Research map / Detection and segmentation / 3D detection and driving
Unifying Voxel-based Representation with Transformer for 3D Object Detection
Read the paperCheck 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 |
|---|---|---|---|
| Preserve voxel height instead of collapsing to BEV (ablation) | UVTR camera branch · nuScenes val, camera | NDS: 31.4 -> 35.6 | Helped |
| Use Conv3D instead of Conv2D for voxel spatial interaction (ablation) | UVTR camera branch · nuScenes val, camera | NDS: 31.9 -> 34.5 | Helped |
| Add voxel encoder spatial interaction (ablation) | UVTR camera branch · nuScenes val, camera | NDS: 12.0 -> 34.5 | Helped |
| Increase camera input from 1 to 5 sweeps (ablation) | UVTR-CS · nuScenes val, camera | NDS: 34.5 -> 39.5 | Helped |
| Use ResNet-101 instead of ResNet-50 for voxel construction (ablation) | UVTR-C · nuScenes val, camera | NDS: 34.5 -> 39.4 | Helped |
| Add LiDAR teacher knowledge transfer (ablation) | Camera student UVTR-L2C · nuScenes val | NDS: 34.5 -> 36.4 | Helped |
| Use multi-modality teacher rather than no teacher (ablation) | Camera student UVTR · nuScenes val | NDS: 34.5 -> 37.1 | Helped |
| Use partial L2 instead of L2 for knowledge transfer (ablation) | UVTR-L2C · nuScenes val, camera student | NDS: 36.0 -> 36.4 | Helped |
| Replace transformer decoder with convolutional head (ablation) | UVTR-L · nuScenes val, LiDAR | NDS: 66.4 -> 65.4 | Hurt |
| Replace transformer decoder with convolutional head (ablation) | UVTR-C · nuScenes val, camera | NDS: 44.1 -> 40.0 | Hurt |
| Fuse camera features with LiDAR features (ablation) | UVTR-M · nuScenes val, LiDAR and camera fusion | NDS: 63.8 -> 66.3 | Helped |
| Increase voxel height for LiDAR from 1 to 11 (ablation) | UVTR-L · nuScenes val, LiDAR | NDS: 62.8 -> 63.8 | Helped |
Closest papers by meaning
- Voxel Field Fusion for 3D Object Detection (CVPR 2022 · similarity 0.90)
- HVPR: Hybrid Voxel-Point Representation for Single-Stage 3D Object Detection (CVPR 2021 · similarity 0.89)
- DeepInteraction: 3D Object Detection via Modality Interaction (NeurIPS 2022 · similarity 0.88)
- Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection (NeurIPS 2024 · similarity 0.86)
- Referred by Multi-Modality: A Unified Temporal Transformer for Video Object Segmentation (AAAI 2024 · similarity 0.86)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 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.