resilient Sensor Fusion Under Adverse Sensor Failures via Multi-Modal Expert Fusion (CVPR 2025 · similarity 0.87)
- LiDAR-PTQ: Post-Training Quantization for Point Cloud 3D Object Detection (ICLR 2024 · similarity 0.86)
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.