Research map / Detection and segmentation / 3D detection and driving
Cross Modal Transformer: Towards Fast and Robust 3D Object Detection
Junjie Yan, Yingfei Liu, Jianjian Sun, Fan Jia, Shuailin Li, Tiancai Wang, Xiangyu Zhang
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 |
|---|---|---|---|
| add masked-modal training (random single-modality training) (ablation) | CMT · nuScenes val, camera-only input | NDS: 0.073 -> 0.447 | Improved |
| add masked-modal training (ablation) | CMT · nuScenes val, LiDAR-only input | NDS: 0.603 -> 0.681 | Improved |
| add masked-modal training (ablation) | CMT · nuScenes val, full multimodal input | NDS: 0.726 -> 0.729 | No clear effect |
| add point cloud position encoding to query generation (ablation) | CMT · nuScenes val | NDS: 0.595 -> 0.669 | Improved |
| add image position encoding to query generation (ablation) | CMT · nuScenes val | NDS: 0.665 -> 0.669 | Improved |
| add point-based query denoising (ablation) | CMT · nuScenes val | NDS: 0.626 -> 0.669 | Improved |
| use VoxelNet instead of PointPillars LiDAR backbone (ablation) | CMT · nuScenes val | NDS: 0.628 -> 0.669 | Improved |
| use VoV-99 instead of ResNet-50 image backbone (ablation) | CMT · nuScenes val | NDS: 0.658 -> 0.669 | Improved |
| increase image size from 800x320 to 1600x640 (ablation) | CMT · nuScenes val | NDS: 0.654 -> 0.669 | Improved |
| reduce voxel size from 0.125 to 0.075 (ablation) | CMT · nuScenes val | NDS: 0.655 -> 0.669 | Improved |
| reduce voxel size from 0.1 to 0.075 (ablation) | CMT · nuScenes val | NDS: 0.671 -> 0.669 | No clear effect |
| fuse camera tokens with LiDAR in CMT vs LiDAR-only CMT-L | CMT · nuScenes val | NDS: 0.686 -> 0.729 | Improved |
Closest papers by meaning
- Unleash the Potential of Image Branch for Cross-modal 3D Object Detection (NeurIPS 2023 · similarity 0.90)
- CAT-Det: Contrastively Augmented Transformer for Multi-Modal 3D Object Detection (CVPR 2022 · similarity 0.89)
- CCF: Complementary Collaborative Fusion for Domain Generalized Multi-Modal 3D Object Detection (CVPR 2026 · similarity 0.89)
- Bidirectional Propagation for Cross-Modal 3D Object Detection (ICLR 2023 · similarity 0.88)
- SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection (AAAI 2024 · similarity 0.87)
- CMDA: Cross-Modal and Domain Adversarial Adaptation for LiDAR-Based 3D Object Detection (AAAI 2024 · similarity 0.87)
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.
