Research map / Label-efficient and robust learning / Recognition and long-tail
DenseTNT: End-to-End Trajectory Prediction From Dense Goal Sets
Junru Gu, Chen Sun, Hang Zhao
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 |
|---|---|---|---|
| use dense goal probability estimation instead of sparse anchors (with NMS) (ablation) | DenseTNT · Argoverse val | Miss Rate: 10.1% -> 8.6% | Improved |
| add offline optimization pseudo-labels instead of NMS selection (ablation) | DenseTNT · Argoverse val | Miss Rate: 8.2% -> 7.0% | Improved |
| add offline optimization pseudo-labels instead of NMS selection (ablation) | DenseTNT · Argoverse val | minFDE: 1.28 -> 1.27 | Improved |
| make the model end-to-end with a goal set predictor (ablation) | DenseTNT · Argoverse val | Miss Rate: 7.0% -> 7.0% | No clear effect |
| replace variety loss with dense probability and optimization (ablation) | TNT-style sparse · Argoverse val | Miss Rate: 13.3% -> 7.0% | Improved |
| optimize minFDE objective instead of MR (ablation) | DenseTNT · Argoverse val | minFDE: 1.27 -> 1.05 | Improved |
| optimize minFDE objective instead of MR (ablation) | DenseTNT · Argoverse val | Miss Rate: 7.0% -> 9.8% | Hurt |
| reduce sampling density from 1.0m to 3.0m (ablation) | DenseTNT · Argoverse val | Miss Rate: 7.0% -> 12.5% | Hurt |
| refine sampling density from 1.0m to 0.5m (ablation) | DenseTNT · Argoverse val | Miss Rate: 7.0% -> 7.0% | No clear effect |
| increase optimization time from 20ms to 100ms (ablation) | DenseTNT · Argoverse val | Miss Rate: 7.6% -> 7.0% | Improved |
| increase optimization time from 100ms to 500ms (ablation) | DenseTNT · Argoverse val | Miss Rate: 7.0% -> 6.9% | No clear effect |
| mix minFDE and MR objectives 70%/30% instead of pure MR (ablation) | DenseTNT · Argoverse val | minFDE: 1.27 -> 1.08 | Improved |
Closest papers by meaning
- HalentNet: Multimodal Trajectory Forecasting with Hallucinative Intents (ICLR 2021 · similarity 0.88)
- TPNet: Trajectory Proposal Network for Motion Prediction (CVPR 2020 · similarity 0.87)
- HPNet: Dynamic Trajectory Forecasting with Historical Prediction Attention (CVPR 2024 · similarity 0.87)
- SparseDriveV2: Scoring is All You Need for End-to-End Autonomous Driving (ECCV 2026 · similarity 0.86)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.86)
- SmartRefine: A Scenario-Adaptive Refinement Framework for Efficient Motion Prediction (CVPR 2024 · similarity 0.85)
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.
