Face landmark detection in Computer Vision - Model Metrics & Evaluation

Metrics & Evaluation - Face landmark detection

Which metric matters for Face Landmark Detection and WHY

Face landmark detection finds key points on a face, like eyes, nose, and mouth corners. The main metric is Mean Squared Error (MSE) or Normalized Mean Error (NME). These measure how close the predicted points are to the true points. Lower error means better accuracy. We use these because this is a regression task, not classification.

Confusion Matrix or Equivalent Visualization

Face landmark detection does not use a confusion matrix because it predicts coordinates, not classes. Instead, we visualize errors as distances between predicted and true points.

True points:       (x1, y1), (x2, y2), ..., (xN, yN)
Predicted points:  (x1', y1'), (x2', y2'), ..., (xN', yN')

Error per point = sqrt((x - x')^2 + (y - y')^2)
Mean Error = average of all point errors
    

Tradeoff: Precision vs Recall Equivalent

In face landmark detection, the tradeoff is between accuracy (how close points are) and robustness (working well on different faces and conditions). Improving accuracy might make the model sensitive to small changes, reducing robustness. Balancing these ensures landmarks are precise and reliable.

What "Good" vs "Bad" Metric Values Look Like

Good models have low mean error, often below 5% of the inter-ocular distance (distance between eyes). For example, an NME of 0.03 means average error is 3% of eye distance, which is good. Bad models have high errors, like 0.1 or more, meaning points are far from true locations.

Common Pitfalls in Metrics

• Ignoring scale: Measuring error in pixels without normalization can mislead if face sizes vary.
• Overfitting: Very low training error but high test error means model memorizes training faces.
• Data leakage: Using same faces in training and testing inflates accuracy.
• Not considering robustness: Good error on clear images but poor on occluded or rotated faces.

Self Check

Your face landmark model has a mean error of 0.02 on training data but 0.08 on new faces. Is it good for real use? Why or why not?

Answer: No, it is not good. The model fits training faces well but performs poorly on new faces, showing overfitting and poor generalization. You need to improve robustness and test on diverse data.

Key Result

Mean error (e.g., Normalized Mean Error) is key to measure how close predicted face landmarks are to true points; lower values mean better accuracy.

Practice

1. What is the main purpose of face landmark detection in computer vision?

easy

A. To detect the presence of multiple faces in an image

B. To classify the face into different age groups

C. To find key points on a face that represent facial features

D. To convert a face image into grayscale

Solution

1. Step 1: Understand face landmark detection

   Face landmark detection identifies specific points on a face like eyes, nose, and mouth corners.

2. Step 2: Differentiate from other tasks

   Other tasks like age classification or face detection do not focus on key facial points.

3. Final Answer:

   To find key points on a face that represent facial features -> Option C

4. Quick Check:

   Face landmark detection = key facial points [OK]

Hint: Landmark detection means finding facial points, not classifying faces [OK]

Common Mistakes:

• Confusing face detection with landmark detection
• Thinking it classifies face attributes like age or emotion
• Assuming it changes image colors

2. Which of the following is the correct way to load a pre-trained face landmark model using MediaPipe in Python?

easy

A. mp_face_mesh = mp.solutions.face_mesh.FaceMesh()

B. mp_face_mesh = mp.solutions.load_face_mesh()

C. mp_face_mesh = mp.FaceMesh()

D. mp_face_mesh = mp.solutions.face_landmark()

Solution

1. Step 1: Recall MediaPipe face mesh syntax

   The correct class to load face landmark detection is FaceMesh under mp.solutions.face_mesh.

2. Step 2: Check each option

   Only mp_face_mesh = mp.solutions.face_mesh.FaceMesh() uses the correct full path and class name.

3. Final Answer:

   mp_face_mesh = mp.solutions.face_mesh.FaceMesh() -> Option A

4. Quick Check:

   Correct MediaPipe class = FaceMesh [OK]

Hint: MediaPipe face landmarks use mp.solutions.face_mesh.FaceMesh() [OK]

Common Mistakes:

• Using incorrect method names like load_face_mesh()
• Omitting the full module path
• Using non-existent classes like face_landmark()

3. Given the following code snippet using MediaPipe to detect face landmarks, what will be the output type of results.multi_face_landmarks if a face is detected?

medium

A. A list of face landmark objects

B. A single integer representing number of faces

C. A dictionary with landmark coordinates

D. None

Solution

1. Step 1: Understand MediaPipe output for face landmarks

   When a face is detected, results.multi_face_landmarks returns a list of landmark objects, one per face.

2. Step 2: Analyze the options

   Only A list of face landmark objects correctly describes this as a list of face landmark objects.

3. Final Answer:

   A list of face landmark objects -> Option A

4. Quick Check:

   Detected faces landmarks = list of objects [OK]

Hint: multi_face_landmarks returns a list, not a number or dict [OK]

Common Mistakes:

• Thinking it returns a count (integer)
• Assuming it returns a dictionary
• Confusing None with detected results

4. You wrote this code to draw face landmarks but get an error: mp_drawing.draw_landmarks(image, results.face_landmarks, mp_face_mesh.FACEMESH_TESSELATION). What is the likely cause?

medium

A. FACEMESH_TESSELATION is misspelled

B. mp_drawing is not imported

C. image must be grayscale

D. results.face_landmarks should be results.multi_face_landmarks[0]

Solution

1. Step 1: Check results object structure

   MediaPipe stores landmarks in results.multi_face_landmarks as a list, not results.face_landmarks.

2. Step 2: Correct usage for drawing

   To draw landmarks for the first face, use results.multi_face_landmarks[0].

3. Final Answer:

   results.face_landmarks should be results.multi_face_landmarks[0] -> Option D

4. Quick Check:

   Use multi_face_landmarks list for drawing [OK]

Hint: Use multi_face_landmarks[0] to access first face landmarks [OK]

Common Mistakes:

• Using a non-existent attribute results.face_landmarks
• Forgetting to import mp_drawing
• Assuming image must be grayscale

5. You want to build an app that applies virtual glasses on a user's face using face landmark detection. Which landmarks should you primarily use to position the glasses correctly?

hard

A. Landmarks on the lips and chin

B. Landmarks around the eyes and nose bridge

C. Landmarks on the forehead and cheeks

D. Landmarks on the ears and jawline

Solution

1. Step 1: Identify glasses placement on face

   Glasses rest on the nose bridge and cover the eyes area.

2. Step 2: Choose landmarks matching these areas

   Landmarks around eyes and nose bridge provide accurate positioning points.

3. Final Answer:

   Landmarks around the eyes and nose bridge -> Option B

4. Quick Check:

   Glasses position = eyes + nose bridge landmarks [OK]

Hint: Glasses align with eyes and nose bridge landmarks [OK]

Common Mistakes:

• Using lip or chin landmarks which are too low
• Choosing forehead or cheeks which don't fit glasses
• Selecting ears which are not detected by standard landmarks