Deep learning for mining data about your fish's wearabouts

If you're interested in checking out how it works see below

\/ \/ \/

Check out our development slides HERE!


Humans really are amazing creatures, our brain can tell the difference between millions of objects. Heck, we named them didn't we? We dared to see if a computer via a Deep Learning (FishNet) could learn to recognize one object. The Betta fish Watson was our adorable test subject for this years hackathon.

How it works

That's a long story! Basically, we trained our Neural Network (NN) to recognize a fish using a tiny Python script. We click the position of the fish and data is generated. We then upload this data to Digits, a machine learning application on one of our desktops, which uses GPU horse power to "learn" what a Betta fish looks like. FishNet-Trainer

We then loaded the image and coordinate data into the Application Digits, to start parsing and learning. This is a process that took us hours of fine tuning! We finally determined a model that would predict if a Betta fish in the frame, all using machine learning!

After training our NN we used a model to determine if the image contained a Betta fish or not and where Watson was. Then using bash and python scripts we updated his image containing a caption with what the Neural Network thought he was doing.

Built With

Share this project: