Contents
What is the effective receptive field of a convolution layer?
As a natural consequence, one can define the relative importance of each input pixel as the effective receptive field (ERF) of the feature. In other words, ERF defines the effective receptive field of a central output unit as the region that contains any input pixel with a non-negligible impact on that unit.
What is receptive field in convolution?
The receptive field is defined by the filter size of a layer within a convolution neural network. The receptive field is also an indication of the extent of the scope of input data a neuron or unit within a layer can be exposed to (see image below).
What is the receptive field RF of a unit in a convolutional neural network?
What is the Receptive Field (RF) of a unit? The receptive field of a unit is the region of input that is seen and responded to by the unit. When dealing with high-dimensional inputs such as images, it is impractical to connect neurons to all neurons in the previous volume.
How to increase the receptive field of deep convolutional?
In essence, there are a plethora of ways and tricks to increase the RF, that can be summarized as follows: Let’s look at the distinct characteristics of these approaches. Option 1 increases the receptive field size linearly, as each extra layer increases the receptive field size by the kernel size [7].
How are Dilated convolutions increase the RF exponentially?
On the other hand, sequentially placed dilated convolutions, increase the RF exponentially. But first, let’s revisit the idea of dilated convolutions. In essence, dilated convolutions introduce another parameter, denoted as r, called the dilation rate. Dilations introduce “holes” in a convolutional kernel [3].
How does pooling increase the receptive field size?
Sub-sampling techniques like pooling (option 2) on the other hand, increases the receptive field size multiplicatively. Modern architectures like ResNet combine these techniques (option 1 and 2). On the other hand, sequentially placed dilated convolutions, increase the RF exponentially.
How many decimation layers increase the receptive field?
This is already giving us a signal which is halved in temporal resolution, which made us achieve a receptive field of 7 with 1 less layer. This is a very little example of course, but imagine the exponential increase in receptive field when we have 2, 3 or more decimation layers with factor 2 (or more).