Exploring the

Possibilities
of Nvidia at
the Edge with
Edge Impulse
edgeimpulse.com
 Exploring the Possibilities of Nvidia at the Edge with Edge Impulse 3

Key Features and Capabilities of NVIDIA Metropolis 3

Leveraging Edge Impulse with Nvidia 6

Sample Hardware Target: Advantech ICAM-540 7

Unlocking value at the Edge 9

Endless Possibilities 10

Conclusion 11

2 edgeimpulse.com
EXPLORING THE POSSIBILITIES
OF NVIDIA AT THE EDGE WITH
EDGE IMPULSE
Nvidia, the leading global provider of AI hardware and software solutions, is driving innovation in
artificial intelligence from the cloud to edge computing. Edge AI is set to expand into new markets
and environments due to advantages such as local processing resulting in decreased inference
times and higher performance, enhanced privacy, and reduced data and bandwidth costs.

The partnership between Nvidia and Edge Impulse bridges the gap from cloud to edge,
enabling the seamless deployment of AI technologies to edge devices and unlocking new
possibilities by helping product engineers throughout the lifecycle from prototyping, to
deployed solutions.

Nvidia has a suite of hardware and software offerings that are both deep and wide across a
variety of AI domains. Many of their core computer vision, audio, and sensor AI frameworks
are contained in their Metropolis product.

Nvidia Metropolis is a comprehensive platform designed to integrate visual data and artificial
intelligence to enhance operational efficiency and safety across various industries. It consists
of an application framework, developer tools, and a partner ecosystem that collectively
facilitate the development, deployment, and scaling of vision AI and generative AI solutions.

KEY FEATURES AND

CAPABILITIES OF
NVIDIA METROPOLIS
1. Application Framework
Metropolis provides a robust application framework that simplifies the creation and deployment
of AI-powered solutions. This framework offers:

Pretrained Models: Access to pretrained AI models for common tasks such as object
detection, facial recognition, and anomaly detection.

Custom Model Training: Tools to train custom AI models tailored to specific use cases.

Model Deployment: Seamless deployment of AI models across edge devices and cloud
infrastructure.

3 edgeimpulse.com
2. Developer Tools
Metropolis includes a comprehensive set of developer tools designed to make AI accessible
to developers of all skill levels:

NVIDIA TAO: A toolkit that simplifies model training, pruning, and optimization, enabling
developers to fine-tune AI models efficiently.

Image cropped from nvidia.com

DeepStream SDK: A software development kit that facilitates the creation of AI-powered
video analytics applications. It supports a wide range of use cases from smart cities to
industrial inspection.

Image cropped from nvidia.com

4 edgeimpulse.com
 Omniverse Replicator: A synthetic data generation tool that creates realistic training data
for AI models, enhancing model accuracy and robustness.

Image cropped from nvidia.com

3. Partner Ecosystem
The Metropolis partner ecosystem includes hardware manufacturers, software developers,
and system integrators, providing a wide range of compatible solutions:

Hardware Integration: Works with a variety of cameras, sensors, and edge devices,
ensuring flexibility in hardware choices.

Software Solutions: Collaborations with software partners to offer end-to-end solutions,

from data capture to AI model deployment.

System Integrators: Partnerships with system integrators to deliver customized AI

solutions tailored to specific industry needs.

Edge Impulse has partnered with Nvidia to simplify and optimize the deployment to edge
devices, from Jetson Developer Kits and Jetson Partner hardware, to resource constrained
devices like microcontrollers. This streamlines your interaction with key Nvidia products into
a single interface provided by Edge Impulse. System Integrators and software engineers can
shorten development cycles and learning curves, and enhance focus on their customer’s
needs instead.

5 edgeimpulse.com
LEVERAGING EDGE IMPULSE
WITH NVIDIA

Edge Impulse is a leading development platform for edge AI, enabling developers to build, train,
and deploy machine learning models on a variety of edge devices. When combined with Nvidia
Metropolis, Edge Impulse enhances the development workflow by providing:

Ease of Use: With a user-friendly interface, Edge Impulse allows developers of all skill levels to
create sophisticated AI models without extensive expertise in machine learning.

Seamless Integration: The platform supports integration with Nvidia’s AI tools, enabling developers
to leverage the full power of Metropolis, including Omniverse Replicator, DeepStream, and TAO.

Real-World Applications: Edge Impulse facilitates the deployment of AI models in real-world

scenarios, making it possible to implement solutions in industries such as manufacturing,
healthcare, and transportation.

Edge Impulse has built integrations to augment Metropolis for edge use cases, as
shown below:

Image cropped from nvidia.com

6 edgeimpulse.com
Each of the blocks represented above contain a step in the machine learning pipeline. Let’s
look at them in detail.

Create: Developers can leverage Edge Impulse and Nvidia Omniverse Replicator to reduce
prototyping costs, speed up development, and create robust AI models with synthetic data
generation and model validation before deploying to physical hardware. This approach
enables rapid experimentation, efficient resource use, and the creation of diverse synthetic
data scenarios that enhance model reliability and performance across various applications,
such as smart cities, retail, and manufacturing. By validating models in virtual environments,
developers can ensure their solutions are ready for real-world deployment, saving time and
resources while improving overall model robustness. The synthetic data can simplify and
speed up the task of creating models that generalize well, all using an intuitive GUI interface.

Train: Edge Impulse has integrated Nvidia TAO directly into the Studio, providing model builders
with a selection of pre-trained Object Detection and Classification Models for computer vision
applications. Developers use their collected dataset and perform transfer learning on the
TAO models through the Studio GUI interface, which exposes model hyperparemeters for
easy customization and experimentation. It’s even possible to train a classification model by
distilling knowledge from ChatGPT-4o.

Build: Nvidia DeepStream enables the seamless creation of interactive video analytics, which
are applications that process multimedia data and use AI in the form of Object Detection
or Image and Audio Classification to derive real time insights from multiple video streams.
DeepStream enables additional functionality such as object tracking, support for multiple
streams, accelerated encoding and decoding and Triton Inference server integration.
Models trained in Edge Impulse can be seamlessly integrated into DeepStream, so that you
can leverage the simple and productive MLOps process provided by Edge Impulse in your
IVA applications.

Deploy: Finally, Edge Impulse supports deployment of models to the range of Nvidia Developer
Kits and Partner devices with built-in Jetson modules, by means of the Nvidia TensorRT
deployment or through conversion to DeepStream. Edge Impulse has developed the leading
tinyML optimization compiler that even allows for deployment of models that have been
fine-tuned with the TAO integration to high-end microcontrollers, opening up a whole host
of new applications.

SAMPLE HARDWARE TARGET:

ADVANTECH ICAM-540
The edge domain has expanded over the years, now spanning a range of devices from powerful
GPU-based devices such as the Nvidia Jetson range to general-purpose devices such as the
Raspberry Pi, even down to microcontrollers. Edge Impulse provides optimizations to ensure
seamless deployment to both accelerated and non-accelerated devices, with use-cases
varying in terms of camera resolution and latency needs.

To fully realize the potential of edge AI, robust and reliable hardware is essential. Advantech
is a key partner that provides cutting-edge hardware solutions tailored for AI applications
at the edge.

7 edgeimpulse.com
Advantech ICAM: The ICAM series from Advantech offers intelligent cameras equipped with
Nvidia’s AI capabilities. These cameras are designed for a wide range of applications, including
smart surveillance, industrial automation, and retail analytics. By integrating with Nvidia
Metropolis and Edge Impulse, the Advantech iCAM-540 enables real-time data processing
and actionable insights.

Image cropped from advantech.com

The ICAM 540 features a high framerate (45FPS) industrial grade IMX334l image sensor
with either integrated optics and illumination, or a model C-mount lens support, housed in
a rugged industrial case. Internally the device contains an Nvidia Orin NX module, featuring
up to 100 TOPS of AI performance. Connectivity is via familiar M12 connectors aligned with
Industrial Automation standards and there are a variety of interfaces, from Ethernet to RS485
and even GPIO, meaning this device is ready for the factory floor.

8 edgeimpulse.com
UNLOCKING VALUE
AT THE EDGE
By combining Nvidia, Edge Impulse, and hardware targets like the Advantech ICAM, it is
possible to unlock applications that add value while reducing tooling costs and complexities
associated with other methods.

Industrial Vision Inspection: Traditionally the field of unattended Vision Inspection systems in
Industrial automation has been dominated by classical machine vision systems. More recent
deep learning-based systems are extremely expensive and include proprietary software, and
in many cases proprietary hardware as well. These systems come with higher support costs
due to skill specialization, and costly support level agreements. Advantech, combined with
Edge Impulse, disrupts this space by reducing the cost, complexity, and long-term TCO while
bringing the power of Nvidia’s accelerators to the heart of the factory floor. The ICAM can
easily integrate into existing PLC and even SCADA systems, allowing organizations to leverage
their existing control and automation systems.

Inspection can be done using simple object detection, or by combining object detection with
TAO and DeepStream pipelines that include object detection, classification and tracking in real-
time on production lines made up of conveyor systems, rotating machinery, or even vehicles.

Types of applications include but are not limited to defect detection, sorting glass, packaging
quality control, high speed printing, pharmaceutical package verification, and more. Wherever
traditional inspection systems have already been deployed, the ICAM offers a low cost, high
performance alternative.

Further use-cases in retail settings, such as people counters, grocery checkout monitoring,
or inventory detection and counting are also good candidates for ICAM deployments.
Due to the high framerate provided by the camera, it can handle these kinds of applications
with ease.

9 edgeimpulse.com
Image cropped from nvidia.com

Whatever the use-case, the Advantech ICAM series combined with Edge Impulse can reduce
costs and improve business efficiencies, while obtaining a faster ROI vs traditional solutions.

ENDLESS POSSIBILITIES
Being able to combine synthetic data generation with the power of transfer learning using
TAO pretrained models, while having maximum flexibility in deployment options from Nvidia
GPUs to MCUs, ensures that as many use-cases as possible can be covered. Additionally,
the streamlined workflow and MLOps environment of Edge Impulse saves developer time by
eliminating the need to switch between tools and environments.

10 edgeimpulse.com
This comes without needing to sacrifice any of the benefits provided by making use of Nvidia’s
pretrained models and technology.

CONCLUSION
The convergence of Nvidia’s hardware and software infrastructure, Edge Impulse’s
development tools, and advanced hardware from Advantech are driving the next wave of
innovation in edge AI. By leveraging the capabilities of Omniverse Replicator, DeepStream,
TAO, and powerful hardware solutions, developers can create and deploy sophisticated AI
applications that transform industries and improve everyday life. Get started now, or Contact
Sales if you need assistance with building your next application.

11 edgeimpulse.com
Thank You!
Exploring the Possibilities of Nvidia
at the Edge with Edge Impulse

For More Information

Contact Us

edgeimpulse.com