YOLO26 Drops NMS and DFL for Lighter End-to-End Detection

YOLO26 introduces a dual-head architecture that eliminates non-maximum suppression at inference and removes Distribution Focal Loss entirely, yielding a lighter detection head with unconstrained regression range. It also addresses small-object label assignment gaps present in prior YOLO versions. No benchmark numbers are available from the provided abstract excerpt.