August 2020
tl;dr: Dynamic training of a CNN as an DRL agent to draw maps.
The following work are focused on road network discovery and are NOT focused on HD maps.
RoadTracer noted the semantic segmentation results are not a reliable foundation to extract road networks. Instead, it uses an iterative graph construction to get the topology of the road directly, avoiding unreliable intermediate representations.
The network needs to make a decision to step a certain distance toward a certain direction, resembling an agent in a reinforcement learning setting. This is somehow similar to the cSnake idea in Deep Boundary Extractor.