It’s tempting to think deep learning can just be thrown at a problem and will solve it. This is a concrete reminder that it's essential to understand your data distribution and its limitations, and to take steps to address those when you can.
Building a Segmentation Model with Contaminated Data
