As a personal capstone project for Part 2 (2019), I took a big bite and attempted to implement my first paper: https://arxiv.org/abs/1603.06208. Initially I didn’t think it was too different from the image classification tasks that we already mastered in Part 1, but I soon learned that it had some interesting ideas that proved quite the challenge not only understand but also also to implement. It was a great learning experience, and I’m happy to say I’m no longer afraid of callbacks:)
I wrote all about it here: https://medium.com/@mathis.alex/rotationnet-in-fast-ai-adventures-in-virtual-set-creation-and-computer-vision-cd694ad7ec1b
Feedback and suggestions are always welcome.