VALNet: Vision-Based Autonomous Landing with Airport Runway Instance Segmentation
Visual navigation is an important upstream task for autonomous landing systems. Such navigation methods would have to rely on robust segmentation methods in order to accurately capture runway geometry. VALNet builds on top of the YOLOv8 architecture to handle specific challenges in the domain of runway segmentation, namely high variance in scale and robustness to rotations. Using a specially curated dataset, the authors claim large improvements in mAP, AP@50, and AP@75 over the baseline YOLOv8-seg approach.