7+ Resources for Getting Started with YOLO-NAS, A Next-Generation, Object Detection Foundation Model

Since the release of YOLO-NAS last week, we’ve been thrilled seeing a lot of use cases and feedback from the community.

With its superior real-time object detection capabilities and production-ready performance, YOLO-NAS can detect small objects better, improve localization accuracy, and increase the performance-per-compute ratio. These capabilities make YOLO-NAS more accessible for real-time edge-device applications such as robotics, autonomous driving, and video analytics.

To recap the numbers, the chart above shows how the YOLO-NAS (m) model delivers?a 50% (x1.5) increase in throughput and 1 mAP better accuracy compared to other SOTA YOLO models on the NVIDIA T4 GPU.

And now, here are some use cases and tutorials, most of which were created by the community, that can help you get started with YOLO-NAS if you haven't yet.

1. General Overview and Webcam Inference by Nicolai Nielsen

Nicolai provides an introduction to YOLO-NAS, including how to set it up and use a pre-trained model to make predictions. He also fine-tunes YOLO-NAS on a custom dataset, and shows you how to export and use it on custom Python script and projects.

2. Personal Protective Equipment (PPE) Detection by Muhammad Moin

In this video, Muhammad walks you through taking a custom dataset, fine-tuning YOLO-NAS on the dataset, and exporting it.

3. YOLO-NAS Explainer with a Variety of Video Inference Results by Vaibhab Singh @ LearnOpenCV

In this article, Vaibhab provides a short brief on YOLO-NAS' architecture, training regime, and metrics comparison against other YOLO models. He also shares several video inference results including traffic, cats, and drone clips.

4. A Visual Deep Dive into YOLO-NAS by CS Board

In this video, learn more about the mechanics behind how YOLO-NAS works. It explains?what neural architecture search is, what quantization is, and how YOLO-NAS is a foundation model.

5. A Q&A Type Introduction to YOLO-NAS by Nir B.

From what is unique about the YOLO-NAS architecture to how it was trained with SuperGradients, Nir shares an overview of the new object detection foundation model. But the best part? DagsHub is integrated with SuperGradients to enable experiment logging with MLflow and DVC when training SOTA object detection models such as YOLO-NAS.

6. A Collection of Notebooks for Various Use Cases by Harpreet Sahota ??

Our DevRel Manager, Harpreet, prepared a lot of amazing notebooks so you can easily get started with YOLO-NAS. Check them out:

• Learn about YOLO-NAS and?the SuperGradients training library with this?introductory notebook
• Use this?template?to solve most object detection tasks
• Make societies not only smarter but also safer with this?pistol detection notebook
• Improve workplace safety with?hard hat and vest detection
• Work better with aerial images with this?stadium detection from aerial images notebook
• Detect ships from aerial images and test YOLO-NAS’ capability to detect small objects
• Make a case for enhancing your city commuting experience with this?pothole detection notebook

7. Fall Detection Using YOLO-NAS by Aarohi Singla

Aarohi takes you through the YOLO-NAS notebook and shows how you can train and fine-tune YOLO-NAS with a custom dataset, using it to detect falls on images and videos.

Have you experimented with YOLO-NAS yet? Give it a try today! Check out YOLO-NAS?technical blog and starter notebook, and be sure to give it a star on the?YOLO-NAS on GitHub.