AI-driven Capybara Detector

18 August 2019 ● 1 min read ● 2 images

I made a web app that uses machine-learning techniques to recognize images of capybaras.

Use case:

You’re at the petting zoo and you’ve just bought some Capybara Chow, but you notice that the picture signs identifying the animals are down for maintenance.

You need to know: “Is that a capybara?”

I trained a Convolutional Neural Net with thousands of images of capybaras and non-capybaras, using Python's PyTorch and FastAI libraries, then deployed a web app in Python.

My trained model has >99% accuracy on overall test data, but it occasionally makes mistakes with beavers and nutrias. 

Here's the app to try out:

And here's a short presentation on how it was made:

Greg Rawson - Creative Director in Tokyo, Japan Greg Rawson Creative Director 20 September 2019

Next time I'm at the zoo and the signs are missing, I know where to turn :)
This is really cool and fun.

You can log in to post a comment.

About Robb

Robb Satterwhite

I'm a writer in Tokyo, Japan

I am a copywriter (officially a Communications Strategist), and I produce several Tokyo-based websites.

See full profile

Welcome to Canvas Tokyo

Canvas is an online network for Tokyo's creative community. It's a place for professional and amateur creatives to show their skills, make new connections and find work.

Sign up to Canvas