Friend or Toad?

correct output!
incorrect output :( (probs due to biases in dataset)

Inspiration

A pretty big motif in this hackathon/learnathon was frogs! We wanted to build something that would be practical, and also cute and fun like frogs. The cane toad is an invasive species in Australia that kills native species vital to the ecosystem. Many Australians don't know whether a frog is an ordinary native toad, or if it's a deadly cane toad, so many native toads are unnecessarily killed every year, while cane toads escape.

What it does

This uses computer vision, trained on a 90-10 train-test split of images of regular frogs and cane toads, to identify if a given frog is a cane toad or not. It is more or less of a basic, from scratch CNN.

How we built it

We used Keras and OpenCV for the machine learning bit, and used Wix for the frontend, and Flask for the backend. https://www.youtube.com/watch?v=dQw4w9WgXcQ

Challenges we ran into

Most of this was new to all members of the team; none of us had much experience in machine learning/computer vision, or working with a backend and API calls.

We didn't know how to connect the python code to a website, so we initially tried AWS Lambda, which was a bit fun, but also didn't seemed to suit our purposes especially given the time frame. Getting anything with triggers working seemed pretty impossible, so we then switched to flask, which was also new to us.

The API's functionality ranges between questionable to dysfunctional, so there's that.

Accomplishments that we're proud of

Learning how to use OpenCV was a pretty big one, and how to use images with PIL was really cool! The model also seems to work reasonably well.

What we learned

We learned about how to use OpenCV and Keras! This was our first time working with those libraries. We learned more about backend development and using tools like flask.