Evaluation and confusion matrix in Computer Vision - Model Pipeline Trace

Model Pipeline - Evaluation and confusion matrix

This pipeline shows how a computer vision model is evaluated using a confusion matrix. It helps us understand how well the model predicts different classes by comparing predictions to true labels.

Data Flow - 5 Stages

1Input Images

1000 images x 64x64 pixels x 3 color channels→Raw images loaded for classification→1000 images x 64x64 pixels x 3 color channels

Image of a cat, Image of a dog, Image of a bird

↓

2Preprocessing

1000 images x 64x64 x 3→Resize and normalize pixel values (0-1)→1000 images x 64x64 x 3

Pixel values scaled from 0-255 to 0-1

↓

3Feature Extraction

1000 images x 64x64 x 3→Extract features using CNN layers→1000 samples x 128 features

Feature vector representing edges, textures

↓

4Model Prediction

1000 samples x 128 features→Fully connected layers output class probabilities→1000 samples x 3 classes

[0.7, 0.2, 0.1] for cat, dog, bird

↓

5Evaluation

1000 samples x 3 classes→Compare predicted classes to true labels→Confusion matrix 3x3

Matrix showing counts of true vs predicted classes

Training Trace - Epoch by Epoch

Loss
1.2 |*       
0.9 | *      
0.7 |  *     
0.5 |   *    
0.4 |    *   
    +---------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	Model starts learning, accuracy low
2	0.9	0.60	Loss decreases, accuracy improves
3	0.7	0.72	Model learns better features
4	0.5	0.80	Good improvement in accuracy
5	0.4	0.85	Model converges with good accuracy

Prediction Trace - 5 Layers

Layer 1: Input Image

Layer 2: Feature Extraction (CNN)

Layer 3: Fully Connected Layer

Layer 4: Prediction

Layer 5: Confusion Matrix Update

Model Quiz - 3 Questions

What does a confusion matrix show in model evaluation?

A Counts of true vs predicted classes

B The model's training loss over epochs

C The input image sizes

D The number of features extracted

Key Insight

The confusion matrix is a simple but powerful tool to see where a model makes mistakes. It helps us understand if the model confuses certain classes more than others, guiding improvements.