Picky Art Critic

• Sketch link

This is sketch using a simple DoodleNet Classifier with varying responses based on confidence level, as well as an ideally more lenient classification.

This is kind of like a drawing board game where you play with a computer. You input the subject you want to draw into the field below the sketch and press “calibrate”, and then the art critic will judge your drawings based on confidence level to the calibrated subject. There are three different states that can happen based on confidence.

One neutral state where the art critic deems your drawing as not resembling the target subject at all, and does not even bother to respond.

One “dissatisfied” state, where the critic has taken note of your work, but tells you that it is not good enough and you should try again.

One “satisfied” state, where if the critic has 80% or more confidence that your drawing is resembling the subject, will be visibly happier and congratulate you on your performance.

• Inspiration: How did you become interested in this idea? Quotes, photographs, products, projects, people, music, political events, social ills, etc.

  • In general, it kind of puts into form a kind of mindset that I tend to deal with when I draw. It’s a little bit of a perfectionist mindset (obviously not complete or else it might be impossible). And if I had to think of something specific, I think I was kind of inspired by Anton Ego in Ratatouille, who would not swallow food unless he loved it.

• Process: How did you make this? What did you struggle with? What were you able to implement easily and what was difficult?

  • I started out with a DoodleNet classifier and decided to simply just make some improvements, as well as hide some debugging info like the confidence level because players don’t exactly need to know that stuff.
  • DoodleNet strictness was the way it was because it took the first element in the array of results. Tried to remedy via taking the first 5 elements in the array instead. Possibly successful? It’s still somewhat janky and annoying but it seems better, so I’ll take what I can get.
  • Tried many different timed response implementation methods and automation of the response -> clear sketch process. Was unsuccessful.

• Audience/Context: Who is this for? How will people experience it? Is it interactive? Is it practical? Is it for fun? Is it emotional? Is it to provoke something?

  • To be honest, there’s no specific target audience in mind. It’s kind of just a fun (or frustrating) game you play where you repeatedly draw until the machine is satisfied with your work.

• Code references: What examples, tutorials, references did you use to create the project? (You must cite the source of any code you use in code comments. Please note the following additional expectations and guidelines at the bottom of this page.)

  • Classifying Drawings with ml5's DoodleNet: output in separate canvas

• Next steps: If you had more time to work on this, what would you do next?

  • I would definitely try to improve the responses to automatically clear your canvas for you when you submit, as I was not able to get that working. I think more responses would also be a good place to explore.