Deformable attention detection transformer with iterative box refinement for improved object localization.