baysik

We decided to see if we could train an image classification model to recognize something as ethereal as 'basic'. We all know originality and banality when we see them, but can one quantify that? assign a score?

Training the Model

Instagram provides a data rich environment where we can analyze images for basicness. We trained our model on ~200 'basic' acounts: @totalfratmove, @kimkardashian, @avacado_toast and then gave our model ~200 photos from the instagram accounts of highly original artists: Tyler the Creator, Kanye West, James Bay, several photographers, The New Yorker, National geographic. The results were hilarious and shocking. We found that highly basic photos consistently scored in the 50%-70% confidence range.

The Results

This photo scored a ~70%. Suuuuper basic.

on the other hand

This photo of James Bay scored a 3%. Cool! Hip! Streets ahead!

The webapp for accessing this classifier is, let's say, still in the works. You can sign up with Instagram, we can access the apis to get your latest instagram images, we can analyze them and get their scores, but something there's some funkiness with asynchronous server-side javascript calls and Meteor's reactivity system that for the life of me I cannot figure out.

Both of our team also got very ill (_well, one of our team got the rest of us very ill *cough cough laya *) and we had to head home before the expo.

That being said. We solved it. We solved for basicness.

Share this project:
×

Updates