Definition

• The receptive field refers to the area of the input data that a particular neuron in a convolutional neural network (CNN) is sensitive to. The size of the receptive field is determined by the size of the filter that the neuron uses to convolve the input data. 

  1. Theoretical receptive field: Provides a basic estimate of the input area impacting the neuron's activation, assuming that there are no interactions between neurons, treating filters independently.

  2. Effective receptive field: Takes into account the actual influence of surrounding pixels on the neuron's response.

• For example, if a neuron has a 3x3 filter, then its receptive field will be a 3x3 patch of the input data. 

Increase Receptive Field

• Adding more layers => increase receptive field size.

• Sub-sampling and Dilated Convolutions.

• Understanding the Effective Receptive Field in Deep Convolutional Neural Networks.

• Dilated CNNs: More Dilations or Larger Filters?.

• Understanding Deep Convolutional Neural Networks with Visual Receptive Fields.

• Rethinking Stride: An Analysis of Convolutional Neural Network Architectures with Non-Strided Convultion Layers.

• Towards a Rigorous Understanding of Receptive Fields in Convolutional Neural Networks.

Spatial Pyramid Pooling (SPP) 

• Integrates Spatial Pyramid Matching (SPM) [39] into the CNN to output one fixed-size dimensional feature vector.

Atrous Spatial Pyramid Pooling (ASPP) 

• The ASPP is the SPP module with dilated convolution operation. 

Receptive Field Block (RFB) 

• Improve dilated convolution to obtain more comprehensive spatial coverage of feature maps.