Jay's 2019 Picnic

Inspiration

Needed a toy problem to learn fast.ai with

What it does

Classifies food

How I built it

Using the fast.ai library and google colab

Challenges I ran into

• Misclassified training data
• Missing categories (e.g. coconuts)
• Disconnects from google colab
• Couldn't figure out what a good image size was for the chosen network structures

Accomplishments that I'm proud of

Only took a few hours for the setup and first classification. #1 ranking at the original deadline. Was able to try a few known structures, like resnet and xresnet.

What I learned

Libraries can be awesome. Deep learning is complicated.

What's next for Jay's 2019 Picnic

Trying resnext

Built With

• fast.ai
• google-colab
• pytorch