Research map / Probabilistic and scientific modeling / Molecules and proteins
Motion Transformer with Global Intention Localization and Local Movement Refinement
Shaoshuai Shi, Li Jiang, Dengxin Dai, Bernt Schiele
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 latent learnable embedding queries with static intention queries from k-means points (ablation) | MTR decoder · WOMD marginal val (20% train) | mAP: 0.2633 -> 0.3059 | Improved |
| add iterative refinement with stacked decoders (ablation) | MTR decoder · WOMD marginal val (20% train) | mAP: 0.3059 -> 0.3171 | Improved |
| add local movement refinement (dynamic searching query) (ablation) | MTR decoder · WOMD marginal val (20% train) | mAP: 0.3171 -> 0.3234 | Improved |
| add dense future prediction (ablation) | MTR · WOMD marginal val (20% train) | mAP: 0.3234 -> 0.3437 | Improved |
| use local self-attention instead of global attention in encoder (ablation) | MTR encoder · WOMD marginal val | mAP: +0.77% (256 polylines), +1.12% (512 polylines) | Improved |
| increase neighbors in local self-attention from 16 to 64 (ablation) | MTR encoder · WOMD marginal val (20% train) | mAP: 34.37% -> 33.67% | Hurt |
| collect 256 instead of 128 map polylines in dynamic map collection (ablation) | MTR decoder · WOMD marginal val (20% train) | mAP: -0.41% mAP | Hurt |
| increase number of decoder layers (ablation) | MTR decoder · WOMD marginal val (20% train) | mAP | Improved |
| replace k-means intention points with uniform 8x8 sampling (ablation) | MTR · WOMD marginal val (20% train) | mAP | Hurt |
| remove NMS using 6 queries (MTR-e2e vs MTR) | MTR · WOMD marginal val | minADE / mAP: minADE 0.6046 -> 0.5160; mAP 0.4164 -> 0.3245 | Mixed |
| use intention-point assignment (alpha) vs predicted-trajectory assignment (beta) as queries grow (ablation) | MTR · WOMD marginal val (20% train) | mAP | Improved |
| ensemble MTR and MTR-e2e predictions | MTR · WOMD marginal val | mAP: 0.4164 -> 0.4323 (MTR-ens) | Improved |
Closest papers by meaning
- MTR++: Multi-Agent Motion Prediction With Symmetric Scene Modeling and Guided Intention Querying (TPAMI 2024 · similarity 0.94)
- Future-Aware Interaction Network For Motion Forecasting (ICCV 2025 · similarity 0.89)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.87)
- UniMotion: A Unified Motion Framework for Simulation, Prediction and Planning (NeurIPS 2025 · similarity 0.86)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.86)
- SmartRefine: A Scenario-Adaptive Refinement Framework for Efficient Motion Prediction (CVPR 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.
