Research map / 3D vision / 3D generation
End-to-End Deep Structured Models for Drawing Crosswalks
Justin Liang, Raquel Urtasun
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 camera to LiDAR input | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 95.8 -> 96.6 | Improved |
| add LiDAR to camera input | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 94.9 -> 96.6 | Improved |
| use multiple passes instead of a single pass | crosswalk model · camera+LiDAR | precision at 40cm: 91.5 -> 96.6 | Improved |
| add structured inference over CNN segmentation output | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 93.1 -> 96.6 | Improved |
| add structured prediction over nearest neighbor on VGG features | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 24.8 -> 96.6 | Improved |
| add +-2 and +-5 degree angle search (ablation) | structured inference · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 94.4 -> 96.6 | Improved |
| add centerline angle hypothesis (ablation) | structured inference · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 96.3 -> 96.6 | Improved |
| add predicted angle from CNN (ablation) | structured inference · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 85.3 -> 96.6 | Improved |
| use ground-truth segmentation (ablation) | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 96.6 -> 97.8 | Improved |
| use ground-truth distance transform (ablation) | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 96.6 -> 96.6 | No clear effect |
| use ground-truth angle (ablation) | crosswalk model · Crosswalk dataset test (offline, multi-pass) | precision at 40cm: 96.6 -> 96.5 | No clear effect |
| retrain with images without crosswalks | crosswalk model · images without crosswalks | false positive rate: 5.7% -> 0.04% | Improved |
Closest papers by meaning
- Convolutional Recurrent Network for Road Boundary Extraction (CVPR 2019 · similarity 0.86)
- Sketch and Refine: Towards Fast and Accurate Lane Detection (AAAI 2024 · similarity 0.83)
- Effective Multi-sensor Conditioning for Street-view Novel-view Synthesis (NeurIPS 2026 · similarity 0.82)
- Controllable Road Marking Generation (NeurIPS 2026 · similarity 0.82)
- L2RSI: Cross-view LiDAR-based Place Recognition for Large-scale Urban Scenes via Remote Sensing Imagery (NeurIPS 2025 · similarity 0.82)
- LidaRF: Delving into Lidar for Neural Radiance Field on Street Scenes (CVPR 2024 · similarity 0.82)
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.
