{"id":7174,"date":"2019-08-26T19:26:41","date_gmt":"2019-08-26T17:26:41","guid":{"rendered":"https:\/\/blog.mi.hdm-stuttgart.de\/?p=7174"},"modified":"2023-06-18T18:23:42","modified_gmt":"2023-06-18T16:23:42","slug":"how-to-recognize-doodles-in-the-cloud","status":"publish","type":"post","link":"https:\/\/blog.mi.hdm-stuttgart.de\/index.php\/2019\/08\/26\/how-to-recognize-doodles-in-the-cloud\/","title":{"rendered":"How to Train a Doodle Image Classifier and Recognize Doodles in the Cloud"},"content":{"rendered":"\n
\"\"<\/figure>\n\n\n\n

As part of the lecture \u201cSoftware Development for Cloud Computing\u201d we developed a doodle image recognition game. The idea came to us when we were searching for possible mini-games for our semester project “Peers \u2013 The Party”, an iOS app using Apple’s MultipeerConnectivity framework. <\/p>\n\n\n\n\n\n\n\n

Many people don\u2019t realize that the story of today\u2019s applied AI is actually a story about the cloud. <\/p>Cassie Kozyrkov (Chief Decision Intelligence Engineer, Google) [1]<\/cite><\/blockquote>\n\n\n\n

Goal<\/h2>\n\n\n\n

During our research, we found Google’s “Quick, Draw!” [2] which was basically what we wanted to achieve. The game principle is quite simple: The player gets shown a word and has to draw it within a certain time. He wins if the doodle can be recognized by the neural network. <\/p>\n\n\n\n

Luckily, Google open-sourced their huge dataset [3] of 50 million doodles that the game is based on. So now it was time for us to think about how to train our own neural network and use it in our game. <\/p>\n\n\n\n

\"\"
Quick Look at the “Quick, Draw!” Dataset [3]<\/figcaption><\/figure>\n\n\n\n

Local Machine Learning Setup<\/h2>\n\n\n\n

As a first step, we wanted to set up a local training environment to later implement the same principle in the cloud. <\/p>\n\n\n\n

We started using Google’s TensorFlow<\/strong>, an open-source platform for machine learning, following the tutorial “Recurrent Neural Networks for Drawing Classification” [4]. For easy collaboration, we set up a shared virtual environment using Vagrant<\/strong> based on an Ubuntu image with all needed technologies (especially Python & TensorFlow) installed. <\/p>\n\n\n\n

Some things to be aware of that we learned after making many mistakes: <\/p>\n\n\n\n