Garfield is a common trope among AI and Machine Learning projects, whether it is generating custom comic strips or taking existing comic strips and erasing the character Garfield altogether.
What it does
When given an image of one of the characters (Garfield, Jon, or Odie) the AI should be able to tell who it is.
How we built it
Used the google collab classification demo (https://github.com/tensorflow/docs/blob/master/site/en/tutorials/images/classification.ipynb) as a basis of where to start, changed the data from flowers to a custom dataset of images I got off the internet of Garfield, Odie, and Jon and ran a training set of about 100 epochs instead of 10.
Challenges we ran into
When an image that is to be tested is taken from a url online, depending on the site's security settings sometimes the crawler is unable to download the image because the hosting website forbids it (tokens, cookies, direct file access, etc). This can sometimes cause hickups in the analysis process.
Accomplishments that we're proud of
After 100 epochs of training it reaches a model training accuracy of 1.0/1/0
What we learned
How to import data into a model and train it
What's next for Is this Garfield?
Most likely add more characters into the model.