Artificial intelligence is taking the automotive industry by storm. For example, AI-powered robotaxis are expected to create a $2 trillion market worldwide by 2030, according to global financial services firm UBS.

This indicates one of the most prominent ways AI is revolutionizing the industry: the development of autonomous driving technology. Self-driving cars use camera-based machine vision systems and radar- and lidar-based detection units to perceive, understand, and safely navigate their surrounding environment.

In this workshop, you’ll learn how to optimize performance of self-driving car perception applications such as lane navigation and pedestrian detection. You’ll learn to build and train a semantic segmentation neural network to identify objects such as roads, pedestrians, and other vehicles. Then, you’ll deploy the neural network on the NVIDIA DRIVE AGX platform to power autonomous navigation of a vehicle. Upon completion of the workshop, you’ll have the skills needed to build AI applications for a variety of autonomous driving scenarios, including highway driving, city roads, and parking.

Learning Objectives

By participating in this workshop, you’ll learn how to:
  • Run example code using different GPU memory configurations on the DRIVE AGX platform to determine which is most effective in a variety of use cases
  • Compare multiple methods of performance optimization using CUDA® and timed-inference test cases on the DRIVE AGX platform
  • Create and train a semantic segmentation network to understand automotive scenes by combining a fully convolutional network (FCN) “head” with a MobileNets convolutional neural network (CNN) “stem” 
  • Train a semantic segmentation model with the NVIDIA DIGITS tool using Cityscapes data to demonstrate pixel-wise semantic segmentation of scenes
  • Convert Keras and Tensorflow semantic segmentation models into optimized NVIDIA® TensorRT models that can be deployed for practical applications on the DRIVE AGX platform
  • Deploy and run an optimized TensorRT model on the DRIVE AGX platform to demonstrate the development and deployment workflow for DRIVE applications

Download workshop datasheet (PDF 299 KB)

Workshop Outline

(15 mins)
  • Meet the instructor.
  • Create an account at
(120 mins)
    Learn techniques to improve GPU performance for DRIVE applications through memory management and optimization techniques:
    • Test and compare performance tradeoffs for conventional memory, pinned memory, and unified memory types.
    • Explore mixed-precision computation optimizations in inference.
    • Optimize performance using CUDA streams and load-balancing techniques.
Break (60 mins)
Training Semantic Segmentation for DRIVE
(120 mins)
    Explore how to build and train an FCN for semantic segmentation and deploy it to analyze automotive scenes:
    • Build a semantic segmentation FCN based on a conventional CNN in TensorFlow.
    • Prepare a Cityscapes dataset to train an FCN using DIGITS.
    • Train an FCN in DIGITS and test it using inference to observe the resulting pixel-level semantic segmentation of the scenes.
Break (15 mins)
Deployment of a Semantic Segmentation Network Using TensorRT
(120 mins)
    Learn the TensorRT development workflow for an autonomous driving semantic segmentation use case:
    • Optimize a pre-trained semantic segmentation model built with Keras to TensorRT for an embedded system.
    • Test and compare performance and accuracy across the Keras implementation, TensorRT FP32, and TensorRT INT8.
    • Build a calibration dataset and deploy the model to the embedded target system for additional performance comparisons.
Assessment and Q&A (15 mins)

Workshop Details

Duration: 8 hours

Price: Contact us for pricing.  


  • Familiarity with C++ and Python
  • Experience with convolutional neural networks (CNNs)

Suggested materials to satisfy prerequisites:  C++ Tutorial, Python Tutorial, Convolutional Neural Network

Technologies: TensorFlow, Keras, NVIDIA TensorRT, CUDA C++, Python, NVIDIA DIGITS, 

Certificate: Upon successful completion of the assessment, participants will receive an NVIDIA DLI certificate to recognize their subject matter competency and support professional career growth.

Hardware Requirements: Desktop or laptop computer capable of running the latest version of Chrome or Firefox. Each participant will be provided with dedicated access to a fully configured, GPU-accelerated server in the cloud.

Languages: English, Simplified Chinese

Upcoming Workshops

If your organization is interested in boosting and developing key skills in AI, accelerated data science, or accelerated computing, you can request instructor-led training from the NVIDIA DLI.