Face recognition concept in Computer Vision - Model Pipeline Trace

Model Pipeline - Face recognition concept

This pipeline identifies who a person is by looking at their face in a photo. It learns from many face pictures, then can recognize new faces by comparing features.

Data Flow - 5 Stages

1Input Image

1 image x 100 x 100 pixels x 3 color channels→Receive a color photo of a face→1 image x 100 x 100 pixels x 3 color channels

Photo of a person’s face with RGB colors

↓

2Face Detection

1 image x 100 x 100 x 3→Find and crop the face area from the image→1 image x 64 x 64 x 3

Cropped face region focusing on eyes, nose, mouth

↓

3Feature Extraction

1 image x 64 x 64 x 3→Convert face image into a list of numbers representing unique features→1 vector x 128 features

[0.12, 0.45, 0.33, ..., 0.07] (128 numbers describing face)

↓

4Face Embedding Comparison

1 vector x 128 features→Compare extracted features to known faces in database→1 vector x N known faces similarity scores

[0.98, 0.12, 0.45, ..., 0.05] (similarity scores)

↓

5Identity Prediction

1 vector x N similarity scores→Pick the known face with highest similarity score→1 label (person’s name or unknown)

"Alice"

Training Trace - Epoch by Epoch

Loss
1.2 |****
0.9 |***
0.7 |**
0.5 |*
0.35| 
Epochs -> 1 2 3 4 5

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	Model starts learning basic face features
2	0.9	0.60	Accuracy improves as model learns better features
3	0.7	0.72	Model refines feature extraction
4	0.5	0.82	Face embeddings become more distinct
5	0.35	0.90	Model achieves good recognition accuracy

Prediction Trace - 5 Layers

Layer 1: Input Image

Layer 2: Face Detection

Layer 3: Feature Extraction

Layer 4: Face Embedding Comparison

Layer 5: Identity Prediction

Model Quiz - 3 Questions

What does the feature extraction step do in face recognition?

A Crops the face from the photo

B Turns the face image into a list of numbers representing unique features

C Compares the face to known faces

D Labels the face with a name

Key Insight

Face recognition works by turning face images into unique number patterns, then matching these patterns to known faces. Training helps the model learn to make these patterns more distinct, improving accuracy.