I know it’s easier to learn robotics when I’ve decided on a robot to build. In the same way, I bet it’s easier to learn ML when you have a specific goal in mind. Maybe even something novel. So.

My challenge to you is to train a ML artist. this would be a network that inputs a bitmap picture and outputs a set of vectors that could be drawn by a plotter robot. I have lots of non-ML image->vector converters and I’m a big fan of turtletoy.net. I would like to see your network run on turtetoy given a source image. They could be any style you want. You don’t have to share your training data, just the final resulting weights and the NN to run in javascript on their site. Your result would be public for everyone to try and enjoy, while your trade secret training stays all yours.

I suspect I’m going to get a lot of “that’s dumb I won’t do that”, and I feel embarrassed making this post. But you miss 100% of the shots you don’t take, right?