The Open Data Science Conference (East) looked a bit different this year. While typically 6,000+ data science professionals gather in Boston for the Expo, the team at ODSC moved the entire event online. Still, over 3,000 "attendees" gathered in a shared Slack Workspace and online livestream to enjoy the dozens of speakers.
I had the privilege of offering a three-hour tutorial on building a custom object detection model from scratch. In some ways, the live online delivery format improved the discussion: participants were able to live react to commentary about space, answer one another's questions with minimal disruption, and have a live-log of all comments for after the presentation.
In my tutorial, we walked through a live example of preparing an object detection problem from end-to-end. This includes identifying a good problem statements, tips on collecting images, best practices for labeling images, practical tips on preprocessing, the importance of considering good image augmentations, training a model, and using that model for inference.
If you missed the tutorial, this blog post on training YOLOv3 with a custom dataset is a good complement to the discussion.