receptive field

The region of the original input image that a given unit's activation is actually influenced by.

A unit's receptive field is the region of the original input image that its activation actually depends on. In a CNN, receptive field grows with depth -- each additional convolutional or pooling layer lets a unit see a larger patch of the input indirectly, through the units it's stacked on top of, even though any one layer's own kernel only looks at a small local window.

Examples

Stacking three $3\times3$ convolutional layers gives a unit a $7\times7$ receptive field on the original input, even though any one of those layers only looks at a $3\times3$ window directly.