The code implementation of LaneMapNet, a unified end-to-end model designed for both HD Map Construction and Lane Network Recognization using a BEV perception framework. The project is based on mmdetection3d and MapTR.
Method | Backbone | 2D to BEV | Config | Checkpoint |
---|---|---|---|---|
LaneMapNet | ResNet50 | GKT | config | model |
Visualization results for two scenarios on the NuScenes val split. Columns 2-4 represent road divider (red), road boundary (green), and pedestrian crossing (blue). Columns 5-8 use green and red dots to indicate the beginning and end of a lane line.
A comprehensive presentation of HD map and lane network. The output of LaneMapNet is shown on the left and GT on the right.