Computer Vision News - December 2022

24 AI Spotlight News Automatic Forest Fire Detection System with AI Enables Early and Efficient Fire Fighting Here’sabrilliantwaybywhichAIhelpscontainoneoftheissuesgeneratedbyclimatechange: forestfires . Deep learningalgorithms teachan imageprocessing systemtosee, recognizeandverify theoccurrence of smoke . Furthermore, AI enables a corresponding image processing system to draw conclusions from what it learns. It’s a French company called Paratronic that has developed this automatic forest fire detection system called ADELIE (Alert Detection Localization of Forest Fires) : four industrial cameras observe a forest area within a radius of up to 20 kilometers and monitor it at 360 degrees in a couple of minutes. Read More Computer Vision News has found great new stories, written somewhere else by somebody else. We share them with you, adding a short comment. Enjoy! Amazon’s Just Walk Out and Amazon One - Beyond the Future of Shopping WIRED Brand Lab wrote a very nice article for Amazon , describing how their Just Walk Out technology and Amazon One are transforming the way we interact with the world, from retail to entertainment and business. Of course, this is scripted text, but it is interesting to see what Amazon sees in the future of shopping. This piece includes many quotes by Gérard Medioni who, besides being a big friend of our magazine and a pillar of our community, is also Vice President and Distinguished Scientist at Amazon. He has a lot to say about “ what machine learning can do to create magical experiences for consumers! ” Read More Researchers at Stanford Have Developed an AI Approach for Fast Model Editing at Scale Largemodels have improvedperformanceon awide rangeofmodern computer vision problems, in particular in NLP (Natural Language Processing) . However, issuing patches to adjust model behavior after deployment is a major challenge in deploying and maintaining such models! To enable easy post-hoc editing at scale, Eric Mitchell and his colleagues at Stanford propose Model Editor Networks with Gradient Decomposition (MEND) , a collection of small auxiliary editing networks that use a single desired input-output pair to make fast, local edits to a pre-trained model’s behavior. Read More