Research map / Probabilistic and scientific modeling / Molecules and proteins
Adaptive Human Trajectory Prediction via Latent Corridors
Neerja Thakkar, Karttikeya Mangalam, Andrea Bajcsy, Jitendra Malik
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 latent corridor prompts to a frozen scene-aware predictor (LC) | YNet-Ours · MOTSynth (437 scenes) | ADE: 47.3 -> 44.6 | Improved |
| add joint finetuning of predictor head to latent corridors (ablation) | YNet-Ours · MOTSynth (437 scenes) | ADE: 44.6 -> 42.6 | Improved |
| add latent corridors on top of per-scene finetuning (ablation) | YNet-Ours · MOT (real, sim-to-real) | ADE: 41.8 -> 37.4 | Improved |
| add latent corridors on top of per-scene finetuning (ablation) | YNet-Ours · EarthCam | ADE: 21.9 -> 17.5 | Improved |
| use latent corridors only vs finetuning only on real scenes (ablation) | YNet-Ours · MOT | ADE: 41.8 (Finetune) vs 43.0 (LC) | Hurt |
| add scene awareness (segmentation input) to learned trajectory predictor | PECNet-Ours · MOTSynth (437 scenes) | ADE: 51.2 -> 47.3 | Improved |
| sum prompt to all input heatmaps instead of other locations or multiplication (ablation) | Latent corridors (LC) · MOTSynth (437 scenes) | ADE: 44.6 vs 45.1-47.6 for other variants | Improved |
| apply latent corridors with joint finetuning to PECNet architecture | PECNet-Ours · MOTSynth (437 scenes) | ADE: 51.2 -> 46.0 | Improved |
| add latent corridors to per-scene finetuning on ETH/UCY (ablation) | YNet-Ours · ETH/UCY | ADE: 22.8 -> 21.8 | Improved |
| use low-rank (rank 1) prompt instead of full-rank prompt | Latent corridors · parameter count | added parameters | No clear effect |
| adapt via latent corridors in scenes where prior predictor already suffices | YNet-Ours · MOTSynth per-scene | ADE: per-scene gains up to 63.4% | Mixed |
Closest papers by meaning
- Towards Predicting Any Human Trajectory In Context (NeurIPS 2025 · similarity 0.89)
- Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · similarity 0.88)
- Stochastic Trajectory Prediction via Motion Indeterminacy Diffusion (CVPR 2022 · similarity 0.88)
- SingularTrajectory: Universal Trajectory Predictor Using Diffusion Model (CVPR 2024 · similarity 0.87)
- SocialCVAE: Predicting Pedestrian Trajectory via Interaction Conditioned Latents (AAAI 2024 · similarity 0.87)
- MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation (CVPR 2025 · 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.
