Research map / Detection and segmentation / 3D detection and driving
BoxeR: Box-Attention for 2D and 3D 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 |
|---|---|---|---|
| Use box-attention instead of self-attention, deformable attention, or dynamic attention | end-to-end object detection · COCO val | AP: 36.2 -> 48.7; 46.9 -> 48.7; 47.2 -> 48.7 | Helped |
| Add translation and scaling transformations to the reference window (ablation) | Box-Attention · COCO val | AP: 46.4 -> 48.7 | Helped |
| Avoid adding proposal refinement in decoder layers (ablation) | BoxeR-2D · COCO val | AP: 47.2 -> 48.7 | Helped |
| Add Instance-Attention for mask prediction (ablation) | BoxeR-2D · COCO val | AP: 48.7 -> 50.0 | Helped |
| Add rotation transformation to box-attention (ablation) | BoxeR-3D · Waymo Open val LEVEL_1 | vehicle AP: 69.4 -> 70.4 | Helped |
| Use multi-angle rather than single-angle reference windows (ablation) | BoxeR-3D · Waymo Open val LEVEL_1 | vehicle AP: 70.0 -> 70.4 | Helped |
| Use multi-angle rather than single-angle reference windows (ablation) | BoxeR-3D · Waymo Open val LEVEL_1 | pedestrian AP: 64.7 -> 64.7 | No clear effect |
| Remove dropout during training (ablation) | BoxeR-2D · COCO val | AP: 47.8 -> 48.7 | Helped |
| Use BoxeR-2D rather than Deformable DETR | COCO 2017 test object detection · COCO 2017 test | AP: 48.7 -> 51.1 | Helped |
| Use BoxeR-2D rather than QueryInst | COCO 2017 test instance segmentation · COCO 2017 test | mask AP: 42.8 -> 43.8 | Helped |
| Use BoxeR-3D rather than Deformable DETR | 3D object detection · Waymo Open val LEVEL_2 | pedestrian AP: 45.8 -> 61.5 | Helped |
Closest papers by meaning
- PADRe: A Unifying Polynomial Attention Drop-in Replacement for Efficient Vision Transformer (ICLR 2025 · similarity 0.86)
- Visual Transformer with Differentiable Channel Selection: An Information Bottleneck Inspired Approach (ICML 2024 · similarity 0.84)
- CAT: Circular-Convolutional Attention for Sub-Quadratic Transformers (NeurIPS 2025 · similarity 0.84)
- Union-over-Intersections: Object Detection beyond Winner-Takes-All (ICLR 2025 · similarity 0.84)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.84)
- Scaling White-Box Transformers for Vision (NeurIPS 2024 · similarity 0.84)
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.