9 years, 4 months, 19 days ago I posted my first image to Meow Reader, a Tumblr dedicated to images of cats reading and cats learning how to read.
A few weeks back I mentioned the (then abadoned) site to my department chair and he (jokingly?) suggested I use Machine Learning to automate the discovery of new Meow Reader images.
Challenge Accepted.
A few Ruby scripts later (plus some research into the Clarifai API) and I’ve got a shiny new collection of reading cats, dogs, rabbits, sloths… you named it! I’ve documented the process below, but you can also skip straight to the images.
Finding reading animals, a play in five acts:
- Act 1 - Collect 140 existing images of reading cats.
- Act 2 - Use Clarifai to detect concepts within images from Act 1.
- Act 3 - Sort the discovered concepts by:
- How often they appear.
- Machine’s “confidence” in the concept.
- Act 4 - Collect 1000s of new animal images Tumblr.
- Act 5 - Filter images from Act 4 using concepts discovered in Act 2:
book, book bindings, book series, education, literature, newspaper, research, technology - Profit!
The Clarifai API could also be used in Act 5 to filter the images even further to limit the reading animals to be cats only. View the full source code here. There’s a separate script for each step.
Oh, and I also created a new version of the Meow Reader Android app using Vue.js and Cordova.
Please install it and leave me a glowing 5 star review. (/◔◡ ◔)/
