Face landmark detection in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main purpose of face landmark detection?

Imagine you want to put virtual glasses on a person's face in a photo. What does face landmark detection help you find?

A The exact positions of key facial points like eyes, nose, and mouth

B The person's age and gender

C The background scenery behind the person

D The color of the person's hair

💡 Hint

Think about what you need to place glasses correctly on a face.

Face landmark detection finds important points on the face, such as eyes, nose tip, and mouth corners. These points help align or modify the face in images.

❓ Predict Output

intermediate

Output of face landmark coordinates extraction code

What will be the output of this Python code snippet that uses a face landmark detector?

Computer Vision

import numpy as np
landmarks = np.array([[30, 40], [35, 45], [40, 50]])
print(landmarks.shape)

A Error: 'shape' attribute not found

B (2, 3)

C [30 40 35 45 40 50]

D (3, 2)

💡 Hint

Check how many points and coordinates each point has.

The landmarks array has 3 points, each with 2 coordinates (x and y), so the shape is (3, 2).

❓ Model Choice

advanced

Choosing the best model architecture for real-time face landmark detection

You want to build a face landmark detector that works fast on a mobile phone. Which model architecture is best?

A A lightweight MobileNetV2 with depthwise separable convolutions

B A large ResNet with 50 layers

C A deep Transformer model with self-attention layers

D A simple linear regression model

💡 Hint

Think about speed and size for mobile devices.

MobileNetV2 is designed to be fast and small, making it ideal for mobile real-time tasks like face landmark detection.

❓ Metrics

advanced

Evaluating face landmark detection accuracy

Which metric best measures how close predicted face landmarks are to the true landmarks?

A Accuracy of classifying face vs non-face

B Mean Squared Error (MSE) between predicted and true landmark coordinates

C Cross-entropy loss on landmark heatmaps

D F1 score on detected face bounding boxes

💡 Hint

Think about measuring distance between points.

MSE calculates the average squared distance between predicted and true landmark points, directly measuring landmark accuracy.

🔧 Debug

expert

Debugging incorrect face landmark predictions

You trained a face landmark model but it predicts landmarks shifted far from the face in test images. What is the most likely cause?

A The optimizer learning rate was too low

B The model used too many layers causing overfitting

C The training data landmarks were normalized but test data landmarks were not

D The test images were grayscale but training images were color

💡 Hint

Think about coordinate scales and consistency between training and testing.

If training landmarks are normalized (e.g., scaled between 0 and 1) but test landmarks are not, predictions will be off-scale and appear shifted.