TopoMask: Instance-Mask-Based Formulation for the Road Topology Problem via Transformer-Based Architecture

Driving scene understanding task involves detecting static elements such aslanes, traffic signs, and traffic lights, and their relationships with eachother. To facilitate the development of comprehensive scene understandingsolutions using multiple camera views, a new dataset called Road Genome(OpenLane-V2) has been released. This dataset allows for the exploration ofcomplex road connections and situations where lane markings may be absent.Instead of using traditional lane markings, the lanes in this dataset arerepresented by centerlines, which offer a more suitable representation of lanesand their connections. In this study, we have introduced a new approach calledTopoMask for predicting centerlines in road topology. Unlike existingapproaches in the literature that rely on keypoints or parametric methods,TopoMask utilizes an instance-mask based formulation with a transformer-basedarchitecture and, in order to enrich the mask instances with flow information,a direction label representation is proposed. TopoMask have ranked 4th in theOpenLane-V2 Score (OLS) and ranked 2nd in the F1 score of centerline predictionin OpenLane Topology Challenge 2023. In comparison to the currentstate-of-the-art method, TopoNet, the proposed method has achieved similarperformance in Frechet-based lane detection and outperformed TopoNet inChamfer-based lane detection without utilizing its scene graph neural network.