Research map / Probabilistic and scientific modeling / Molecules and proteins
Recover to Predict: Progressive Retrospective Learning for Variable-Length Trajectory Prediction
Hao Zhou, Lu Qi, Xiangtai Li, Jie Zhang, Yi Liu, Xu Yang, Mingyu Fan, Fei Luo
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 |
|---|---|---|---|
| add RDM (retrospective distillation) (ablation) | DeMo · Argoverse 2 val obs 10 | mADE6: 0.876 -> 0.655 | Improved |
| add RPM (retrospective prediction) (ablation) | DeMo + RDM · Argoverse 2 val obs 10 | mFDE6: 1.257 -> 1.241 | Improved |
| add Rolling-Start Training Strategy (ablation) | DeMo + RDM + RPM · Argoverse 2 val obs 10 | mADE6: 0.652 -> 0.617 | Improved |
| add extra rolling windows [0,40],[0,30],[0,20] (ablation) | PRF · Argoverse 2 val obs 10 | mADE6: 0.652 -> 0.617 | Improved |
| progressive distillation instead of direct one-shot distillation (ablation) | PRF · Argoverse 2 val obs 10 | mADE6: 0.663 -> 0.652 | Improved |
| Mamba instead of GRU for state queries in RPM (ablation) | PRF · Argoverse 2 val obs 10 | mADE6: 0.662 -> 0.652 | Improved |
| increase RDM attention layers from 1 to 3 (ablation) | PRF · Argoverse 2 val obs 10 | mFDE6: 1.280 -> 1.257 | Improved |
| increase RPM layers from 1 to 3 (ablation) | PRF · Argoverse 2 val obs 10 | mADE6: 0.647 -> 0.652 (mADE6), 1.252 -> 1.241 (mFDE6) | Mixed |
| add PRF to DeMo versus training separate models per length (IT) | DeMo · Argoverse 2 val obs 10 | mADE6: 0.675 -> 0.617 | Improved |
| PRF with full-length observation on Argoverse 1 leaderboard | DeMo-PRF · Argoverse 1 test | b-mFDE6: 1.67 (Tamba) vs 1.73 | Hurt |
| PRF inference with shorter observations | PRF · Argoverse 2 val | Inference time (s): 0.140 -> 0.268 | Hurt |
Closest papers by meaning
- Towards Generalizable Trajectory Prediction using Dual-Level Representation Learning and Adaptive Prompting (CVPR 2025 · similarity 0.86)
- LaKD: Length-agnostic Knowledge Distillation for Trajectory Prediction with Any Length Observations (NeurIPS 2024 · similarity 0.86)
- PLMTrajRec: A Scalable and Generalizable Trajectory Recovery Method with Pre-trained Language Models (NeurIPS 2025 · similarity 0.85)
- C2F-TP: A Coarse-to-Fine Denoising Framework for Uncertainty-Aware Trajectory Prediction (AAAI 2025 · similarity 0.85)
- CaDeT: a Causal Disentanglement Approach for Robust Trajectory Prediction in Autonomous Driving (CVPR 2024 · similarity 0.85)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · 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.
