June 2020
tl;dr: Detached model to perform domain adaptation sim2real.
Two stage model to bridge domain gap. This is very similar to GenLaneNet for 3D LLD prediction.
The first stage model already extracted away domain-dependent features and thus the second stage model can be used as is.
- The first stage generates the pseudo-lidar point cloud in BEV.
- The second stage converts pseudo-lidar point cloud to BEV segmentation.
- During inference, only finetune first stage. Use second stage as is.
- The simulated segmentation GT can be obtained in abundance with simulator such as CARLA.
- Summary of technical details