Only content delivered through the publication boundary on this date is included.
arXiv This review traces the Ultralytics YOLO family from YOLOv5 through YOLOv8, YOLO11, YOLO26, and YOLO27, highlighting YOLO27's scale-adaptive dual-architecture strategy in which compact n/s models use streamlined CNNs with dual-scale prediction and optional NMS-free inference while m/l models adopt query-based transformer decoding for native NMS-free detection and YOLO27l adds an UltraViT backbone with deep-stage self-attention, and it consolidates preliminary COCO accuracy and TensorRT latency figures, cross-generation comparisons, export and quantization paths, and deployment scenarios in robotics, agriculture, surveillance, and manufacturing, closing with open directions such as dense occlusion, domain generalization, open-vocabulary perception, temporal reasoning, and hardware-aware opti
This review traces the Ultralytics YOLO family from YOLOv5 through YOLOv8, YOLO11, YOLO26, and YOLO27, highlighting YOLO27's scale-adaptive dual-architecture strategy in which compact n/s models use streamlined CNNs with dual-scale prediction and optional NMS-free inference while m/l models adopt query-based transformer decoding for native NMS-free detection and YOLO27l adds an UltraViT backbone with deep-stage self-attention, and it consolidates preliminary COCO accuracy and TensorRT latency figures, cross-generation comparisons, export and quantization paths, and deployment scenarios in robotics, agriculture, surveillance, and manufacturing, closing with open directions such as dense occlusion, domain generalization, open-vocabulary perception, temporal reasoning, and hardware-aware opti
This review traces the Ultralytics YOLO family from YOLOv5 through YOLOv8, YOLO11, YOLO26, and YOLO27, highlighting YOLO27's scale-adaptive dual-architecture strategy in which compact n/s models use streamlined CNNs with dual-scale prediction and optional NMS-free inference while m/l models adopt query-based transformer decoding for native NMS-free detection and YOLO27l adds an UltraViT backbone with deep-stage self-attention, and it consolidates preliminary COCO accuracy and TensorRT latency figures, cross-generation comparisons, export and quantization paths, and deployment scenarios in robotics, agriculture, surveillance, and manufacturing, closing with open directions such as dense occlusion, domain generalization, open-vocabulary perception, temporal reasoning, and hardware-aware opti
This review traces the Ultralytics YOLO family from YOLOv5 through YOLOv8, YOLO11, YOLO26, and YOLO27, highlighting YOLO27's scale-adaptive dual-architecture strategy in which compact n/s models use streamlined CNNs with dual-scale prediction and optional NMS-free inference while m/l models adopt query-based transformer decoding for native NMS-free detection and YOLO27l adds an UltraViT backbone with deep-stage self-attention, and it consolidates preliminary COCO accuracy and TensorRT latency figures, cross-generation comparisons, export and quantization paths, and deployment scenarios in robotics, agriculture, surveillance, and manufacturing, closing with open directions such as dense occlusion, domain generalization, open-vocabulary perception, temporal reasoning, and hardware-aware opti