Deep learning firm Deci AI, has launched YOLO-NAS, its latest deep learning model that delivers superior real-time object detection capabilities and high performance ready for production. The foundation model is based on Deci’s Neural Architecture Search Technology, AutoNAC™, which ensures unmatched precision and speed, surpassing other models like YOLOv6, YOLOv7, and YOLOv8, including the recently launched YOLOv8.
Deci’s AutoNAC technology is a revolutionary tool that democratises the use of Neural Architecture Search for all organizations, enabling teams to generate custom, fast, accurate, and efficient deep learning models promptly. AutoNAC delivers best-in-class deep learning model architectures for any task in any environment, achieving the best balance between accuracy and inference speed. It takes into account other components in the inference stack, including compilers and quantisation, in addition to being data and hardware aware.
The YOLO-NAS model delivers 50% (x1.5) more throughput and 1 mAP higher accuracy than other YOLO models, as shown in the chart, making it ideal for downstream Object Detection tasks in production environments.
The model is pre-trained on popular datasets such as COCO, Objects365, and Roboflow 100, making it highly suitable for real-world applications. The open-source model is available with pre-trained weights for research use (non-commercial) on Deci’s PyTorch-based, open-source, computer vision training library called SuperGradients.
Check out the GitHub repository here.
