As mentioned a month ago, I write a series of posts on detecting solar panels from satellite images at my company, Appsilon. In this post I focus on using fastai python library to deliver a working ML model in few hours!

The series covers an example ML project journey in python:

  1. Data collection and preparation in pandas and pillow.
  2. Creating a simple ML model for image segmentation using fastai.
  3. Model visualization and serving using streamlit.

In general I’m not a fan of fastai for long-lasting projects as it is often weird, code is ugly and documentation is irritating to say the least. On the other hand it really allows doing super quick PoCs. I spend once a few hours on tinkering my resnet-based model written in pytorch-lightning that was able to achieve 0.94 F1 on some classification task only to get 0.96 F1 out-of-the-box with fastai. Their introductory course to ML in pytorch is a valuable resource for those who want to get a quick walkthrough on how to approach different ML problems.

Check out my post if you want to see fastai in action!