Case Study

Find the defective stage when a cropper sends the wrong image patch to a species classifier.

Case context: A wildlife recognition pipeline is meant to spot a rabbit in a photo. Instead of isolating the animal, the first stage extracts a patch of empty pavement. The downstream species model receives that crop and outputs y=0, meaning no rabbit is present.

Question: Using the situation described here, determine which part of the pipeline caused the bad result. Explain your choice by referring to the downstream model's output and to what a person would conclude from the same crop.

Sample answer: The localization stage is the source of the failure. The downstream classifier handled its input appropriately: it was given a crop that showed only pavement, so returning y=0 was a sensible decision. A human looking at that same crop would also say there is no rabbit in it. Because the second stage responded correctly to the data it received, the mistake belongs to the earlier stage that selected the wrong region.

Key points:

  • The localization stage is the component at fault.
  • The downstream classifier is not to blame because it processed the provided crop correctly.
  • y=0 is the right outcome for an image patch with no rabbit.
  • A human reviewing that crop would also answer y=0.

Rubric: The learner must identify the localization stage as responsible, explain that y=0 is appropriate for the pavement crop, and state that a human would agree with that judgment.

0

1

Updated 2026-08-12

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

Related