Research map / Detection and segmentation / 3D detection and driving
An End-to-End Transformer Model for 3D Object Detection
Ishan Misra, Rohit Girdhar, Armand Joulin
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 |
|---|---|---|---|
| replace parametric queries with non-parametric queries (and Sine embeddings) (ablation) | 3DETR · ScanNetV2 val | AP25: 15.4 -> 55.8 | Improved |
| replace sinusoidal with Fourier positional embeddings (ablation) | 3DETR · ScanNetV2 val | AP25: 55.8 -> 61.8 | Improved |
| add Fourier positional embedding to the encoder (ablation) | 3DETR · ScanNetV2 val | AP25: 62.7 -> 61.8 | Hurt |
| use set loss instead of Box loss (ablation) | 3DETR · ScanNetV2 val | AP25: 49.6 -> 62.7 | Improved |
| use set loss instead of Vote loss (ablation) | 3DETR · ScanNetV2 val | AP25: 54.0 -> 62.7 | Improved |
| use Transformer decoder and set loss instead of Box decoder and loss (ablation) | 3DETR · ScanNetV2 val | AP25: 31.0 -> 62.7 | Improved |
| use Transformer decoder and set loss instead of Vote decoder and loss (ablation) | 3DETR · ScanNetV2 val | AP25: 46.1 -> 62.7 | Improved |
| use Transformer encoder instead of PointNet++ encoder (ablation) | 3DETR · ScanNetV2 val | AP25: 61.4 -> 62.7 | No clear effect |
| add masked self-attention encoder (3DETR-m) | 3DETR · ScanNetV2 val | AP25: 62.7 -> 65.0 | Improved |
| add masked self-attention encoder (3DETR-m) | 3DETR · SUN RGB-D val | AP50: 30.3 -> 32.7 | Mixed |
| add NMS at test time (ablation) | VoteNet · ScanNetV2 val | AP25: 10.7 -> 60.4 | Improved |
| add NMS at test time (ablation) | 3DETR · ScanNetV2 val | AP25: 59.5 -> 62.7 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- H3DNet — H3DNet: 3D Object Detection Using Hybrid Geometric Primitives (ECCV 2020) · better on 0 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
- Uni3DETR: Unified 3D Detection Transformer (NeurIPS 2023 · similarity 0.91)
- 3D Object Detection With Pointformer (CVPR 2021 · similarity 0.90)
- Bridged Transformer for Vision and Point Cloud 3D Object Detection (CVPR 2022 · similarity 0.90)
- Improving 3D Object Detection With Channel-Wise Transformer (ICCV 2021 · similarity 0.90)
- CT3D++: Improving 3D Object Detection with Keypoint-Induced Channel-wise Transformer (IJCV 2025 · similarity 0.90)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.88)
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.
