Course contentsShow
← Machine Learning and Deep Learning
Lesson 879 of 3,53820. Convolutional Neural Networks: FundamentalsFree lesson

What is a Receptive Field?

Define receptive field as the region of input that influences a single neuron's activation in a CNN layer.

What is a Receptive Field?

What you'll learn: You'll understand how each neuron in a CNN "sees" only a limited portion of the original input image, and why this matters for architecture design.

The Core Idea

A receptive field is the region of the input image that influences the activation of a single neuron in a particular CNN layer. Think of it as the neuron's "field of view" — the area it can "see" when making its decision.

When you apply convolution operations with kernels (like the 3×3 filters you've learned about), each output neuron is computed from a small patch of the input. For the first convolutional layer, if you use a 3×3 kernel, each neuron's receptive field is exactly 3×3 pixels from the original image.

How It Grows Deeper

Here's where it gets interesting: as you stack more convolutional layers, the receptive field grows. A neuron in the second layer might directly look at a 3×3 patch of the first layer's output, but since each of those neurons already saw a 3×3 patch of the input, the second-layer neuron indirectly sees a 5×5 region of the original image.

This is why deeper networks can recognize more complex patterns — neurons in later layers have larger receptive fields and can capture broader spatial context.

Why It Matters

Understanding receptive fields helps you design architectures appropriately. If you need to detect large objects or patterns, you need neurons with large receptive fields, which means either deeper networks, larger kernels, or techniques like dilated convolutions and pooling (which you've already learned increase the effective receptive field).

Key Takeaway: The receptive field defines how much of the input image a neuron can "see," and it grows as you go deeper in the network, allowing higher layers to capture increasingly global patterns.