Capstone_Chihuahua_Muffin
1.0.0
Background - Meme: In 2016, a meme went viral that asked people a question they'd likely never thought would be challenging before: Can you tell the difference between these images that you never before thought looked alike?
(source: Elle Magazine, https://www.elle.com/culture/news/a34939/animals-or-food/)
As the owner of a Chihuahua, my interest zeroed in on this pairing:
Background - Image Classification: A statement often made about image classification algorithms is that though they can quickly distinguish between thousands of images with pretty good accuracy, a child can distinguish between images with much better accuracy.
The question of Chihuahua versus Blueberry Muffin fascinated me because--in the case of the particular close up angles selected for the meme--this is not a case when a human can easily distinguish between these images.
Problem Statement: After building an image classification model that can predict whether an image is of a Chihuahua or a Blueberry Muffin trained on zoomed out distinctly different photos, can that image classification model accurately predict the classification for the challenging zoomed in photos from the meme?
I began with a CNN trained from scratch on the scraped images. The maximum accuracy I gathered from this was 85%. I then incorporated Image Data Generation to increase my training size. This did not improve my accuracy but did increase the speed of running the model. Finally, I implemented transfer Learning and achieved >99% accuracy.
