Trainingdeepmodelsforlanedetectionischallengingduetotheverysubtleandsparsesupervisorysignalsin-herentinlaneannotations.Withoutlearningfrommuchrichercontext,thesemodelsoftenfailinchallengingsce-narios,e.g.,severeocclusion,ambiguouslanes,andpoorlightingcon