This blog post was originally published at NVIDIA’s website. It is reprinted here with the permission of NVIDIA.
With NVIDIA TAO Toolkit, developers around the world are building AI-powered visual perception and computer vision applications. Now the process is faster and easier than ever, thanks to significant platform enhancements and strong ecosystem adoption.
NVIDIA TAO Toolkit supports more than 10 computer vision and vision AI modalities, including image classification, object detection, three types of segmentation, optical character recognition (OCR), action recognition, key point estimation, body pose estimation, embedding models, Siamese networks, and more.
Getting started with TAO Toolkit is faster than ever, with support for over 40 pretrained models on NVIDIA NGC. Recipes for leveraging TAO continue to expand with workflows for tuning models for various industries. To learn more, see Transforming Industrial Defect Detection with NVIDIA TAO and Vision AI Models and Customizing AI Models: Train Character Detection and Recognition Models with NVIDIA TAO.
Massive adoption
TAO has been downloaded over 100,000 times with nearly 1 million downloads of TAO pretrained models. And now that TAO is open source, solution providers can access more custom integration of TAO into their services and explore the inner workings of the platform with finer granularity.
Leading MLOps and cloud services are leveraging TAO to improve their services. Integration with MLOps platforms such as Weights & Biases and ClearML helps streamline machine learning (ML) workflows, enabling better experiment tracking for model training. TAO can also work with various cloud ML and Kubernetes services such as Azure ML, GCP Vertex AI, Azure AKS, AWS EKS, and GCP GKE.
Major enterprises are also building TAO into their own industry-specific AI development workflows, including PepsiCo in consumer supply chain, Pegatron, Siemens, and others in manufacturing, and ExxonMobil in energy. Large cities and airports around the world also use TAO.
Built for AI workflows
Enterprises large and small are building their own vision AI factories and development workflows using TAO. These AI workflows are increasingly requiring the following three capabilities now provided by TAO.
First, TAO can handle the latest AI models and algorithms, including foundation model tuning, generative AI, and vision transformers. Second, TAO can seamlessly connect with synthetically generated datasets from simulation approaches such as NVIDIA Omniverse Replicator and Stable Diffusion. Third, TAO includes enterprise-level support as an essential tool in the development workflow through NVIDIA AI Enterprise.
Advanced AI on trillions of devices
With TAO, NVIDIA offers the flexibility to deploy the latest AI models on trillions of devices at the far edge through ONNX and TFLite model export, along with strong ecosystem adoption from the world’s leading providers and edge AI software platforms.
ARM is leveraging TAO to optimize AI runtime on Ethos NPU devices. STMicroelectronics uses TAO to run complex vision AI for the first time on STM32 microcontrollers. And partners like Edge Impulse and Nota are integrating TAO into their edge AI platforms to bring edge-optimized solutions to their customers.
Get started
Download NVIDIA TAO Toolkit to get started building AI-powered visual perception and computer vision applications. Access expert help on the NVIDIA Developer TAO Forum.
Join the ST Edge AI Summit on December 6 to hear NVIDIA VP and GM of Embedded and Edge Computing Deepu Talla speak about building the next generation of edge AI platforms, and how NVIDIA TAO continues to evolve.
Related resources
- GTC session: AI Models Made Simple Using TAO (Spring 2023)
- GTC session: Solving Computer Vision Grand Challenges in One-Click (Spring 2023)
- SDK: TAO Toolkit
- Webinar: Unleash the Power of Vision Transformers, NVIDIA TAO, and the Latest NVIDIA GPUs
- Webinar: Accelerate AI Model Creation Using AutoML in NVIDIA TAO 4.0
- Webinar: Top 5 Reasons Why the TAO Toolkit Should Be Part of Your AI Development Effort Today
Adam Scraba
Accelerated Computing Platform Worldwide Evangelism and Marketing, NVIDIA
