Research map / Label-efficient and robust learning / Recognition and long-tail
End-to-End Trajectory Distribution Prediction Based on Occupancy Grid Maps
Ke Guo, Wenxi Liu, Jia Pan
Check 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 the OGM reverse-cross-entropy term by increasing β from 0 to 10 (ablation) | Ours trajectory distribution model · TrajNet split | Offroad Rate: 0.035 -> 0.020 | Improved |
| add the OGM reverse-cross-entropy term by increasing β from 0 to 10 (ablation) | Ours trajectory distribution model · TrajNet split | RF20: 7.52 -> 2.13 | Hurt |
| use a ConvLSTM with deconvolution to predict OGMs (ablation) | OGM decoder · TrajNet split, OGM prediction | OGM NLL: 10.40 -> 10.31 | Improved |
| use non-stationary rewards and value iteration instead of behavior cloning (ablation) | Ours trajectory distribution model · TrajNet split | Reverse cross-entropy: 8.10 -> 7.536 | Improved |
| use non-stationary rewards instead of stationary rewards (ablation) | Ours trajectory distribution model · TrajNet split | Reverse cross-entropy: 9.52 -> 7.536 | Improved |
| add the Transformer refinement network (ablation) | Ours representative trajectory model · TrajNet split | minADE20: 8.78 -> 6.77 | Improved |
| replace K-means with the Transformer refinement network (ablation) | Ours representative trajectory model · TrajNet split | minADE20: 7.64 -> 6.77 | Improved |
| add end-to-end fine-tuning (ablation) | Ours representative trajectory model · TrajNet split | minADE20: 7.36 -> 6.77 | Improved |
| use the staged training process with end-to-end fine-tuning instead of multi-task training (ablation) | Ours representative trajectory model · TrajNet split | minADE20: 6.94 -> 6.77 | Improved |
| use the staged training process rather than variety loss alone (ablation) | Ours representative trajectory model · TrajNet split | minADE20: 8.16 -> 6.77 | Improved |
| add the refinement network (ablation) | Ours representative trajectory model · TrajNet split | Offroad Rate: 0.045 -> 0.066 | Hurt |
Closest papers by meaning
- R2P2: A ReparameteRized Pushforward Policy for Diverse, Precise Generative Path Forecasting (ECCV 2018 · similarity 0.87)
- Trajectron++: Dynamically-Feasible Trajectory Forecasting With Heterogeneous Data (ECCV 2020 · similarity 0.87)
- GoIRL: Graph-Oriented Inverse Reinforcement Learning for Multimodal Trajectory Prediction (ICML 2025 · similarity 0.87)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.86)
- Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling (CVPR 2025 · similarity 0.85)
- STraj: Self-training for Bridging the Cross-Geography Gap in Trajectory Prediction (AAAI 2025 · 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.
