Reusing Attention for One-stage Lane Topology Understanding

2025-07-23

AI Summary

To address low accuracy and slow inference in lane topology understanding for autonomous driving, this paper proposes a single-stage end-to-end framework that jointly predicts traffic elements, lane centerlines, and their topological relationships. The core innovation lies in reusing the attention mechanisms within the Transformer decoder to explicitly model topological dependencies—eliminating error propagation and redundant computation inherent in two-stage approaches. It is the first work to enable cross-task sharing of attention resources from the detection module within a single-stage architecture. Additionally, we introduce a graph-free knowledge distillation strategy that transfers prior knowledge from standard HD maps to a lightweight, map-agnostic model. Evaluated on OpenLane-V2, our method achieves new state-of-the-art performance, improving AP by +3.2% while accelerating inference by 2.1×. The source code is publicly available.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Computer Vision: Large Vision Models
• Machine Learning: Mixture of Experts (MoE)

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

Understanding lane toplogy relationships accurately is critical for safe autonomous driving. However, existing two-stage methods suffer from inefficiencies due to error propagations and increased computational overheads. To address these challenges, we propose a one-stage architecture that simultaneously predicts traffic elements, lane centerlines and topology relationship, improving both the accuracy and inference speed of lane topology understanding for autonomous driving. Our key innovation lies in reusing intermediate attention resources within distinct transformer decoders. This approach effectively leverages the inherent relational knowledge within the element detection module to enable the modeling of topology relationships among traffic elements and lanes without requiring additional computationally expensive graph networks. Furthermore, we are the first to demonstrate that knowledge can be distilled from models that utilize standard definition (SD) maps to those operates without using SD maps, enabling superior performance even in the absence of SD maps. Extensive experiments on the OpenLane-V2 dataset show that our approach outperforms baseline methods in both accuracy and efficiency, achieving superior results in lane detection, traffic element identification, and topology reasoning. Our code is available at https://github.com/Yang-Li-2000/one-stage.git.

Problem

• Improving accuracy and speed of lane topology understanding
• Reusing attention to avoid expensive graph networks
• Enabling performance without standard definition maps

Innovation

• One-stage architecture predicts elements and topology simultaneously
• Reuses attention in transformer decoders for efficiency
• Distills knowledge from SD maps to non-SD models