Research map / Probabilistic and scientific modeling / Molecules and proteins
SKETCH: Semantic Key-Point Conditioning for Long-Horizon Vessel Trajectory Prediction
Linyong Gan, Zimo Li, Wenxin XU, Li Xingjian, Jianhua Z. Huang, Enmei Tu, Shuhang Chen
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 |
|---|---|---|---|
| condition trajectory prediction on predicted Next Key Point (SKETCH) (ablation) | MiniMind decoder-only trajectory model · Private AIS dataset | MFD: 1.86 -> 0.63 (pure 4ch -> SFT-O) | Improved |
| feed a wrong NKP instead of no NKP (ablation) | SKETCH trajectory predictor · Private AIS dataset | MFD: 1.86 -> 2.09 | Hurt |
| use open-set NKP prediction (SFT-O) instead of closed-set SFT-C (ablation) | NKP predictor · Private AIS dataset | MFD: 0.69 -> 0.63 | Improved |
| inject NKP into temporal model without decoupled contrastive representation (pretrain-C) (ablation) | NKP predictor · Private AIS dataset | MFD: 0.63 -> 4.24 (SFT-O -> Pretrain-C) | Hurt |
| use predicted NKP versus oracle NKP (ablation) | SKETCH · Private AIS dataset | MFD: 0.61 -> 0.63 | No clear effect |
| use SKETCH instead of strongest baseline on public AIS data (no fine-tuning) | SKETCH · Public AIS dataset (out-of-domain) | MFD: 2.77 -> 0.97 | Improved |
| use continuous linear input projection instead of token embeddings | trajectory model vs TrAISformer · Public AIS dataset (out-of-domain) | MFD: 49.50 -> 0.97 | Improved |
| closed-set pretrain-c versus open-set sft-o-s2 NKP classifier (ablation) | NKP predictor · NKP prediction accuracy | accuracy (%): 98.98% (pretrain-c) vs 95.46% (sft-o-s2) | Mixed |
Closest papers by meaning
- Perceiving the Near, Reasoning the Distant: Coherent Long-Horizon Trajectory Prediction for Autonomous Driving (CVPR 2026 · similarity 0.88)
- Long-Horizon Visual Planning with Goal-Conditioned Hierarchical Predictors (NeurIPS 2020 · similarity 0.86)
- Drift-Resistant Navigation World Model with Anchored Epipolar Guidance (NeurIPS 2026 · similarity 0.84)
- TrajMamba: An Efficient and Semantic-rich Vehicle Trajectory Pre-training Model (NeurIPS 2025 · similarity 0.84)
- Holistic Semantic Representation for Navigational Trajectory Generation (AAAI 2025 · similarity 0.84)
- Trajectory Diffusion for ObjectGoal Navigation (NeurIPS 2024 · similarity 0.84)
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.
