A.I. : State-of-the-art of Computer Vision

By | January 5, 2017

In another life (almost 14 years ago) for some years I had the privilege to be a member of one of the top European computer vision groups VISICS/PSI VISICS/PSI.

My focus then was object categorization using neural networks, not a very popular topic at that time due to limitations in data availability and computation power. Everybody was looking for just good enough faster algorithms not at something that emulates the biological vision system. As is normal I still have nice memories from that time and I still have an affinity for computer vision and neural networks.

So how to you think I feel when I see that this exact domain becomes mainstream and is the focus of everyone 🙂 Something like, I told you so 14 years ago 🙂

My humble contribution to the domain is my published paper View based categorization of Complete Objects. An extended version of the same paper Hierarchical Chorus of Prototypes was part of the deliverables of Cognitive Vision Systems (CogViSys) an European financed project I was part of. That evolved after a decade as the AXES project. Check it out is very cool and also free to use. I know at least 6 of the people from that photo of the project team and they pose on the stairs of the building where I worked for 3 years and a half.

A friend shared with me this great article from New York Times The Great A.I. Awakening How Google used artificial intelligence to transform Google Translate, one of its more popular services — and how machine learning is poised to reinvent computing itself. By GIDEON LEWIS-KRAUSDEC. 14, 2016

This is a must read and an incredible well written piece with links to several great papers that made this “awakening” possible. Believe me I printed every one of the papers linked by the article and the result is a full book about the scientific evolution of an A.I that peaked in our time with the google A.I that does image detection and translations.
Google being google has exactly what I dreamed of 14 years ago: vast training data , vast computing capabilities (see also the discussion about the new tensor processors).

But even if you are not google with the vast data and computation power you still can make great things as this guy Joseph Redmon ( I know, ponies, do not ask 🙂 ) the creator of Darknet: Open Source Neural Networks in C. The best part is that the project is an opensource project and is used by the author in YOLO: Real-Time Object Detection and ImageNet Classification

Please read all the papers and play with the darknet applications. I hope you will be enthusiastic as I am about A.I.s and neural networks after the read.

Advertisements