Research map / Graph learning / Knowledge and temporal graphs
Lightweight Spatio-Temporal Modeling via Temporally Shifted Distillation for Real-Time Accident Anticipation
Patrik Patera, Yie-Tarng Chen, Wen-Hsien Fang
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 contrastive video-text loss in pre-training (ablation) | student with spatial and temporal distillation · DAD | mAP (%): 70.1 -> 75.3 | Improved |
| add temporally shifted distillation to spatial-only distillation (ablation) | student · DAD | mAP (%): 71.2 -> 75.3 | Improved |
| add spatial distillation (ablation) | student · DAD | mAP (%): 74.1 -> 75.3 | Improved |
| use a one-frame temporal shift instead of none (ablation) | student · DAD | mAP (%): 71.2 -> 75.3 | Improved |
| increase temporal shift from 1 to 3 frames (ablation) | student · DAD | mAP (%) and mTTA: mAP 75.3 -> 70.7; mTTA 4.04 -> 4.13 | Mixed |
| increase RWKV layers from 6 to 8 (ablation) | student · DAD | mAP (%): 75.33 -> 71.6 | Hurt |
| use moderate 30% masking in memory strategy (ablation) | student pre-training · DAD | mAP (%): 74.0 -> 75.3 | Improved |
| raise mask ratio to 50% or 75% (ablation) | student pre-training · DAD | mAP (%): 75.3 -> 71.8 (50%), 71.0 (75%) | Hurt |
| swap image teacher MobileCLIP for video teacher V-JEPA2 (ablation) | distillation · DAD | mAP (%): 75.3 -> 66.0 | Hurt |
| add temporal shift when distilling from video teacher V-JEPA2 (ablation) | distillation · DAD | mTTA (s): 4.12 -> 4.21 | Improved |
| add temporally shifted distillation to temporal RWKV (vs no pre-training) (ablation) | student · DAD | mAP (%): 39.4 -> 75.3 | Improved |
| add temporal recurrence (T-RWKV) to spatial RWKV with TSD (ablation) | student · DAD | mAP (%): 55.6 -> 75.3 | Improved |
Closest papers by meaning
- Accident Anticipation via Temporal Occurrence Prediction (NeurIPS 2025 · similarity 0.85)
- STSBench: A Spatio-temporal Scenario Benchmark for Multi-modal Large Language Models in Autonomous Driving (NeurIPS 2025 · similarity 0.85)
- FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving (NeurIPS 2025 · similarity 0.85)
- TUMTraf VideoQA: Dataset and Benchmark for Unified Spatio-Temporal Video Understanding in Traffic Scenes (ICML 2025 · similarity 0.85)
- SEPT: Towards Efficient Scene Representation Learning for Motion Prediction (ICLR 2024 · similarity 0.85)
- DriveSpatial: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving (NeurIPS 2026 · similarity 0.84)
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.
