Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Calibrating YOLO Models for More Reliable Waste Detection
This talk addresses waste detection model miscalibration in recycling factories using temperature scaling to improve prediction reliability and reduce confidence errors.
I will explain the issues with traditional waste management in recycling factories and how we can leverage AI to streamline and optimize these processes. Then I will discuss how waste detection models can suffer from miscalibration and why addressing this is important. To tackle miscalibration, I apply temperature scaling, a widely used and effective method for improving calibration. I will explain why temperature scaling is a good fit for this problem and how it helps reduce confidence errors in our predictions.