Research map / Probabilistic and scientific modeling / Molecules and proteins
EigenTrajectory: Low-Rank Descriptors for Multi-Modal Trajectory Forecasting
Inhwan Bae, Jean Oh, Hae-Gon Jeon
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 |
|---|---|---|---|
| transform both observed and predicted trajectories into ET space instead of Euclidean (ablation) | SGCN · ETH/UCY average | FDE: 0.63 -> 0.49 | Improved |
| replace only the input with ET space, keep output Euclidean (ablation) | SGCN · ETH/UCY average | FDE: 0.63 -> 0.53 | Improved |
| add data-driven trajectory anchor refinement in ET space (ablation) | SGCN with ET input/output · ETH/UCY average | FDE: 0.49 -> 0.36 | Improved |
| cluster trajectory anchors in ET space instead of Euclidean space (ablation) | SGCN · ETH/UCY average | FDE: 0.58 -> 0.38 | Improved |
| use ET descriptor (k=12) instead of Bezier curve for trajectory approximation (ablation) | trajectory descriptor · ETH/UCY average, prediction reconstruction error | L2 error (mm): 25 -> 14 (Bezier -> ET k=12) | Improved |
| apply ET space to SGCN on non-linear trajectories (ablation) | SGCN · ETH/UCY non-linear trajectories | FDE drop vs all trajectories: -0.32 -> -0.13 | Improved |
| add Gaussian input noise (std 0.10) with ET space (ablation) | SGCN · ETH/UCY average | FDE drop: -0.42 -> -0.13 | Improved |
| apply ET space to AgentFormer | AgentFormer · ZARA1 | FDE gain: -31.1% | Hurt |
| apply ET low-rank space to baselines | SGCN · ETH/UCY average | collision rate (COL): 2.816 -> 3.663 | Hurt |
| apply ET space to STGCNN, SGCN, PECNet | baselines · ETH/UCY average | FDE gain: 49.2% (STGCNN), 42.1% (SGCN), 35.3% (PECNet) | Improved |
| apply ET space to AgentFormer | AgentFormer · ETH/UCY average | FDE gain: 5.3% | Mixed |
| use separate learned descriptor instead of shared; learned SVD basis vs linear interpolation (ablation) | trajectory descriptor · ETH/UCY average | L2 error (mm, prediction): 384 -> 14 (linear interp -> ET k=12) | Improved |
Closest papers by meaning
- SocialCVAE: Predicting Pedestrian Trajectory via Interaction Conditioned Latents (AAAI 2024 · similarity 0.89)
- SocialMOIF: Multi-Order Intention Fusion for Pedestrian Trajectory Prediction (CVPR 2025 · similarity 0.88)
- Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · similarity 0.87)
- Higher-order Relational Reasoning for Pedestrian Trajectory Prediction (CVPR 2024 · similarity 0.87)
- SingularTrajectory: Universal Trajectory Predictor Using Diffusion Model (CVPR 2024 · similarity 0.87)
- Multi-modal Knowledge Distillation-based Human Trajectory Forecasting (CVPR 2025 · similarity 0.87)
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.
