Research map / Probabilistic and scientific modeling / Molecules and proteins
Unveiling Transferability in Trajectory Prediction via Latent Scene Embeddings
Theodor Westny, David Axelsson, Björn Olofsson, Erik Frisk
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 |
|---|---|---|---|
| use KL divergence instead of L1, Wasserstein or MMD as dissimilarity (ablation) | latent dataset embeddings · zero-shot transfer | Spearman rho: 0.482 (best alt.) -> 0.811 | Improved |
| use KL divergence instead of mean speed difference (ablation) | latent dataset embeddings · zero-shot transfer | Spearman rho: 0.624 -> 0.811 | Improved |
| increase latent dim from 32 to 128 (ablation) | embedding model · zero-shot transfer | Spearman rho: 0.811 -> 0.736 | Hurt |
| increase latent dim from 32 to 128 (ablation) | embedding model · post-fine-tuning degradation | Spearman rho: 0.729 -> 0.874 | Improved |
| use KL-based source selection instead of random | source selection · zero-shot transfer | top-3 rate: 13.2% -> 83.3% | Improved |
| use KL-based selection instead of largest-source selection | source selection · zero-shot transfer | mean absolute minADE6 gap: 0.984 -> 0.100 | Improved |
| pretrain on lower-KL source then fine-tune on Argoverse | QCNet · Argoverse | minADE6 after fine-tuning | Improved |
| larger KL divergence between datasets | QCNet fine-tuning · forgetting on source dataset | minADE6 change: rho=0.729 | Hurt |
| regularize covariance with alpha 1e-4 instead of 1e-7 (ablation) | covariance estimator · L=32 | Spearman rho: 0.817 -> 0.811 | No clear effect |
Closest papers by meaning
- Improving Transferability for Cross-Domain Trajectory Prediction via Neural Stochastic Differential Equation (AAAI 2024 · similarity 0.88)
- SingularTrajectory: Universal Trajectory Predictor Using Diffusion Model (CVPR 2024 · similarity 0.86)
- GigaTraj: Predicting Long-term Trajectories of Hundreds of Pedestrians in Gigapixel Complex Scenes (CVPR 2024 · similarity 0.86)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.86)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.85)
- Learned Trajectory Embedding for Subspace Clustering (CVPR 2024 · similarity 0.85)
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.
