Research map / Video / Video segmentation and tracking
Predicting Scene Parsing and Motion Dynamics in the Future
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly predict parsing and flow, using flow features to support parsing | Ours · Cityscapes val, 1-step | mIoU: 62.6 -> 66.1 | Helped |
| add parsing-derived semantic information and object-group flow branches | Ours · Cityscapes val, 1-step | EPE: 2.71 -> 2.30 | Helped |
| add the transform layer to combine flow features with parsing features (ablation) | Ours · Cityscapes val, 1-step | mIoU: 64.7 -> 66.1; improves the mIoU by 1.4 | Helped |
| add the transform layer to combine flow features with parsing features (ablation) | Ours · Cityscapes val, 1-step | EPE: 2.42 -> 2.30; reduces EPE by 0.12 | Helped |
| add recurrent fine-tuning for multi-step prediction (ablation) | Ours · Cityscapes val, 10-step | mIoU: 52.6 -> 53.9; improves the mIoU by 1.3% | Helped |
| add recurrent fine-tuning for multi-step prediction (ablation) | Ours · Cityscapes val, 10-step | EPE: 6.63 -> 6.31; reduces the EPE by 0.32 | Helped |
| use cross-entropy instead of L1 plus gradient difference loss on automatically generated parsing targets (ablation) | S2S · Cityscapes val, 1-step | mIoU: reduces the mIoU by 1.5 from 66.1 | Hurt |
| use preceding segmentation maps rather than preceding frames or frames plus segmentation maps as parsing-network input (ablation) | Parsing anticipating network · Cityscapes | parsing prediction performance | Helped |
| add the flow anticipating network to support parsing prediction | Ours · Cityscapes val, 10-step | mIoU: improves the mIoU by 3.1% | Helped |
| add parsing prediction to support flow prediction | Ours · Cityscapes val, 10-step | EPE: reducing the EPE of PredFlow by 1.79 | Helped |
| use the learned static-object flow features to predict steering angle | Ours · Comma.ai steering angle | MSE (degrees²): 2.96 vs ~ 4 | Helped |
Closest papers by meaning
- SEPT: Towards Efficient Scene Representation Learning for Motion Prediction (ICLR 2024 · similarity 0.87)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.87)
- DINO-Foresight: Looking into the Future with DINO (NeurIPS 2025 · similarity 0.87)
- Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution (NeurIPS 2025 · similarity 0.86)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.86)
- MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting (NeurIPS 2026 · 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.