An overview of how deep convolutional neural networks have gotten good at recognizing images, after a landmark 2012 paper called AlexNet transformed the field
Timothy B. Lee / Ars Technica : Tweets: @iveccc , @binarybits , @binarybits , @binarybits , and @binarybits Tweets: Ivan Vecerina / @iveccc : A truly excellent article on the recent history of machine vision and deep learning, also doing a great job making the technical approach (relatively) easy to understand. #deeplearning #machinevision https://lnkd.in/dGcRPDa Timothy B. Lee / @binarybits : It's astonishing how good convolutional networks have gotten at recognizing images. Here are some examples from the famous 2012 AlexNet paper. http://twitter.com/... Timothy B. Lee / @binarybits : AlexNet had a total of five layers. Here are some examples of the kinds of images recognized by layers 2, 3, 4, and 5 of an AlexNet-like network. (These images are from this excellent paper by Zeiler and Fergus. http://cs.nyu.edu/... ) http://twitter.com/... Timothy B. Lee / @binarybits : AlexNet-style networks use all parts of an image to figure out what's in it. Here's a fun example of that. On the right of this screenshot are nine images matched by a particular filter. http://twitter.com/... Timothy B. Lee / @binarybits : Since 2012, the performance of convolutional neural networks at image recognition has steadily improved. http://twitter.com/...