Computer Vision News - January 2023

4 Computer Vision Tools Happy New Year for all the people who changed one, and Sea- son’s Greetings for everyone! This is an exciting month. New beginnings. And one of the most exciting news lately is the ChatGPT. I am sure you read, you saw tweets, everybody got excited, their systems were overloaded. It can do many things, with impressive accuracy and determi- nation. OK, I am so excited myself, that I will give space for the next paragraph. Me : Can you tell me a phrase about yourself, describing your technology to a very knowledgeable audience and write an introductory paragraph for my computer vision tutorial article in the new year’s issue of RSIP Vision’s magazine? ChatGPT : Certainly! Here is a phrase that describes my technology: "Assistant is a large language model trained by OpenAI, using state-of-the-art machine learning techniques to generate human-like text ." As an introduction to your computer vision tutorial article in the new year’s issue of RSVIP magazine, you might consider mentioning the increasing importance and widespread ap- plications of computer vision in various fields, from self-driving cars to medical image anal- ysis. You could also mention the potential for computer vision to transform industries and improve quality of life and introduce the concept of using machine learning techniques to teach computers to interpret and analyze visual data. You might also provide some context on the current state of the field and mention any exciting recent developments or advances. Wow! So, with this introduction, I’ll start my tutorial which is a computer vision analysis using computer vision models with convnets. Let’s dive in! Introduction The idea is that, as we saw on the last time, we use different layers of a convnet to extract features, this time we’ll use those feature extraction layers and put them together to build a network of our own  We have also seen in my previous tutorials, that convolutional networks perform feature extraction through three operations: filter, detect , and condense . A single round of feature extraction can only extract relatively simple features from an image, things like simple lines or contrasts. These are too simple to solve most classification problems. Instead, convnets will repeat this extraction over and over, so that the features become more complex and refined as they travel deeper into the network.