Research map / Probabilistic and scientific modeling / Variational and generative sampling
Fast Inference and Update of Probabilistic Density Estimation on Trajectory Prediction
Takahiro Maeda, Norimichi Ukita
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 |
|---|---|---|---|
| reuse flow transformations and update with newest observed position instead of re-predicting from scratch (rapid update) (ablation) | FlowChain · ETH/UCY log-probability t+6..t+12 | log-probability: -0.26 -> 0.31 | Improved |
| update density from t-1 instead of predicting at t | FlowChain · Simfork | EMD: 1.408 -> 2.539 | Hurt |
| update density instead of full prediction (compute time) | FlowChain · ETH/UCY | computation time (ms): 37.0 -> 0.6 | Improved |
| use continuously-indexed flows instead of vanilla conditional flows (ablation) | FlowChain · ETH/UCY ADE | ADE (m): 0.35 -> 0.29 | Improved |
| use continuously-indexed flows instead of vanilla conditional flows (ablation) | FlowChain · SDD ADE | ADE (pixels): 9.70 -> 9.93 | No clear effect |
| use continuously-indexed flows for density estimation (ablation) | FlowChain · Simfork | EMD: 3.313 -> 1.408 | Improved |
| use RealNVP instead of MAF (ablation) | FlowChain · SDD ADE | ADE (pixels): 14.58 -> 9.93 | Improved |
| use Trajectron++ social encoder instead of a per-agent transformer layer (ablation) | FlowChain · SDD ADE | ADE (pixels): 18.92 -> 9.93 | Improved |
| analytic normalizing-flow density instead of Gaussian mixture (Trajectron++) | trajectory density estimation · ETH/UCY | log-probability: -281.86 -> -0.26 | Improved |
| analytic flow density instead of KDE on sampled trajectories (MID) | trajectory density estimation · ETH/UCY | computation time (ms): 23904.1 -> 37.0 | Improved |
Closest papers by meaning
- BCDiff: Bidirectional Consistent Diffusion for Instantaneous Trajectory Prediction (NeurIPS 2023 · similarity 0.85)
- Den-TP: A Density-Balanced Data Curation and Evaluation Framework for Trajectory Prediction (CVPR 2026 · similarity 0.85)
- MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation (CVPR 2025 · similarity 0.85)
- Likelihood-Based Diverse Sampling for Trajectory Forecasting (ICCV 2021 · similarity 0.85)
- MGF: Mixed Gaussian Flow for Diverse Trajectory Prediction (NeurIPS 2024 · similarity 0.84)
- Stop the Sampler! Classifier-Based Adaptive Stopping for Sampling Kernels (NeurIPS 2026 · similarity 0.83)
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.
