Three Insights Into the World of Video Analytics - Atmarkit Interviews Graymatics' Abhijit Shanbhag
During a recent trip to Tokyo, Graymatics’ CEO Abhijit Shanbhag met with Japan-based tech publication Atmarkit to share some unique insights on video analytics. Read on to check out our top three insights from the interview:
- Understanding video data is much more complex.
What does it take to find the video you’re looking for? For the most part, a lot of effort. While it’s easy to store and share specific videos, search functions are often limited. This is because video data is a lot more complex.
For instance, a camera can capture details such as colour, facial expressions, product labels, music and more through moving images but the average video platform would only understand a file’s content by the metadata text (eg. file name) provided.
Even YouTube, the Internet’s most popular video platform with over 60 hours of video uploaded in a minute, is limited by these constraints. Finding a specific video on the platform first depends on the video uploader manually providing an accurate title (eg. “Charlie Bit My Finger”) and an appropriate set of tags (eg. “funny”, “child”, “home video”). Next, the user looking for the video has to use the same or similar terms in their search query.
- Processing this data requires machines to “think differently”
A large majority of today’s platforms are limited to analysing text- or binary-based data. Yet, across the Internet, over 26 billion photos and over one billion seconds of video are uploaded daily. This is a huge amount of information which Big Data and search engines have thus far been unable to make sense of.
For computers to begin to understand the complexities of video data, they must learn to process images in the way humans do. At Graymatics, we’re beginning to uncover the infinite possibilities in this field. Powered by machine learning and neural networking, our technology can interpret visual data to capture unique insights.
For example in this video, our software can identify that there are two men in the room with one speaking English and the other speaking Japanese, and can even gauge the interviewee’s emotions from facial expressions.
- To young engineers: the possibilities with video and image analysis are endless
The video analytics landscape has evolved rapidly, but it still has far to go. A few decades ago, it would take over a week to process massive amount of video and image content. Today, however, it’s possible in as little as 70 microseconds.
Reflecting on his experience, Abhijit mentions that Graymatics has benefited from adopting a disruptive mindset when developing its technology. “If 5% of an engineer’s eight-hour work day was dedicated to creating disruptive innovations, that would amount to a total of 25 minutes.”
“Now, if we consider this on a bigger scale, India has approximately 1.5 million engineers. That would roughly amount to 26,000 working days or approximately 70 years of developing disruptive technologies a day. Can you imagine the potential?”