How Beamr Technology Can Boost Machine Learning

Image credit: Envato elements

Beamr BMR is advancing on a new front and reveals its capability to boost Machine Learning for video. Machine Learning and Artificial Intelligence for video have demonstrated immense achievements and have even more tremendous potential. This hot field is expanding fast, as part of the computer vision market that is already estimated at more than $20 billion and expected to grow exponentially in the coming years.

But one of the biggest pain points that slows down progress is managing extremely large files and libraries. That is because video files are relatively large, and for training computer networks to recognize moving objects, you need lots and lots of them.

Think of how you recognize a car or a human. For us, it is an easy task, but not for a computer, either in a single image and certainly in a video. Each movement changes how the object looks, its shape, size and angle.  That’s why computer networks must scan and analyze countless videos to learn how to recognize if there is a human, a car,  a cat or anything else in them. Players in Machine Learning deal with, not to say are stuck with, large clusters of video files that are extremely difficult to manage, store and transfer. 

All these technical details sum up to a very clear bottom line for the many companies and start-ups in this field: heavy expenses that hinder their growth. 

Tamar Shoham, Beamr CTO, explained how Beamr technology can help such enterprises cut their cost: “Overcoming one of the most difficult and expensive challenges of Machine Learning rests on Beamr technology's proven and tested ability to perform a remarkable thing, scanning each and every frame of a video file and concluding how much it can be compressed without losing its quality”. 

Shoham recently led an experiment that showed that Machine Learning workflows benefit from Beamr's proven ability to create a compressed file that looks exactly the same as the original one. These files were downsized on average by 40% — streamlining Machine Learning processes and allowing significant savings in storage and costs. The experiment confirmed that when performing people detection on the optimized and smaller files, essentially the same results were obtained. Further research and evaluation of possible contributions to Machine Learning workflows is already underway by Beamr’s R&D teams.

Image credit: Envato elements

The tests were conducted on NVIDIA DeepStream SDK - a tool for AI-based multi-sensor processing, video, audio and image understanding, which was a natural choice for Beamr as an NVIDIA Metropolis partner.

Shoham said: "We showed that the video optimization process, which cuts down the file size, didn't affect the detection results obtained with the DeepStream SDK, an enabler for vision AI applications and services. We are thankful to the Nvidia DeepStream team for supporting our research”.

In the last decade, Beamr’s inventive technology, a winner of the Emmy award for Technology and Engineering in 2021 and backed by 53 patents, aims to provide the best possible tradeoffs between quality and compression of video files, whether it is used for streaming films on Netflix, which is a long-time customer of Beamr, or examined by professionals who scan every pixel and in a wide range of use-cases.

About Beamr

Beamr BMR is a world leader in content-adaptive video solutions. Backed by 53 granted patents, and winner of the 2021 Technology and Engineering Emmy® award and the 2021 Seagate Lyve Innovator of the Year award, Beamr's perceptual optimization technology enables up to a 50% reduction in bitrate with guaranteed quality.

For more details, visit www.beamr.com

This post was authored by an external contributor and does not represent Benzinga's opinions and has not been edited for content. This contains sponsored content and is for informational purposes only and not intended to be investing advice.