Convolutional Layers

Introduction

Convolutional layers detect spatial patterns in images through local receptive fields and weight sharing.

Conv2D Layer

from tensorflow.keras import layers
import tensorflow as tf

# Basic Conv2D
model = keras.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    layers.MaxPooling2D((2, 2)),
    layers.Conv2D(64, (3, 3), activation='relu'),
    layers.MaxPooling2D((2, 2)),
    layers.Flatten(),
    layers.Dense(10, activation='softmax')
])

Padding

# Valid (no padding) - output smaller
layers.Conv2D(32, (3, 3), padding='valid')

# Same - output same size as input
layers.Conv2D(32, (3, 3), padding='same')

# Causal - for 1D
layers.Conv1D(32, 3, padding='causal', activation='relu')

Stride

# Stride of 2 reduces dimensions
layers.Conv2D(32, (3, 3), strides=(2, 2), padding='same')

# Dilated convolutions
layers.Conv2D(32, (3, 3), dilation_rate=(2, 2))

Separable Convolution

# More efficient than Conv2D
layers.SeparableConv2D(32, (3, 3), activation='relu')

# Depthwise + pointwise
layers.DepthwiseConv2D((3, 3), depth_multiplier=1)
layers.Conv2D(32, (1, 1))  # Pointwise

Practice Problems

1. Build CNN for image classification
2. Use different padding modes
3. Apply strided convolutions
4. Implement separable convolutions
5. Add multiple conv layers