Now Reading
How Unreal AI Is Using Proprietary Algorithms To Turn Videos Into Usable Data

How Unreal AI Is Using Proprietary Algorithms To Turn Videos Into Usable Data

This New Delhi-based startup, founded in 2018 by Saurabh Singh and Nischal Gaba aims to decentralise artificial intelligence (AI) on Google and AWS cloud by developing proprietary AI neural network model that can run on low-cost edge devices.

Register for FREE Workshop on Data Engineering>>

According to the founders, the pressing challenge before them was to run machine learning models on low-cost edge devices so as to reduce their dependence on the cloud for intelligence.

After a year of hard work and deep research, the youngsters were able to address the challenge and when they showcased it to their friends and told them that they were able to run the models using $5 Raspberry Pi, their friends found it ‘unreal’.

“AI on the edge is the next frontier in AI.  Just like with the advent of PCs in 70s decentralised computing power from mainframe computers we believe that our proprietary AI methods can help decentralise AI and make every device intelligent in itself,” Saurabh Singh, Co-Founder of Unreal AI said when asked about the motivation behind establishing the startup.

Taxing ML  models

Prior to the genesis of Unreal AI, the pair was working on an AI and NLP powered-chatbot that could answer queries by searching for words or phrases within videos, “For example, you could ask “When was this taught in this video lecture?” and the chatbot will pinpoint the location inside the video when it happened,” says Singh.

However, the duo later discovered that their machine learning models for extracting information  were computationally taxing and soon they were looking at means to make their ML models run faster, “We realised that it’s a fundamental problem with ML that it requires a huge amount of computation power for real-time application,” Singh explains

It is this quest to make their models faster and efficient that led them to the problem statement “Is there a way to run deep learning models on the low-cost devices/edge devices?” and finally to the creation of Unreal AI.

Building a sleek and efficient system

Currently, the startup’s unique product offering is an easy to install security device that uses face recognition to authenticate employees at locations such as offices and schools that uses deep machine learning models to achieve this.

The product uses face recognition for authentication and for added transparency. It also takes a snapshot of the moment when the person was authenticated which can later be reviewed for close inspection. The speed of the device also enables the device to identify as many as seven people at a time. This added layer of transparency and speed is a bonus for the device claims Singh and adds that it sets the product apart from conventional authentication systems which can identify just one face at a time.

See Also

“Vizor ID uses a pipeline of multiple CNNs that have been implemented using our proprietary methods which enable them, otherwise computationally expensive, deep learning models to run on the low-cost devices,” explains Singh who depended on open source frameworks like Caffe, Keras, Pytorch etc to design the system.

However, he does point out that the initial challenge before them was to design a sleek system that was easy to install and ensure that the work is done seamlessly, this also posed the challenge of working with substantially underpowered for real-time deep learning applications such as face recognition.

Another challenge that they encountered was the client’s’ reluctance to send their video footage over cloud fearing security concerns, hence,  the team had to devise a mechanism to make the system work offline,  “In order to make Vizor ID work we used our own proprietary methods that enables deep learning models to run on low-cost devices. Our methods are a result of over a year of research that we invested at the start of this company. Thanks to our proprietary methods, Vizor ID can now run on low-cost devices such as the $35raspberry Pi!” Singh exclaims.

The road ahead

Though it took them over a year’s time to set up the startup, within a short time of inception, it has closely worked with over a dozen clients and has taken up over 50 projects. Recently, Unreal AI was one among the eight startups which graduated from Zeroth. AI’s three-month-long cohort.

Market expansion and installation of 100 plus Vizor ID devices is their focus for 2019 and are in talks with Chinese hardware manufacturers for developing cheaper hardware systems to achieve this, “This year we plan on installing 100 Vizor ID devices. We are talking to hardware manufacturers in China to build these devices. We plan on making Vizor ID available in New Delhi and Bengaluru,” Singh concludes.

What Do You Think?

Subscribe to our Newsletter

Get the latest updates and relevant offers by sharing your email.
Join our Telegram Group. Be part of an engaging community

Copyright Analytics India Magazine Pvt Ltd

Scroll To Top