Lane Segmentation Refinement with Diffusion Models

2024-05-01

arXiv.org

AI Summary

To address incomplete masks, blurred boundaries, and topological discontinuities in aerial-image-based lane segmentation—leading to geometric and topological distortions in HD map construction—this paper proposes a diffusion-based post-processing method integrating a conditional diffusion probabilistic model (DPM). It introduces DPMs for the first time into lane-level semantic segmentation refinement, synergistically combining initial UNet predictions, graph-structured loss supervision, and joint GEO/TOPO F1 optimization. The approach significantly improves geometric continuity and topological consistency of non-intersection lanes. On public benchmarks, it achieves +6.2% TOPO F1 and +3.8% GEO F1 over baselines, effectively eliminating topological breaks and spurious branches while enhancing lane-graph connectivity and structural reliability—thereby delivering more robust lane maps for high-precision navigation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Planning, Routing, and Scheduling: Planning with Language Models
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

The lane graph is a key component for building high-definition (HD) maps and crucial for downstream tasks such as autonomous driving or navigation planning. Previously, He et al. (2022) explored the extraction of the lane-level graph from aerial imagery utilizing a segmentation based approach. However, segmentation networks struggle to achieve perfect segmentation masks resulting in inaccurate lane graph extraction. We explore additional enhancements to refine this segmentation-based approach and extend it with a diffusion probabilistic model (DPM) component. This combination further improves the GEO F1 and TOPO F1 scores, which are crucial indicators of the quality of a lane graph, in the undirected graph in non-intersection areas. We conduct experiments on a publicly available dataset, demonstrating that our method outperforms the previous approach, particularly in enhancing the connectivity of such a graph, as measured by the TOPO F1 score. Moreover, we perform ablation studies on the individual components of our method to understand their contribution and evaluate their effectiveness.

Problem

• Refining lane segmentation masks from aerial imagery using diffusion models
• Improving lane graph connectivity and completeness despite occlusions
• Enhancing segmentation-to-graph conversion for autonomous driving applications

Innovation

• Refines lane masks using diffusion models
• Improves graph connectivity and segmentation quality
• Outperforms CNN-only and prior diffusion approaches