Understanding the Effective Receptive Field in Deep Convolutional Neural\n Networks

We study characteristics of receptive fields of units in deep convolutional\nnetworks. The receptive field size is a crucial issue in many visual tasks, as\nthe output must respond to large enough areas in the image to capture\ninformation about large objects. We introduce the notion of an effective\nreceptive field, and show that it both has a Gaussian distribution and only\noccupies a fraction of the full theoretical receptive field. We analyze the\neffective receptive field in several architecture designs, and the effect of\nnonlinear activations, dropout, sub-sampling and skip connections on it. This\nleads to suggestions for ways to address its tendency to be too small.\n