Training Example: Ambarella – Review the Data, Give Your Score & Compare to the Real AI Evaluation

[H1] AI Vision Processors For Edge Applications
Our solutions make cameras smarter by extracting valuable data from high-resolution video streams.

Explore our key applications

AutomotiveSecurityConsumerIoT, Industrial & Robotics
Explore our Developer Zone

People
Signs
Objects
Vehicles
Environment & depth

Replay

360-degree object detection
At long range
And short range

[H2] From ADAS to Autonomy
With advanced driver assistance systems, smart electronic mirrors, drive recorders, autonomous vehicles and more, Ambarella is committed to preparing you for the road ahead.

Discover now

Replay

[H2] Smart Surveillance Solutions
Our intelligent video devices use face detection, analytics, and multi-object CNN classification to enhance security in businesses and homes, without the need for the cloud.

Discover now

Replay

[H2] News & Events

May
28
[H3] Ambarella, Inc. Announces First Quarter Fiscal Year 2027 Financial Results

May
21
[H3] Ambarella Announces Inaugural Quarterly Briefing Call Targeting Non-Financial Industry Research Analysts

Feb
26
[H3] Ambarella, Inc. Announces Fourth Quarter and Fiscal Year 2026 Financial Results

[H2]
[IMG: CVflow]
Our CVflow® chip architecture is based on a deep understanding of core computer vision algorithms. Unlike general-purpose CPUs and GPUs, CVflow includes a dedicated vision processing engine programmed with a high-level algorithm description, allowing our architecture to scale performance to trillions of operations per second with extremely low power consumption.
With CVflow, customers can efficiently map their own CNN networks trained with industry-standard tools (e.g., Caffe, TensorFlow, PyTorch) to run on Ambarella processors.

See how CVflow works

[H2] How CVflow works

Trained Neural Network
Our customers provide their own trained neural networks in popular formats including Caffe, TensorFlow, and ONNX.

Sparsification and Quantization
Using our CNN Generation tool, we analyze the trained neural network and optimize it for our hardware via sparsification and quantization.

Conversion to High-Level Language
The optimized network moves to our compiler tool, where we generate a program – written in a high-level language rather than generic low-level operators – for execution on our CVflow hardware.

Compiled DAG Binary
The compiled program, known as a DAG executable binary, is sent to our CVflow hardware for execution. The result is validated and the network is retrained, as needed, with our customers’ support.

Play animation

Close

[H2] Stereovision

Stereo processing transforms the world from a flat 2D plane into a full 3D environment. This not only provides a richer, denser representation of the target scene, but also allows the detection of generic obstacles that the system hasn’t been trained to recognize, making navigation safer and more effective for warehouse robots, autonomous vehicles, and more.

Replay

[H2] Low power
Our technology is designed to be deployed in embedded platforms on the edge (e.g., battery-powered cameras), where power consumption is a critical concern. Ambarella processors consistently outperform competitive solutions in power efficiency—sometimes by 5x or more—while delivering equal or superior results.

Replay

[H2] Compression efficiency
Cameras are quickly becoming ubiquitous—with multiple devices in our vehicles, homes, workplaces, and intersections—generating massive amounts of data.

[H2] Reduce bandwidth & storage costs
Our compression technologies dramatically reduce data transmission size, slashing storage costs and bandwidth usage.

Replay

[H2] Image quality
One of the hallmarks of Ambarella processors is our image processing pipeline. Encompassing everything from HDR to EIS to dewarping, our state-of-the-art ISP ensures that every frame of every video is exquisite, improving safety and enhancing the viewing experience.

Replay

[H2] Functional Safety

As an AI silicon company offering SoCs for applications such as ADAS and autonomous driving, Ambarella has implemented an array of internal processes and procedures to ensure that our functional safety-certified products comply with ISO 26262.
These strict processes and procedures touch every aspect of the work we do—from technical documentation to software engineering to VLSI—and have become an inextricable part of our culture, from top to bottom.

[H2] Computer vision

Present day
Computer vision
Our vision processors were designed using more than twenty-five years of pioneering research in environmental perception. The result? Cameras with an extraordinary degree of onboard intelligence, including the ability to classify objects, interpret gestures, recognize faces, track people of interest, identify obstacles, and more.

1998
Computer vision
Our vision processors were designed using more than twenty-five years of pioneering research in environmental perception. The result? Cameras with an extraordinary degree of onboard intelligence, including the ability to classify objects, interpret gestures, recognize faces, track people of interest, identify obstacles, and more.

[H2] Autonomy

From drones to robots, from taxis to long-haul semi trucks, Ambarella’s vision processors were designed to revolutionize the way we work and travel. Our hardware platforms deliver the flexibility required to implement all levels of automation, including partial, conditional, high, and full autonomy.

2013 footage of an L4 autonomous driving demonstration in Parma, Italy

[H2] SDK & Reference Designs

Our company was built on transforming core video processing technologies into mass-market products. We offer tools, reference platforms, and a team of dedicated engineers with a proven track record for developing best-selling consumer and professional devices.

Replay

[close]

[H2] Technology

[H2] Technology in Action

[H3] CV3-AD Launch Video
Introducing the new CV3-AD AI domain controller SoC family

[IMG: AD Stack Introduction]

[H3] AD Stack Introduction
Learn about our full software stack for autonomous and semi-autonomous driving

[H3] Central 4D Imaging Radar Architecture
Overview of our centrally processed 4D imaging radar architecture for autonomous mobility systems

[H3] Ambarella + Oculii™ Announcement
Ambarella acquires Oculii, the radar perception AI algorithm leader

[H3] Oculii™ AI Radar Technology
Explainer on our breakthrough Oculii adaptive AI radar software

[IMG: Exterior Sensing Overview]

[H3] Exterior Sensing Overview
Learn more about our advantages in L2+ and front ADAS applications.

[IMG: Interior Sensing Overview]

[H3] Interior Sensing Overview
A short animated video about our interior sensing solutions.

[IMG: CVflow® Examples]

[H3] CVflow® Examples
Sample clips from our CVflow platform: tracking, segmentation, depth, and more.

[IMG: EVA Stereo Camera Perception]

[H3] EVA Stereo Camera Perception
Actual short- and long-range camera outputs, including the resulting dense point cloud.

[IMG: CVflow® Overview]

[H3] CVflow® Overview
Learn more about our CVflow® advanced computer vision architecture.

[IMG: EVA in Italy]

[H3] EVA in Italy
Our EVA autonomous vehicle on the famous Lingotto rooftop test track in Turin.

[IMG: Embedded Vehicle Autonomy]

[H3] Embedded Vehicle Autonomy
Our EVA autonomous vehicle perception platform in action.

[IMG: Smart Electronic Mirrors]

[H3] Smart Electronic Mirrors
Learn more about the benefits of electronic mirrors.

[IMG: Driver Monitoring System (DMS) in Action]

[H3] Driver Monitoring System (DMS) in Action
A demonstration of Driver Sense by Eyesight Technologies, an Ambarella partner.

[IMG: Stereo Camera Performance]

[H3] Stereo Camera Performance
Long- and short-range stereo performance, both day and night.

[IMG: Autonomous Vehicle Timeline]

[H3] Autonomous Vehicle Timeline
A brief history of Ambarella’s autonomous vehicle technologies (1998-2016).

[IMG: Driverless Car Ride - 2013]

[H3] Driverless Car Ride – 2013
A car ride from Parma University to Parma downtown, with no human intervention.

[IMG: From Italy to Shanghai]

[H3] From Italy to Shanghai
Four autonomous vehicles left Italy on July 26, 2010 and arrived in Shanghai on Oct 28, 2010.

[IMG: The Surround View Perspective]

[H3] The Surround View Perspective
See your vehicle from any angle with surround view (AVM), increasing safety and security.

[IMG: Wide Field of View]

[H3] Wide Field of View
Get a clear picture of the lanes on either side, thanks to on-camera dewarping.

[IMG: Spot A Roadway Accident]

[H3] Spot A Roadway Accident
Capture detail such as roadway debris, driver images, and license plate numbers with ease.

[IMG: A Leisurely Drive in Ultra-HD]

[H3] A Leisurely Drive in Ultra-HD
Get behind the wheel for a pleasant ride. (UHD available.)

[H2] Automotive Applications

[H3] Front ADAS Cameras

Front ADAS Cameras

This forward-facing camera is positioned behind the rearview mirror, enabling functions such as automatic emergency braking (AEB), lane departure warning (LDW), forward collision warning (FCW), and more. Front ADAS cameras require rapid computer vision processing at high resolutions and extremely low power (due to an inherently small form factor), all of which are critical for long-distance object detection with a wide field-of-view (for functions such as cross-traffic detection, required by Euro NCAP), making Ambarella solutions ideally suited for the task.

[IMG: Driver Monitoring Systems and In-Cabin Monitoring]

[H3] Driver Monitoring Systems and In-Cabin Monitoring

Driver Monitoring Systems and In-Cabin Monitoring

DMS and in-cabin monitoring systems allow the tracking of driver alertness as well as passenger-related data that can assist with the deployment of safety features (e.g., airbags). Ambarella’s hardware processes our customers’ interior-sensing algorithms at high speeds and with low power consumption, even at night (via onboard RGB-IR processing), which results in exceptional performance without compromising efficiency. Our DMS solutions can even be integrated with complementary camera applications, including electronic mirror, front-camera ADAS, high-resolution recording, and more.

[H3] Smart Electronic Mirrors

Smart Electronic Mirrors

Ambarella’s rearview and side-view electronic mirror solutions enhance the visibility of traffic, filter out distracting glare and LED flickering from headlights, and offer intelligent safety features such as Blindspot Monitoring and Overtaking Vehicle Detection – and we do it all with ultra-low latency, delivering the same fluid viewing experience provided by optical mirrors.

[H3] 360° Surround View Systems with Parking Assist

360° Surround View Systems with Parking Assist

Ambarella’s 360-degree surround view solution uses multiple cameras to present a single, unified image of the vehicle from a top-down, bird’s eye view. Offering both 2D and 3D operations, intelligent parking assist features, as well as recording and streaming capabilities, Ambarella’s four-channel surround view systems provide drivers with unobstructed, distortion-free views of the vehicle from every side.

[H3] Partial to Fully Autonomous Solutions

Partial to Fully Autonomous Solutions

Ambarella’s autonomous technologies have performed successfully in a variety of demanding scenarios, including the DARPA Grand Challenge (2005), a trip from Italy to China in full autonomous mode (2010), and driverless navigation at highway speeds (2015). Today Ambarella continues to advance its research in critical areas of autonomous vehicle development, such as vehicle detection, obstacle detection, pedestrian detection, lane detection, traffic sign recognition, stereovision processing, and terrain mapping, as well as issues related to commercialization and mass production, enabling us to design optimal platforms for applications ranging from L2+ autopilot to full autonomy.

[H3] Automotive Video Recorders

Automotive Video Recorders

When the primary goal is to capture the highest-quality video—whether to test vehicle performance, to gather critical evidence, or purely for the enjoyment of the road—Ambarella is an established leader in the field, offering both aftermarket and OEM drive recording solutions. We offer single- and multi-channel recorders with automotive-level certification and optional front ADAS features.

[H3] Fleet Management Solutions

Fleet Management Solutions

Building on our proven success in powering aftermarket and OEM dash cameras (over 20 million units sold to date), Ambarella has become a leader in AI-based fleet management hardware, offering advanced automotive-grade processors designed to evaluate risky driving behaviors while delivering high-quality, recordable imagery.

[H2] View Our Selection of Automotive AI SoC and Radar Solutions

See All Automotive Products

[H2] Automotive Software Partners

[H2] Automotive Milestones

1998

Traversed 1,240 miles on Italian highways in autonomous mode
(obstacle and lane detection, stereo matching)

2005

Successfully completed the DARPA Grand Challenge – a 132-mile off-road course in driverless mode
(trinocular system, edge-based stereo processing)

2010

Drove autonomously from Italy to China – over 8,000 miles
(obstacle detection and terrain mapping with dense disparity map)

2011

Established a leadership position in the aftermarket dash camera segment

2013

Performed complex urban driving in L4 autonomous mode
(downtown driving in real traffic conditions)

2014

Launched our first 4-channel automotive system

2015

Demonstrated driverless navigation at highway speeds
(360° multi-sensor coverage fusing stereovision and range sensors)

2016

Featured in Toyota and Honda OEM drive recording cameras

2016

Entered the electronic mirror market in cooperation with Gentex Corporation

2018

Unveiled EVA – our stereovision-based, self-driving demonstration vehicle with the entire perception stack running on our CV chips

2019

Announced partnerships with HELLA Aglaia (top-50 global supplier) and Smart Eye (leader in Driver Monitoring Systems and In-Cabin Monitoring)

2019

Announced partnerships with Momenta (for the development of an open HD map platform) and OFilm (for the development of an advanced automated valet parking platform)

2020

Unveiled multiple AI chips with ASIL-B compliance to enable safety-critical applications

2021

Acquired Oculii™, the radar perception AI algorithm global leader

2022

Launched world’s first centrally processed 4D imaging radar architecture for autonomous mobility systems

2023

Unveiled full software stack for autonomous and semi-autonomous driving

[H2] Latest posts

[H2] The AI That Runs the Physical World Looks Nothing Like Your Favorite Chatbot

AIoT, Industrial & RoboticsAutomotiveConsumerSecurity
Apr 27, 2026

[H2] Next-Gen Fleet Telematics and Dashcams Shift to On-Device AI

Automotive
Aug 28, 2025

[H2] Achieving High-Speed AEB with AI-Driven 4D Imaging Radar

Automotive
Jun 30, 2025

See automotive posts

[close]

[H2] Technology

[H2] Technology in Action

[IMG: Smart Security in the Home]

[H3] Smart Security in the Home
An animated overview of our consumer security approach.

[IMG: Intelligent Professional Security]

[H3] Intelligent Professional Security
Our vision for smart retail surveillance.

[IMG: Traffic Camera With Computer Vision]

[H3] Traffic Camera With Computer Vision
Traffic camera showing vehicle, person, and sign detection.

[IMG: Security on the Roadway]

[H3] Security on the Roadway
Our CV22 chip performing vehicle make, model, and license plate recognition.

[IMG: Record Critical Details]

[H3] Record Critical Details
Security cam footage with enough detail to read signs, license plates, even facial expressions.

[H2] Security Applications

[H3] Indoor/Outdoor Security

Indoor/Outdoor Security

Our powered indoor and outdoor solutions offer complex person/vehicle detection and face recognition on the edge, surpassing the performance of most GPU-based systems—all at a fraction of the power budget.

[H3] Multi-Sensor/Multi-Directional Security Cameras

Multi-Sensor/Multi-Directional Security Cameras

Combining multiple sensors in a single casing, our multisensor cameras deliver 360° coverage of the target environment at extremely high resolutions, all using a single-chip design.

[IMG: ITS & Traffic Cameras with Computer Vision]

[H3] ITS & Traffic Cameras with Computer Vision

ITS & Traffic Cameras with Computer Vision

Our traffic cameras and intelligent transportation systems (ITS) solutions are designed to enhance roadway security with a combination of person and vehicle detection, vehicle make/model recognition, and automatic number plate recognition (ANPR) capabilities.

[IMG: Retail Cameras with Onboard Analytics]

[H3] Retail Cameras with Onboard Analytics

Retail Cameras with Onboard Analytics

In-store cameras are useful for more than just security—they offer retailers the opportunity to analyze customer behavior, traffic patterns, and areas of high or low interest. Ambarella chips combine cutting-edge surveillance features with next-generation analytics capabilities, enabling better inventory control, product placement, customer tracking, ad campaign targeting, and staffing adjustments—without relying on the cloud.

[IMG: Powered Indoor AI Cameras]

[H3] Powered Indoor AI Cameras

Powered Indoor AI Cameras

Ideal for home use, our smart AI cameras offer person/pet detection, face detection and recognition, and fall detection (e.g., for elder care), delivering a security system tailor-made for the individual consumer.

[IMG: Low-power Wire-Free Cameras with Intelligent Alerts]

[H3] Low-Power Wire-Free Cameras with Intelligent Alerts

Low-Power Wire-Free Cameras with Intelligent Alerts

Delivering reliable person detection and person/pet/vehicle classification, our wire-free cameras provide the alerts consumers need most, without pestering them with false alarms.  And because we offer a Quick Motion Detect feature, our solutions don’t drain the camera’s battery with lengthy (and unnecessary) wakeups.

[IMG: Smart Video Doorbells]

[H3] Smart Video Doorbells

Smart Video Doorbells

Our doorbell solutions deliver person detection, face detection and recognition (including “familiar/unfamiliar face” alerts), and package detection on the edge, offering low-latency performance without the cost and lag-time of the cloud.

[H2] View Our Selection of Security SoCs

See All Security Products

[H2] Latest posts

[H2] The AI That Runs the Physical World Looks Nothing Like Your Favorite Chatbot

AIoT, Industrial & RoboticsAutomotiveConsumerSecurity
Apr 27, 2026

[H2] A Smarter Vision for AI at ISC West 2026

CorporateSecurity
Mar 18, 2026

[H2] Looking Under the Hood of AI Image Enhancement Technologies

AIoT, Industrial & RoboticsConsumerSecurity
Oct 29, 2025

All security posts