Research map / Detection and segmentation / 3D detection and driving
EVT: Efficient View Transformation for Multi-Modal 3D Object Detection
Yongjin Lee, Hyeon-Mun Jeong, Yurim Jeon, Sanghyun Kim
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 camera features via vanilla view transformation to LiDAR-only (ablation) | EVT-L · nuScenes val (ResNet, 704x256) | NDS: 71.7 -> 72.7 | Improved |
| add adaptive sampling (AS) to vanilla VT (ablation) | EVT · nuScenes val (ResNet, 704x256) | NDS and mAP: improvements of 0.8% NDS and 1.5% mAP | Improved |
| add adaptive projection (AP) on top of AS (ablation) | EVT · nuScenes val (ResNet, 704x256) | NDS: 73.5 -> 74.1 | Improved |
| full ASAP versus vanilla VT (ablation) | EVT · nuScenes val (ResNet, 704x256) | NDS and mAP: 1.4% NDS and 2.0% mAP with only 3ms latency increase | Improved |
| use heatmap init instead of learnable query init (ablation) | EVT-L · nuScenes val (6 decoder layers) | NDS: 69.6 -> 70.7 | Improved |
| add mixed query initialization over heatmap init (ablation) | EVT-L · nuScenes val (6 decoder layers) | NDS: 70.7 -> 71.1 | Improved |
| add group-wise embedding to mixed init (ablation) | EVT-L · nuScenes val (6 decoder layers) | NDS: 71.1 -> 71.7 | Improved |
| replace standard attention with corner-aware sampling (ablation) | EVT-L · nuScenes val | NDS: 69.6 -> 71.3 | Improved |
| add AdaMixer feature mixing to corner-aware sampling (ablation) | EVT-L · nuScenes val | NDS: 71.3 -> 71.2 | No clear effect |
| add position-aware feature mixing to corner-aware sampling (ablation) | EVT-L · nuScenes val | NDS and mAP: 0.4% NDS and 0.7% mAP | Improved |
| add geometry-aware cross-attention to StreamPETR (24 epochs) (ablation) | StreamPETR · nuScenes val camera-only | NDS: 47.8 -> 49.1 | Improved |
| add geometry-aware cross-attention to StreamPETR (90 epochs) (ablation) | StreamPETR · nuScenes val camera-only | NDS: 53.5 -> 54.7 | 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.
- UniTR — UniTR: A Unified and Efficient Multi-Modal Transformer for Bird's-Eye-View Representation (ICCV 2023) · better on 4 of 4 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
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.89)
- RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection (CVPR 2024 · similarity 0.88)
- BEVNeXt: Reviving Dense BEV Frameworks for 3D Object Detection (CVPR 2024 · similarity 0.88)
- RobuRCDet: Enhancing Robustness of Radar-Camera Fusion in Bird's Eye View for 3D Object Detection (ICLR 2025 · similarity 0.87)
- IS-Fusion: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection (CVPR 2024 · similarity 0.87)
- RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion (CVPR 2025 · 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.
