Research map / Detection and segmentation / 3D detection and driving
DeTra: A Unified Model for Object Detection and Trajectory Forecasting
Sergio Casas, Ben T Agro, Jiageng Mao, Thomas Gilles, ALEXANDER Y CUI, Enxu Li, Raquel Urtasun
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 |
|---|---|---|---|
| formulate detection and forecasting as one unified trajectory refinement transformer (DeTra) instead of cascaded or two-stage E2E models | DeTra · Argoverse 2 Sensor | TrajAP: 44.1 -> 50.0 | Improved |
| unified refinement transformer vs best baseline on Waymo Open | DeTra · Waymo Open | OccAP: 64.9 -> 70.4 | Improved |
| add iterative pose refinement over 3 blocks (ablation) | DeTra · Argoverse 2 Sensor | OccAP: 42.7 -> 73.0 | Improved |
| feed refined poses to each block instead of the initial poses (ablation) | DeTra · Argoverse 2 Sensor | TrajAP: 44.6 -> 50.0 | Improved |
| third refinement block versus second (ablation) | DeTra · Argoverse 2 Sensor | OccAP: 72.8 -> 73.0 | No clear effect |
| initialize poses from a lightweight detector head instead of a fixed grid (ablation) | DeTra · Argoverse 2 Sensor | OccAP: 70.8 -> 73.0 | Improved |
| add learned offsets on top of grid pose initialization (ablation) | DeTra grid init · Argoverse 2 Sensor | OccAP: 70.8 -> 68.5 | Hurt |
| add deformable LiDAR cross-attention (ablation) | DeTra · Argoverse 2 Sensor | OccAP: 42.5 -> 73.0 | Improved |
| use deformable LiDAR attention instead of global attention (ablation) | DeTra · Argoverse 2 Sensor | OccAP: 48.1 -> 73.0 | Improved |
| add k-nearest map attention (ablation) | DeTra · Argoverse 2 Sensor | MR K=6: 20.2 -> 15.4 | Improved |
| use k-nearest map attention instead of global map attention (ablation) | DeTra · Argoverse 2 Sensor | MR K=6: 20.3 -> 15.4 | Improved |
| add time self-attention (ablation) | DeTra · Argoverse 2 Sensor | OccAP: 62.4 -> 73.0 | Improved |
Closest papers by meaning
- Forecasting From LiDAR via Future Object Detection (CVPR 2022 · similarity 0.90)
- Argoverse 2: Next Generation Datasets for Self-Driving Perception and Forecasting (NeurIPS 2021 · similarity 0.88)
- AS-Det: Active Sampling for Adaptive 3D Object Detection in Point Clouds (AAAI 2025 · similarity 0.87)
- UnO: Unsupervised Occupancy Fields for Perception and Forecasting (CVPR 2024 · similarity 0.86)
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · similarity 0.86)
- DiffuBox: Refining 3D Object Detection with Point Diffusion (NeurIPS 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.
