Research map / Video / Video segmentation and tracking
SIGMA-Lane: Scale-pyramId Gated MAmba for Temporally Consistent Video Lane Detection
Zhang Tiancheng, Mengmeng Wang, Yan Gao, Xiangjie Kong, Guojiang Shen, Jiaxin Du
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 input gating (DG-I) (ablation) | Mamba2 baseline · VIL-100 | mIoU: 0.775 -> 0.781 | Improved |
| add output gating (DG-O) (ablation) | Mamba2 baseline · VIL-100 | mIoU: 0.775 -> 0.778 | Improved |
| add both input and output gating (dual-gating) (ablation) | Mamba2 baseline · VIL-100 | mIoU: 0.775 -> 0.787 | Improved |
| add dual-gating (ablation) | Mamba2 baseline · VIL-100 | missing rate RM@0.5: 0.084 -> 0.032 | Improved |
| add lane-guided affine warp (ablation) | dual-gated model · VIL-100 | mIoU: 0.787 -> 0.794 | Improved |
| add structural spatial retrieval (SSR) (ablation) | dual-gated model with warp · VIL-100 | mIoU: 0.794 -> 0.798 | Improved |
| add geometry-aware start token (ablation) | SIGMA-Lane without token · VIL-100 | mIoU: 0.798 -> 0.801 | Improved |
| remove both gates while keeping the original obstacle mask (ablation) | SIGMA-Lane · OpenLane-V | F1: -0.0180 F1, -0.0150 mIoU, +0.0740 RM | Hurt |
| dilate obstacle mask by 7x7 (ablation) | SIGMA-Lane · OpenLane-V | F1: -0.0014 F1, -0.0010 mIoU | No clear effect |
| erase 20% of obstacle mask pixels (ablation) | SIGMA-Lane · OpenLane-V | F1: +0.0007 F1, -0.0005 mIoU | No clear effect |
| add memory perturbations along occlusion boundaries (with gates vs ungated) (ablation) | SIGMA-Lane · OpenLane-V | RM@0.5: 0.162 -> 0.182 (gated) vs 0.236 -> 0.367 (ungated) | Mixed |
| use SIGMA-Lane instead of OMR on Moderate and Heavy occlusion bins | SIGMA-Lane vs OMR · OpenLane-V | mIoU: +0.028 mIoU | Improved |
Closest papers by meaning
- OMR: Occlusion-Aware Memory-Based Refinement for Video Lane Detection (ECCV 2024 · similarity 0.89)
- Sketch and Refine: Towards Fast and Accurate Lane Detection (AAAI 2024 · similarity 0.87)
- Lane2Seq: Towards Unified Lane Detection via Sequence Generation (CVPR 2024 · similarity 0.86)
- Rethinking Lanes and Points in Complex Scenarios for Monocular 3D Lane Detection (CVPR 2025 · similarity 0.86)
- GSENet:Global Semantic Enhancement Network for Lane Detection (AAAI 2024 · similarity 0.85)
- Selective Focus: Investigating Semantics Sensitivity in Post-training Quantization for Lane Detection (AAAI 2024 · 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.
