Research map / Video / Video segmentation and tracking
End-To-End Learning of Driving Models From Large-Scale Video Datasets
Check 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 |
|---|---|---|---|
| Add semantic segmentation loss as side supervision (ablation) | FCN-LSTM on the 1000-video subset · Discrete action prediction | perplexity: 0.718 -> 0.697 | Helped |
| Add semantic segmentation loss as side supervision (ablation) | FCN-LSTM on the 1000-video subset · Discrete action prediction | accuracy: 71.31% -> 72.4% | Helped |
| Replace end-to-end side supervision with a segmentation-first pipeline | Driving model on the 1000-video subset · Discrete action prediction | accuracy: 72.4% -> 61.66 | Hurt |
| Use a longer temporal-convolution history window (ablation) | CNN temporal-convolution model · Discrete feasible action prediction | perplexity: 0.445 -> 0.411 | Helped |
| Use data-driven equal-probability angle bins instead of linear bins (ablation) | Continuous lane-following model · Continuous lane following | Angle perplexity: -2.82 -> -4.83 | Helped |
| Use data-driven equal-probability angle bins instead of log bins (ablation) | Continuous lane-following model · Continuous lane following | Angle perplexity: -3.66 -> -4.83 | Helped |
| Add speed input to CNN-LSTM (ablation) | CNN-LSTM · Discrete feasible action prediction | perplexity: 0.419 -> 0.449 | Hurt |
| Add speed input to CNN-LSTM (ablation) | CNN-LSTM · Discrete feasible action prediction | accuracy: 84.5% -> 84.2% | Hurt |
| Apply inverse-frequency action weighting (ablation) | Discrete action prediction model · Discrete feasible action prediction | perplexity | No clear effect |
| Replace CNN visual encoder with FCN (ablation) | LSTM-based discrete action model · Discrete feasible action prediction | accuracy: 84.5% -> 84.1% | No clear effect |
Closest papers by meaning
- Learning to Drive is a Free Gift: Large-Scale Label-Free Autonomy Pretraining from Unposed In-The-Wild Videos (CVPR 2026 · similarity 0.88)
- DrivingDojo Dataset: Advancing Interactive and Knowledge-Enriched Driving World Model (NeurIPS 2024 · similarity 0.86)
- Uncertainty-Guided Never-Ending Learning to Drive (CVPR 2024 · similarity 0.86)
- Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models (NeurIPS 2025 · similarity 0.86)
- Vista: A Generalizable Driving World Model with High Fidelity and Versatile Controllability (NeurIPS 2024 · similarity 0.85)
- Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene (CVPR 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.