Scale AI New Grad MLE Image Processing Phone Screen

Before the phone screen, they completed Complete a Small CNN Oriented-Box Regression Pipeline, a take-home about detecting rotated stars. They added IoU to the loss and reported a score above 70%.

The notebook task had four parts. First, Extract the Prefix of a Filename to get an image ID, then look up its class in JSON. Next, load and shuffle 500 images into a TensorFlow tensor with their labels. Then evaluate two provided models and discuss the metrics. The last part asked for the images that both models predicted correctly.

They used F1 as one of the metrics and didn’t finish the last part. They were rejected after the screen.