Separated RoadTopoFormer

Understanding driving scenarios is crucial to realizing autonomous driving.Previous works such as map learning and BEV lane detection neglect theconnection relationship between lane instances, and traffic elements detectiontasks usually neglect the relationship with lane lines. To address theseissues, the task is presented which includes 4 sub-tasks, the detection oftraffic elements, the detection of lane centerlines, reasoning connectionrelationships among lanes, and reasoning assignment relationships between lanesand traffic elements. We present Separated RoadTopoFormer to tackle the issues,which is an end-to-end framework that detects lane centerline and trafficelements with reasoning relationships among them. We optimize each moduleseparately to prevent interaction with each other and aggregate them togetherwith few finetunes. For two detection heads, we adopted a DETR-likearchitecture to detect objects, and for the relationship head, we concat twoinstance features from front detectors and feed them to the classifier toobtain relationship probability. Our final submission achieves 0.445 OLS, whichis competitive in both sub-task and combined scores.