Kalfaoglu M. Esat ; Ozturk Halil Ibrahim ; Kilinc Ozsel ; Temizel Alptekin

Abstract

Recently, the centerline has become a popular representation of lanes due toits advantages in solving the road topology problem. To enhance centerlineprediction, we have developed a new approach called TopoMask. Unlike previousmethods that rely on keypoints or parametric methods, TopoMask utilizes aninstance-mask-based formulation coupled with a masked-attention-basedtransformer architecture. We introduce a quad-direction label representation toenrich the mask instances with flow information and design a correspondingpost-processing technique for mask-to-centerline conversion. Additionally, wedemonstrate that the instance-mask formulation provides complementaryinformation to parametric Bezier regressions, and fusing both outputs leads toimproved detection and topology performance. Moreover, we analyze theshortcomings of the pillar assumption in the Lift Splat technique and adapt amulti-height bin configuration. Experimental results show that TopoMaskachieves state-of-the-art performance in the OpenLane-V2 dataset, increasingfrom 44.1 to 49.4 for Subset-A and 44.7 to 51.8 for Subset-B in the V1.1 OLSbaseline.