AD-SAM: Adapting the Segment Anything Model for Semantic Segmentation in Autonomous Driving
AD-SAM adapts Meta's Segment Anything Model (SAM) for autonomous driving semantic segmentation by integrating a dual-encoder system and a hybrid loss function. It combines global context from a Vision Transformer with local detail from ResNet-50, using deformable fusion and multi-stage refinement to achieve superior accuracy on road scenes while demonstrating faster training and better robustness.