Face embedding and comparison in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What does a face embedding represent?

Imagine you have a photo of a face. A face embedding is a way to turn that photo into a list of numbers. What do these numbers represent?

A The coordinates of the face in the original image.

B The exact pixel colors of the face image stored as numbers.

C A random set of numbers unrelated to the face's appearance.

D A unique numeric summary capturing important features of the face for comparison.

💡 Hint

Think about how a fingerprint summarizes unique patterns.

Face embeddings are numeric vectors that capture unique facial features, allowing easy comparison between faces.

❓ Predict Output

intermediate

Output of cosine similarity between two face embeddings

Given two face embeddings as numpy arrays, what is the output of the cosine similarity calculation?

Computer Vision

import numpy as np
embedding1 = np.array([1, 0, 0])
embedding2 = np.array([0, 1, 0])
cos_sim = np.dot(embedding1, embedding2) / (np.linalg.norm(embedding1) * np.linalg.norm(embedding2))
print(round(cos_sim, 2))

A 0.00

B 1.00

C 0.50

D -1.00

💡 Hint

Cosine similarity measures angle between vectors; orthogonal vectors have similarity zero.

The two embeddings are orthogonal (at 90 degrees), so cosine similarity is 0.

❓ Model Choice

advanced

Best model type for generating face embeddings

Which type of neural network model is best suited to generate face embeddings that capture facial features effectively?

A Recurrent Neural Network (RNN) trained on text data

B Decision tree trained on tabular data

C Convolutional Neural Network (CNN) trained on face images

D Fully connected network trained on random noise

💡 Hint

Think about models good at processing images.

CNNs are designed to extract spatial features from images, making them ideal for face embeddings.

❓ Metrics

advanced

Choosing a metric for face embedding comparison

You have two face embeddings and want to decide if they belong to the same person. Which metric is most commonly used for this comparison?

A Mean squared error

B Cosine similarity

C Accuracy score

D Cross-entropy loss

💡 Hint

Think about measuring angle similarity between vectors.

Cosine similarity measures how close two vectors point in the same direction, useful for face embeddings.

🔧 Debug

expert

Why does this face embedding comparison code fail?

Consider this code snippet to compare two face embeddings:

embedding1 = [0.1, 0.2, 0.3]
embedding2 = [0.1, 0.2, 0.3]
cos_sim = sum(embedding1 * embedding2) / (sum(embedding1**2)**0.5 * sum(embedding2**2)**0.5)
print(round(cos_sim, 2))

What error will this code raise?

A TypeError: can't multiply sequence by sequence

B ZeroDivisionError: division by zero

C NameError: name 'embedding1' is not defined

D No error, outputs 1.00

💡 Hint

Look at how lists are multiplied in Python.

Lists cannot be multiplied element-wise with * operator; this causes a TypeError.