Why CNNs detect spatial patterns in PyTorch - Model Pipeline Impact

Model Pipeline - Why CNNs detect spatial patterns

This pipeline shows how a Convolutional Neural Network (CNN) learns to detect spatial patterns in images by processing pixel data through convolution layers, pooling, and fully connected layers to classify images.

Data Flow - 5 Stages

1Input Image

1000 rows x 28 columns x 28 pixels x 1 channel→Raw grayscale images of handwritten digits→1000 rows x 28 columns x 28 pixels x 1 channel

Image of digit '7' represented as 28x28 pixel grayscale values

↓

2Convolution Layer

1000 rows x 28 x 28 x 1→Apply 16 filters of size 3x3 to detect edges and simple shapes→1000 rows x 26 x 26 x 16

Feature maps highlighting edges and corners in the digit image

↓

3Pooling Layer

1000 rows x 26 x 26 x 16→Max pooling with 2x2 window to reduce spatial size→1000 rows x 13 x 13 x 16

Smaller feature maps keeping strongest features

↓

4Flatten Layer

1000 rows x 13 x 13 x 16→Flatten 3D feature maps into 1D vectors→1000 rows x 2704 features

Vector representing combined spatial features

↓

5Fully Connected Layer

1000 rows x 2704→Dense layer to classify features into digits 0-9→1000 rows x 10

Output scores for each digit class

Training Trace - Epoch by Epoch

Loss
1.2 |*       
0.8 | **     
0.5 |   ***  
0.3 |     ****
0.2 |      *****
     ----------------
      1  2  3  4  5  Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.55	Model starts learning basic spatial features
2	0.8	0.75	Filters detect clearer edges and shapes
3	0.5	0.85	Pooling helps focus on important features
4	0.3	0.92	Model refines spatial pattern recognition
5	0.2	0.95	High accuracy shows strong spatial pattern detection

Prediction Trace - 5 Layers

Layer 1: Input Image

Layer 2: Convolution Layer

Layer 3: Pooling Layer

Layer 4: Flatten Layer

Layer 5: Fully Connected Layer

Model Quiz - 3 Questions

Why does the convolution layer reduce the image size from 28x28 to 26x26?

A Because the image is resized before convolution

B Because pooling reduces the size

C Because the 3x3 filters cannot slide over the edges without padding

D Because the fully connected layer requires smaller input

Key Insight

CNNs detect spatial patterns by applying filters that scan small regions of the image, capturing edges and shapes. Pooling layers reduce size while preserving important features. Over training, the model learns to recognize these patterns better, improving accuracy.