ao link

You are viewing 1 of 1 articles without an email address.

All our articles are free to read, but complete your details for free access to full site!

Already a Member?
Login Join us now

AI algorithm capable of multi-task deep learning

DeepManta’s architecture and enhancements to AI extract different types of information simultaneously and in real time

Visual object recognition for smart cities will be one of the applications demonstrated
Visual object recognition for smart cities will be one of the applications demonstrated

List, a research institute of CEA Tech, has developed a category of artificial intelligence (AI) algorithm capable of extracting different types and levels of information simultaneously and in real-time.


According to List, its DeepManta “multi-task deep neural-network” algorithm is suitable for a wide range of applications including visual object recognition for smart cities, such as identifying vehicles, their type and position and counting them.


At CES 2018 in Las Vegas, List will be partnering with Valeo, a global supplier of advanced automotive technology, to demonstrate DeepManta’s support for autonomous driving.


The demonstration will include a video stream captured by a stationary camera and displayed live on a screen. Different objects, such as miniature cars, move into the camera’s field of view, where the AI will selectively recognise them.


When a car is recognised, the algorithm generates a visual annotation, labelling the car with the logo of the brand and model information, and encompassing it with 2D and 3D boxes to locate it spatially in the video in real time.


“DeepManta delivers one of the promises of AI: providing assistance to users by automatising and parallelising tasks that normally would require their full attention,” said Stéphane David, industrial partnership manager at List.


“It excels at each individual task, but requires much less overall memory and processing power than parallel architectures that use one algorithm per task.”


The result of more than 10 years of research, DeepManta is a flexible algorithm developed to perform advanced and efficient real-time analysis of video streams. The native multi-task architecture combined with enhancements to conventional deep-learning algorithms powers a system capable of extracting different types and levels of information simultaneously and in real-time.


The system uses a standard video camera connected to a laptop equipped with a powerful GPU. The video feed is processed by the algorithm running on the laptop and the result of the analysis, including incrustations, is broadcast with a very low latency onto the screen. This provides an efficient and autonomous system with all the necessary resources performing live.


In addition to automotive applications, the algorithm’s automated perception capacity opens new services with significant social and business impacts. These range from guidance for blind people to video surveillance or aspect control of products on manufacturing lines, said List.


List, an institute of CEA Tech, the CEA Technological Research Division, carries out research on smart digital systems. Working on major economic and social challenges, its R&D programmes focus on advanced manufacturing, embedded systems, data intelligence and digital patient applications.


CEA Tech is the technology research branch of the French Alternative Energies and Atomic Energy Commission (CEA).


If you like this, you might be interested in reading the following:


Key technology trend predictions for 2018

Predictions directly align with the society’s commitment to keeping its community well-informed and prepared for the changing technological landscape

Read more


Artificial intelligence at the edge

NVIDIA Jetson TX1 Supercomputer module to power neuBox with visual computing and deep learning capabilities

Read more


Electric car powertrain partnership formed

A joint venture between Siemens and Valeo aims to provide affordable components and systems for the electric car mass market

Read more

Add New Comment
You must be a member if you wish to add a comment - why not join for free - it takes just 60 seconds!