Computing Receptive Fields of Convolutional Neural Networks

By Andre F. de Araújo, W. Norris, Jack Sim. Distill, 2019

323 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/150f1260a3b73866865626a0e99b486242d32fc9

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

YOLACT++ Better Real-Time Instance Segmentation (2019, background): “For instance, ResNet 101 and ResNet 50 have 511px and 239px of padding in each direction respectively [50].”
Unifying Deep Local and Global Features for Image Search (2020, methodology): “The feature positions with maximum activations can be mapped back to the relevant input image regions by computing the model’s receptive field parameters [2].”
SLEAP: A deep learning system for multi-animal pose tracking (2022, background): “These types of models are composed of building blocks that imbue the network with different capabilities, such as increasing the maximum receptive field size (RF) to enable the model to learn and reason about image features across spatial scales 20 .”
MLP-Mixer: An all-MLP Architecture for Vision (2021, background): “Neurons in deeper layers have a larger receptive field [1, 29].”
Spatial-Phase Shallow Learning: Rethinking Face Forgery Detection in Frequency Domain (2021, background): “For this purpose, SPSL drops many convolutional layers to reduce the receptive field [3] and forces CNNs to pay more attention to local regions which are abundant in textures and lack high-level semantic information.”
On Completeness-aware Concept-Based Explanations in Deep Neural Networks (2019, methodology): “Here, each x ij corresponds to a 164 ˆ 164 square in the input image (with effective stride 16), which is the receptive ﬁeld of convolution layer 5 of VGG-16 [43].”
On Translation Invariance in CNNs: Convolutional Layers Can Exploit Absolute Spatial Location (2020, background): “Yet, CNNs stack convolution layers, yielding receptive fields typically several times the input image size [4].”