You are right. In our code, we only provide the experiments limited to the systhetic dataset. If to consider real-world application, a lot additional work both on methods and dataset needs to be done. I would suggest 1. training segmentation network on huge 2D real world dataset. 2. More synthetic binary maps associated with 3D lanelines considering highly diversed terrian maps.
You are right. In our code, we only provide the experiments limited to the systhetic dataset. If to consider real-world application, a lot additional work both on methods and dataset needs to be done. I would suggest 1. training segmentation network on huge 2D real world dataset. 2. More synthetic binary maps associated with 3D lanelines considering highly diversed terrian maps.