We present a generalized and scalable method, called Gen-LaneNet, to detect3D lanes from a single image. The method, inspired by the lateststate-of-the-art 3D-LaneNet, is a unified framework solving image encoding,spatial transform of features and 3D lane prediction in a single network.However, we propose unique designs for Gen-LaneNet in two folds. First, weintroduce a new geometry-guided lane anchor representation in a new coordinateframe and apply a specific geometric transformation to directly calculate real3D lane points from the network output. We demonstrate that aligning the lanepoints with the underlying top-view features in the new coordinate frame iscritical towards a generalized method in handling unfamiliar scenes. Second, wepresent a scalable two-stage framework that decouples the learning of imagesegmentation subnetwork and geometry encoding subnetwork. Compared to3D-LaneNet, the proposed Gen-LaneNet drastically reduces the amount of 3D lanelabels required to achieve a robust solution in real-world application.Moreover, we release a new synthetic dataset and its construction strategy toencourage the development and evaluation of 3D lane detection methods. Inexperiments, we conduct extensive ablation study to substantiate the proposedGen-LaneNet significantly outperforms 3D-LaneNet in average precision(AP) andF-score.