Feature map visualization in PyTorch - Model Pipeline Trace

Model Pipeline - Feature map visualization

This pipeline shows how an image passes through a simple convolutional neural network. It highlights how feature maps are created and transformed at each convolution layer, helping us understand what the model 'sees' inside the image.

Data Flow - 5 Stages

1Input Image

1 image x 3 channels x 32 height x 32 width→Raw image loaded and normalized→1 image x 3 channels x 32 height x 32 width

A 32x32 color image with 3 color channels (RGB)

↓

2First Convolution Layer

1 x 3 x 32 x 32→Apply 6 filters of size 5x5 with stride 1, padding 0→1 x 6 x 28 x 28

6 feature maps showing edges and simple patterns

↓

3ReLU Activation

1 x 6 x 28 x 28→Apply ReLU to keep positive values, zero out negatives→1 x 6 x 28 x 28

Same 6 feature maps with negative values replaced by zero

↓

4Second Convolution Layer

1 x 6 x 28 x 28→Apply 16 filters of size 5x5 with stride 1, padding 0→1 x 16 x 24 x 24

16 feature maps capturing more complex patterns

↓

5ReLU Activation

1 x 16 x 24 x 24→Apply ReLU activation→1 x 16 x 24 x 24

16 feature maps with negative values zeroed

Training Trace - Epoch by Epoch

Loss
1.2 |****
0.9 |***
0.7 |**
0.5 |*
0.4 |

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	Model starts learning, loss is high, accuracy low
2	0.9	0.60	Loss decreases, accuracy improves as model learns features
3	0.7	0.72	Better feature extraction reflected in improved accuracy
4	0.5	0.80	Model converging, loss dropping steadily
5	0.4	0.85	Good feature maps help model classify better

Prediction Trace - 5 Layers

Layer 1: Input Image

Layer 2: First Convolution Layer

Layer 3: ReLU Activation

Layer 4: Second Convolution Layer

Layer 5: ReLU Activation

Model Quiz - 3 Questions

What does the first convolution layer mainly detect in the input image?

A Noise in the image

B Final classification labels

C Edges and simple shapes

D Color histograms

Key Insight

Feature maps show how the model transforms raw images into meaningful patterns step-by-step. Visualizing these maps helps us understand what the model learns and why accuracy improves during training.