Image Classification

Paul Butera, Altis Data Scientist, Melbourne

In the space of image classification, not much has changed for years. Up until recently, even utilising deep learning, the pinnacle of multiclass object detection remained, at most, detecting one or two different kinds of objects within a single frame. Put more simply, we have been able to train a classifier to answer the question “Is there a dog or cat in this image?” However, academic breakthroughs in the last 6 months have proved it possible to classify the whole zoo on 60+ frames per second video in 1080p!

Leveraging a cutting edge methodology centered on a tiered system of convolutional neural networks, an off the shelf GPU and a few days of work, we were able to build a system capable of classifying over 100 different classes of object on a live 60 frame per second 1080p webcam feed, which we demonstrated our prototype solution at the Gartner BI & AI Summit in February 2018.  Our prototype demonstrates that with the right team and technology you can quickly, easily and cheaply take advantage of the latest AI technologies.

The business applications of this breakthrough are endless. From automated video alarm systems, monitoring patrons for weapons – to building better automated production line machinery, capable of distinguishing products and objects faster and more accurately than ever before.

Contact us if you would like to find out more about how this breakthrough application could benefit your organisation.

Join the conversation

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.


Post has no comments.