Blur Detector: Base Rates, Thresholds and Launch Failures

MediumClassical ML

Reported in interviews at Apple

Problem statement

How would you use a blur detector to choose photos for highlight albums? Blurry photos are uncommon, and hiding a sharp photo annoys users more than keeping a soft one. You need to understand what a flag means, choose a suitable threshold and investigate failures after launch.

Work through probability, evaluation and debugging the image pipeline.

Show hint

Before trusting a flag, ask how many sharp photos there are for every blurry one.

Relevant roles: ML Engineer, Applied Scientist and Data Scientist.

Related free problems