How can Tensorflow be used to create a convolutional base using Python?

A convolutional neural network generally consists of a combination of Convolutional layers, Pooling layers, and Dense layers. TensorFlow with Keras provides an easy way to create these networks using the Sequential API.

Read More: What is TensorFlow and how Keras work with TensorFlow to create Neural Networks?

Creating a Convolutional Base

The convolutional base is the feature extraction part of a CNN. It uses Conv2D and MaxPooling2D layers to progressively reduce spatial dimensions while increasing feature depth ?

from tensorflow.keras import models, layers

print("Creating the convolutional base")
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))

print("Description of architecture is")
model.summary()

Creating the convolutional base
Description of architecture is
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 30, 30, 32)       896       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 15, 15, 32)       0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 13, 13, 64)       18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 6, 6, 64)         0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 4, 4, 64)         36928     
=================================================================
Total params: 56,320
Trainable params: 56,320
Non-trainable params: 0

Architecture Breakdown

The convolutional base follows a common pattern of stacking Conv2D and MaxPooling2D layers:

• Input Shape: (32, 32, 3) ? designed for CIFAR-10 images with 32×32 pixels and 3 color channels

• Conv2D Layers: Extract features using 3×3 filters with ReLU activation

• MaxPooling2D: Reduces spatial dimensions by taking maximum values in 2×2 regions

• Feature Maps: Start with 32 filters, then increase to 64 for deeper feature extraction

How It Works

As the network gets deeper, spatial dimensions (height, width) shrink from 32×32 to 4×4, while the number of feature channels increases from 3 to 64.

Conclusion

This convolutional base creates a feature extraction pipeline that progressively reduces spatial dimensions while increasing feature depth. The pattern of Conv2D followed by MaxPooling2D is fundamental to CNN architecture design.