TAO Toolkit 3.0-21.08
- Highlights:
- New pretrained models and training for computer vision:
- Body Pose estimation
- Emotion recognition
- Facial landmark
- License plate detection and recognition
- Heart rate estimation
- Gesture recognition
- Gaze estimation
- People segmentation
- Introducing ASR and NLP models with inference samples for:
- Speech to Text
- Named Entity Recognition (NER)
- Question/Answering
- Punctuation
- Text classification
- Turnkey training support on AWS, GCP and Azure.
- TAO Toolkit 3.0 brings support for NVIDIA Ampere GPUs with third generation tensor core additions and various performance optimizations
- Improved PeopleNet model to detect difficult scenarios such as people sitting down, rotated/ warped objects
- Train with popular networks: EfficientNet, ResNet, YOLOV3/V4, FasterRCNN, SSD, DetectNet_v2, MaskRCNN and UNET
- Out of the box deployment on NVIDIA Triton and DeepStream SDK for vision AI and Riva for conversational AI
- Enable faster training with jobs split up across multi-GPUs
| Operating System |
|
|---|---|
| Dependencies |
|
Latest Product News
Developer Tutorial
Learn how to create highly optimized production quality pose estimation model.
Supercharge Your Workflows With Transfer Learning
Learn how NVIDIA TAO Toolkit and Pretrained Models can transform your development efforts.
Developer Tutorial
Learn how to create a real-time number plate detection and recognition app.
Developer Tutorial
Learn how to build and deploy conversational AI models using the NVIDIA TAO Toolkit.
Getting Started Resources
Conversational AI
- Conversational AI models and container collection: Download from NGC
- To deploy with Riva, go to download resources
- Get started with Jupyter Notebooks:
Vision AI
- Computer vision models and container collection: Download from NGC
- Collection of Jupyter Notebooks and training specs for vision AI model
- To deploy TAO Toolkit models using DeepStream, go to download resources.
- To deploy TAO toolkit models using Triton, go to TAO Triton repo
- DeepStream sample apps to deploy YoloV3/V4, FasterRCNN, UNET, etc.
- Models trained with TAO Toolkit Sample App for license plate detection and recognition using DeepStream
- Sample App for Gaze Estimation, Facial Landmark, HeartRateNet, Gesture and Emotion - x86 | Jetson
- To convert TAO Toolkit model (etlt) to TensorRT engine for deployment with DeepStream, select the appropriate
tao-converterfor your hardware and software stack.
| Platform | Compute | Download |
|---|---|---|
| x86 + GPU | CUDA 10.2 / cuDNN 8.0 / TensorRT 7.1 | Download |
| x86 + GPU | CUDA 10.2 / cuDNN 8.0 / TensorRT 7.2 | Download |
| x86 + GPU | CUDA 11.0 / cuDNN 8.0 / TensorRT 7.1 | Download |
| x86 + GPU | CUDA 11.0 / cuDNN 8.0 / TensorRT 7.2 | Download |
| x86 + GPU | CUDA 11.1 / cuDNN 8.0 / TensorRT 7.2 | Download |
| x86 + GPU | CUDA 11.3 / cuDNN 8.1 / TensorRT 8.0 | Download |
| Jetson | Jetpack 4.4 | Download |
| Jetson | Jetpack 4.5 | Download |
| Jetson | Jetpack 4.6 | Download |
| Clara AGX | CUDA 11.1 / CuDNN 8.0.5 / TensorRT 7.2.2 | Download |
Documentation
Developer Forums
- Post your questions or feedback in the following associated forums:
Success Stories
- Learn how CVEDIA taps NVIDIA TAO Toolkit to cut down model development time for Cubic’s GRIDSMART Traffic Systems
- Learn how Lexmark slashes AI skills development cycle by 25%
- Explore how INEX revolutionizes toll road systems with real-time video processing
- Find out how Nota cuts development time by 50% for real-time traffic control system
- Explore how other top AI teams utilize NVIDIA pre-trained models and TAO Toolkit to transform the world around us. Read now
Blogs & Tutorials
- Training and Optimizing a 2D Pose Estimation Model with the NVIDIA TAO Toolkit, Part 1 | Part 2
- Preparing State-of-the-Art Models for Classification and Object Detection with the NVIDIA TAO Toolkit
- Creating a Real-Time License Plate Detection and Recognition App
- Fast-Tracking Hand Gesture Recognition AI Applications with Pretrained Models from NGC
- Building and Deploying Conversational AI Models Using the NVIDIA TAO Toolkit
- Implementing a Real-time, AI-Based, Face Mask Detector Application for COVID-19
- Improving INT8 accuracy using Quantization Aware Training and NVIDIA TAO Toolkit
- Training Instance Segmentation Models using MaskRCNN on NVIDIA TAO Toolkit
- Building Intelligent Video Analytics Apps Using NVIDIA DeepStream 5.0
- Using the NVIDIA Isaac SDK Object Detection Pipeline with Docker and the NVIDIA TAO Toolkit
- How To Train with Custom Pre-Trained Models using TAO Toolkit
- What is transfer learning?
- Build and Deploy Accurate Deep Learning Models for Intelligent Image and Video Analytics
- Pruning models with NVIDIA TAO Toolkit
Community Projects
- Take a look at our innovative developer community projects and submit your cool project to be featured on our community project hub
Webinars
- GTC 2021: Train Smarter, not Harder, with NVIDIA Pre-Trained Models and TAO Toolkit 3.0
- GTC 2021: Building and Deploying a Custom Conversational AI App with NVIDIA TAO Toolkit and Riva
- Using NVIDIA Pre-trained Models and TAO Toolkit 3.0 to Create Gesture-Based Interaction With a Robot.
- GTC 2020: Accelerating Vision AI Applications Using NVIDIA TAO Toolkit and Pre-Trained Models
- GTC 2020: NVIDIA Tools to Train, Build, and Deploy Intelligent Vision Applications at the Edge
- Accelerating deep learning training using NVIDIA TAO Toolkit
- Real-time object detection for disaster response
- DeepStream SDK- Accelerating Real-Time AI Based Video And Image Analytics
Free Self-Paced DLI Online Courses
- Learn how to build end-to-end intelligent video analytics pipelines using DeepStream and Jetson Nano >> Enroll now
- Learn how to get started with AI using Jetson Nano >> Enroll now
Ethical AI
NVIDIA’s platforms and application frameworks enable developers to build a wide array of AI applications. Consider potential algorithmic bias when choosing or creating the models being deployed. Work with the model’s developer to ensure that it meets the requirements for the relevant industry and use case; that the necessary instruction and documentation are provided to understand error rates, confidence intervals, and results; and that the model is being used under the conditions and in the manner intended.