YOLO-NAS - A New Best Object Detection Model!

In May 3, 2023 a new best YOLO model was introduced called YOLO-NAS which achieves state of the art performance on object detection tasks.
In this video we will explain how YOLO NAS model has been created by Deci AI.

Please subscribe to the channel to support the creation of more educational content.

We will talk about Neural Architecture Search which is a subdomain of AutoML to automatically create model architectures, which Deci has a proprietary implementation of called AutoNAC which stands for Automated Neural Architecture Construction.

Additionally we will discuss about the quantization aware architecture of YOLO-NAS which enabled the model to be optimized to run with very low latencies while preserving very high accuracy.
To do that they have used QARepVGG, an improved version of RepVGG, which was recently introduced in a research paper by Meituan - https://arxiv.org/abs/2212.01593

They also used a hybrid quantization technology to balance between accuracy and latency.

YOLO NAS was trained on large datasets such as Object365 and COCO and it can serve as a foundational model for object detection by fine-tuning it on datasets which are relevant for downstream tasks.

Original YOLO paper - https://arxiv.org/abs/1506.02640
YOLO NAS GitHub - https://github.com/Deci-AI/super-grad...


Chapters:
0:00 Introducing YOLO-NAS
0:46 Neural Architecture Search
2:12 Quantization Aware Architecture
3:59 Object Detection Foundational Model