Meet YOLO-NAS: An Open-Sourced YOLO-based Architecture Redefining State-of-the-Art in Object Detection

Deci AI has introduced a new object detection model called YOLO-NAS. YOLO-NAS stands for “You Only Look Once – Neural Architecture Search,” and it is a game-changer in object detection. This new model provides superior real-time object detection capabilities and production-ready performance.

Deci’s  Neural Architecture Search technology, AutoNAC™, generated the YOLO-NAS model. This engine lets users input tasks, data characteristics, inference environment, and performance targets. AutoNAC™ then guides the user to find the optimal architectures to give the best possible balance between accuracy and speed for their specific application. This engine is not only data and hardware aware but also considers other components in the inference stack, such as compilers and quantization. YOLO-NAS delivers state-of-the-art performance with unparalleled accuracy-speed performance. It outperforms other models, such as YOLOv5, YOLOv6, YOLOv7, and YOLOv8, in terms of accuracy and speed. Compared to YOLOv8 and YOLOv7, YOLO-NAS is about 0.5 mAP points more accurate and 10-20% faster.

The architecture of YOLO-NAS employs quantization-aware blocks and selective quantization for optimized performance. Quantization is a technique that converts floating-point models to integer models, which allows for more efficient inference on hardware that supports integer operations. When converted to the INT8 quantized version, YOLO-NAS experiences a much smaller precision drop than all other models that lose 1-2 mAP points during quantization. These techniques culminate in an innovative architecture with superior object detection capabilities and top-notch performance.

YOLO-NAS’s architecture is designed to be hardware and data-agnostic, allowing it to run efficiently on various hardware platforms, including CPUs, GPUs, and accelerators. Additionally, the architecture is designed to be flexible and scalable, allowing it to be used in various applications, such as autonomous vehicles, security systems, and robotics.

Deci’s mission is to provide AI teams with tools to help them attain efficient inference performance more quickly, and YOLO-NAS is a testament to that mission. By leveraging the power of AutoNAC™, Deci has developed a model that not only outperforms other models but also considers various components in the inference stack. This approach results in an efficient, scalable, and flexible model, making it suitable for various applications.

In conclusion, it is a game changer in object detection. Its superior real-time object detection capabilities and production-ready performance outperforms other models and delivers state-of-the-art performance. Deci’s mission to provide AI teams with tools to help them attain efficient inference performance more quickly is evident in the development of YOLO-NAS. By leveraging the power of AutoNAC™, Deci has developed a model that is efficient, scalable, and flexible, making it suitable for various applications.

Check out the Github link and Colab Note. Don’t forget to join our 20k+ ML SubReddit, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more. If you have any questions regarding the above article or if we missed anything, feel free to email us at Asif@marktechpost.com

🚀 Check Out 100’s AI Tools in AI Tools Club