Avengers: Infinity War is one of my favorite movies not only from the superhero genre, but of all time. Besides an insane character crossover, it features quite a few scenes with pop culture references. In one of them, Peter Quill (aka Star-Lord) refers to Thanos as "Grimace"
For those who don't know, Grimace is one of McDonaldland's characters. These were quite popular in the 80s and 90s (which fits with Quill's time in Earth as a kid). This reference inspired cosplayers Kittie Cosplay and Banana Steve in making an even crazier crossover: Grimos. This made me wonder: can we create an algorithm capable of distinguishing between these two purple characters?
In this project, I created a convolutional-neural-network- (CNN-) based classifier to discriminate images of Thanos and Grimace. I did so under two different approaches: creating a CNN from scratch and using a pre-trained CNN as a base (i.e., transfer learning). The classifiers reached an accuracy of 50 (?) and >98%, respectively. Although I'm sure the models can be improved and analyzed even deeper (specially in the from-scratch case), generating a perfect classifier wasn't the purpose of this repository.
You can find the notebook of the project here.
Although this is definitely not the first time that Thanos has received the Machine Learning/AI treatment, I still found working with him and Grimace very fun. If you have any comments or suggestions, feel free to raise an issue here or drop me a line in Twitter (@amoncadatorres).