Case Study

What should you do with the 90 incorrect bounding box examples to improve the cat detector component?

Case context: You are building a cat-detection pipeline. During the initial component attribution phase, you identify that the cat detector component outputted incorrect bounding boxes in 90 specific error examples. Rather than collecting a brand-new dataset of cat detector failures, you decide to use these 90 examples to investigate further.

Question: Based on the principles of pipeline error analysis, describe how you should utilize these 90 conveniently found error examples to improve the cat detector's performance.

Sample answer: You should reuse these 90 specific examples where the cat detector outputted incorrect bounding boxes to carry out a deeper level of error analysis specifically on the cat detector component. By analyzing only these 90 failures, you can categorize the types of errors the cat detector is making and determine the most effective ways to improve that specific component.

Key points:

  • Reuse the 90 incorrect bounding box examples attributed to the cat detector.
  • Perform a deeper level of error analysis specifically on the cat detector component.
  • Identify specific failure modes or improvement areas within that component using the isolated subset.

Rubric: The learner must identify that the 90 error examples should be reused to perform a deeper level of error analysis specifically on the cat detector component. They should mention analyzing these specific failures to identify patterns or causes of failure to guide improvements for that component.

0

1

Updated 2026-05-27

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI