Understanding the receptive field of deep convolutional networks
Understanding the receptive field of deep convolutional networks
In this article, we will discuss multiple perspectives that involve the receptive field of a deep convolutional architecture. We will address the influence of the receptive field starting for the human visual system. As you will see, a lot of terminology of deep learning comes from neuroscience. As a short motivation, convolutions are awesome but it is not enough just to understand how it works. The idea of the receptive field will help you dive into the architecture that you are using or developing . If you are looking for an in-depth analysis to understand how you can calculate the receptive field of your model as well as the most effective ways to increase it, this article was made for you. In the end, fundamentals are to be mastered! Let’s begin.
According to Wikipedia [1], the receptive field (of a biological neuron) is “ the portion of the sensory space that can elicit neuronal responses, when stimulated” . The sensory space can be defined in any dimension (e.g. a 2D perceived image for an eye). Simply, the neuronal response can be defined as the firing rate (i.e. number of action potentials generated by a neuron). It is related to the time dimension based on the stimuli. What is important is that it affects the received frames per second (FPS) of our visual system. It is not clear what is the exact FPS of our visual system, and it is definitely changing in different situations (i.e. when we are in danger). Wikipedia [2] says:
Insight: The human visual system can process 10 to 12 images per second and perceive them individually, while higher rates are perceived as motion.
Let’s observe this image to further clarify these concepts:
The visual human system. Source: brainconnection
Based on the image, the entire area (the grid in the figure) an eye can see is called the field of view. The human visual system consists of millions of neurons, where each one captures different information. We define the neuron’s receptive field as the patch of the total field of view. In other words, what information a single neuron has access to . This is in simple terms the biological cell’s receptive field.
Let’s see how we can extend this idea in convolutional networks.
Contents
What is the receptive field in deep learning?
Why do we care about the receptive field of a convolutional network?
Closed-form calculations of the receptive field for single-path networks
How can we increase the receptive field in a convolutional network?
Skip-connections and receptive field
Receptive field and transposed convolutions, upsampling, separable convolutions, and batch normalization
Understanding the effective receptive field
What is the receptive field in deep learning?
Similarly, in a deep learning...
