Face embedding and comparison in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is a face embedding in face recognition?

A face embedding is a list of numbers that represents the unique features of a person's face. It is like a digital fingerprint for the face, used to compare and recognize faces.

beginner

Why do we use face embeddings instead of raw images for comparison?

Face embeddings simplify the face data into numbers that capture important features. This makes comparing faces faster and more accurate than using raw images, which have too much unnecessary detail.

intermediate

What is the common method to compare two face embeddings?

We usually calculate the distance between two face embeddings using methods like Euclidean distance or cosine similarity. Smaller distances mean the faces are more likely to be the same person.

intermediate

How does a neural network create a face embedding?

A neural network processes the face image through layers that learn to extract important features. The output is a fixed-size vector (embedding) that captures the face's unique traits.

beginner

What role does a threshold play in face comparison?

A threshold is a set value for the distance between embeddings. If the distance is below this threshold, the faces are considered a match; if above, they are different people.

What does a face embedding represent?

A A unique numeric representation of a face

B A raw image of a face

C A text description of a face

D A video of a face

Face embeddings are numeric vectors that capture unique features of a face for comparison.

Which distance metric is commonly used to compare face embeddings?

A Jaccard index

B Hamming distance

C Euclidean distance

D Manhattan distance

Euclidean distance measures the straight-line distance between two embeddings and is commonly used.

Why is a threshold needed in face comparison?

A To detect faces in images

B To resize face images

C To convert images to grayscale

D To decide if two faces match based on embedding distance

The threshold helps decide if the distance between embeddings means the faces are the same person.

What is the output size of a face embedding vector typically?

A A fixed-length numeric vector

B A variable-length text string

C A color image

D A binary mask

Face embeddings are fixed-length numeric vectors representing face features.

Which of these is NOT a benefit of using face embeddings?

A Faster face comparison

B Shows the original face image

C Requires less storage than raw images

D More accurate recognition

Face embeddings do not show the original image; they are numeric summaries.

Explain how face embeddings are created and used for face comparison.

Think about how a face image turns into numbers and how those numbers tell if two faces are the same.

Describe why face embeddings improve face recognition compared to using raw images.

Consider the difference between comparing full pictures versus comparing summaries.