Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models
Summary
Ultralytics YOLO26 introduces a unified real-time vision model family with end-to-end, NMS-free inference and a lighter detection head, plus training innovations like MuSGD, Progressive Loss, and STAL. The model supports multiple tasks (detection, segmentation, pose, orientation) across five scales with an open-vocabulary extension, and code/models are available on GitHub. This represents an advance in real-time vision performance and deployment ease.