Research map / 3D vision / 3D generation
HSDF-Lane: Height-Aligned Signed Distance Field with Semantic Lane Prior for 3D Lane Detection
Jiyong Boo, ByeongIn Joung, Hyemin Yang, KUK-JIN YOON
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 |
|---|---|---|---|
| replace direct height regression with HSDF formulation (ablation) | HSDF-Lane · OpenLane (full, height map) | overall MAE: 0.183 -> 0.159 | Improved |
| add height-directional Eikonal regularization (ablation) | HSDF-Lane · OpenLane (full, height map) | overall RMSE: 0.347 -> 0.331 | Improved |
| add Eikonal regularization on top of height supervision (ablation) | HSDF-Lane · OpenLane-300 | F1-score (%): 74.2 -> 74.7 | Improved |
| add explicit height map supervision L_H (ablation) | HSDF-Lane · OpenLane-300 | F1-score (%): 71.8 -> 74.2 | Improved |
| add LSPE (ablation) | HSDF-Lane · OpenLane-300 | F1-score (%): 74.7 -> 75.6 | Improved |
| add LSPE (single learnable vector gated by heatmap) (ablation) | HSDF-Lane · OpenLane val (full) | F1-score (%): 65.9 -> 66.3 | Improved |
| encode heatmap with full learnable MLP instead of single vector (ablation) | LSPE · OpenLane val (full) | F1-score (%): 66.3 -> 65.7 | Hurt |
| concatenate heatmap instead of gating a learnable vector (ablation) | LSPE · OpenLane val (full) | F1-score (%): 66.3 -> 66.0 | Hurt |
| vary max slope angle theta from 5 to 4 or 6 degrees (ablation) | HSDF-Lane · OpenLane-300 | F1-score (%): 75.6 -> 75.6 (theta 4), 75.6 -> 74.8 (theta 6) | Hurt |
| use FPN multi-scale features with 4-layer decoder | HSDF-Lane · OpenLane val | F1-score (%): 66.3 -> 66.9 | Improved |
| replace dense LiDAR height supervision with lane-derived pseudo heights (ablation) | HSDF-Lane · OpenLane val | F1-score (%): 66.3 -> 66.0 | No clear effect |
| replace anchor-based height regression with HSDF dense sampling | HeightLane/SC-Lane · OpenLane height map | MAE: 0.176 -> 0.158 (vs SC-Lane) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SC-Lane — SC-Lane: Slope-aware and Consistent Road Height Estimation Framework for 3D Lane Detection (ICCV 2025) · better on 2 of 3 shared results
- SparseLaneSTP — SparseLaneSTP: Leveraging Spatio-Temporal Priors with Sparse Transformers for 3D Lane Detection (ICCV 2025) · better on 1 of 1 shared result
- LATR — LATR: 3D Lane Detection from Monocular Images with Transformer (ICCV 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Lanes and Points in Complex Scenarios for Monocular 3D Lane Detection (CVPR 2025 · similarity 0.89)
- LaneCPP: Continuous 3D Lane Detection using Physical Priors (CVPR 2024 · similarity 0.88)
- DV-3DLane: End-to-end Multi-modal 3D Lane Detection with Dual-view Representation (ICLR 2024 · similarity 0.88)
- Sketch and Refine: Towards Fast and Accurate Lane Detection (AAAI 2024 · similarity 0.88)
- PVALane: Prior-Guided 3D Lane Detection with View-Agnostic Feature Alignment (AAAI 2024 · similarity 0.87)
- GLane3D: Detecting Lanes with Graph of 3D Keypoints (CVPR 2025 · 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.
