What is a feature map in a CNN?

A feature map in a CNN is the output produced when a convolutional filter is applied across an input.

Each feature map shows where a particular learned pattern appears and how strongly the filter responds at different spatial locations.

How is a feature map created?

A convolutional filter slides across the input and computes a response at each position.

Those responses are arranged into a new grid:

Input → Filter → Feature map

If a convolutional layer contains multiple filters, it produces multiple feature maps.

For example:

• Filter 1 may respond strongly to vertical edges.
• Filter 2 may respond to horizontal edges.
• Filter 3 may respond to textures or other local patterns.

Each filter therefore contributes a different channel to the layer’s output.

What happens in deeper layers?

Early feature maps often represent simple visual patterns such as edges and color transitions.

Deeper feature maps can represent more complex structures formed by combining information from earlier layers.

A simplified progression is:

Edges → Textures → Shapes → Higher-level features

Why are feature maps important?

Feature maps allow a CNN to preserve where useful patterns appear while progressively transforming raw input into more meaningful representations.

Their spatial size is affected by settings such as kernel size, stride, and padding.

For related concepts, see “What is a kernel or filter in a CNN?”, “What is stride in a convolutional neural network?”, and “What is padding in a CNN?”.

In short, a feature map is the spatial output of a convolutional filter, showing where the learned pattern represented by that filter is detected in the input.