image 1
Ever since I was a little kid, I have been interested in entomology, and more generally, in categorizing and classifying things. This makes something like Pokemon a natural interest of mine (and interestingly, it was in fact Satoshi Tajiri’s own interest in bug collecting that inspired him to come up with the idea of Pokemon).
Thus, for my “teachable machines” assignment, I have decided to see if I can get my machine to classify 4 types of pokemon - electric (yellow LED), fire (red LED), water (blue LED) and plant (green LED).
To begin this project, I decided to go on teachable machines and give 4 types of image samples - an image of pikachu, bulbasaur, squirtle and charmander for each of the 4 classes respectively. I started with just one image of each of them as the source of my samples.
Testing with the exact photos I used for training
Testing with the exact photos I used for training
After training this model, I tested it out to see if it works - and it’s extremely accurate when I use the exact pictures I trained (I had about 60 samples of each photo).
However, as the second video shows, I ran into trouble when I used a different photo, in this case of charmander. It was not as accurate.
I took this a step further and just showed a whole grid from google images search of each pokemon, to see what it would give me.
My Model URL can be found at this link: https://teachablemachine.withgoogle.com/models/zTObVO8TT/
Arduino Board with 4 LED’s, one for each Pokemon Type I will be testing
Per our lesson in class, I modified the p5 code for the Image classifier to accommodate my 4 scenarios of Pokemon Types