How deeply does Deep Learning with R, Third Edition cover strided convolution?

Explains the practical role and architecture placement, but does not derive convolution-stride mechanics mathematically.

Strided convolution

As explained in Deep Learning with R, Third Edition

A strided convolution is a convolution that moves its kernel by more than one input location at a time to reduce spatial resolution.

A strided convolution moves its kernel by more than one input location at a time, downsampling a feature map while continuing to extract features. This is useful when spatial location matters: unlike 2 × 2 max pooling, it preserves more information about where features occur. section 11.3.2

Compressing an image for segmentation

An image-segmentation model can use strided Conv2D layers to reduce spatial dimensions three times, producing a compact feature map such as (25, 25, 256). Conv2DTranspose layers then expand that representation back to the input size so the model can predict a class for every pixel. In R/Keras, a typical encoder uses `conv(64, 3, strides = 2)`, followed by further strided convolutions, while the decoder uses corresponding transpose convolutions. section 11.3.2

The same choice appears in object detection. A YOLO model uses a strided convolution after a ResNet backbone to reduce feature maps from (14, 14, 2048) to (6, 6, 512), making the representation manageable for dense prediction layers. section 12.2.2

How Deep Learning with R, Third Edition covers it

Cover of Deep Learning with R, Third Edition

liveBook manning.com

Read Deep Learning with R, Third Edition on liveBook Get the book at manning.com

Also explained in

liveBook manning.com

Inside Deep Learning: Math, Algorithms, ModelsA convolution performed with stride greater than one so it reduces spatial resolution instead of using a separate pooling layer.

Compare how 2 books explain strided convolution →

Related concepts

max poolingcontrasts with · Max pooling is an aggregation operation that replaces each local or grouped region with its maximum value, often to downsample a representation.
convolutionkind of · Convolution is an operation that applies a kernel across an input, combining values from overlapping local regions with shared weights to produce outputs.
downsamplingimplements · Downsampling is the reduction of data quantity, spatial resolution, temporal resolution, or sampling rate by retaining, selecting, or aggregating fewer representative values or observations.
strideuses · A stride is the step interval between successive positions or elements accessed, selected, or processed along a dimension or sequence.

This explainer was generated by Manning's knowledge graph from the text of Deep Learning with R, Third Edition. The section links point to the parts of the book it draws on.