Skip to content
    Flow-alerta.cab

    Dilated Convolution (Atrous Convolution)

    AI fundamentals & modelsAtrous convolutionDilated convolution
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 21 words

    In short

    Dilated convolutions increase the receptive field with the same number of filter parameters due to greater distances between the scanning positions.

    Flow-alerta.cab (2026): Dilated Convolution (Atrous Convolution). https://flow-alerta.cab/en/glossar/dilated-convolution-atrous-convolution. Editorially reviewed on: 09 October 2026.

    How much detail would you like?

    Practical

    Relevant for image segmentation and the processing of larger spatial or temporal contexts.

    Possible example: A segmentation model takes a larger image context into account when examining surface images.

    Possible building block for image segmentation or audio analysis. Whether it is suitable for an operational application must be checked on the basis of suitable test data.

    At a glance

    Core
    A dilated convolution increases the distances between the scanning positions of a filter. With the same filter size, a larger context is thereby detected without additional filter parameters. Whether the output resolution remains the same also depends on stride and padding.
    Use
    Relevant for image segmentation and the processing of larger spatial or temporal contexts.
    Example
    Possible example: A segmentation model takes a larger image context into account when examining surface images.
    Limits
    Unfavorable sampling gaps can create lattice artifacts; confuse dilation, stride and padding; do not test the utility on your own data.

    In context

    Convolutional Neural Network (CNN)

    Neural network that uses learnable filters to recognize local patterns in structured data.

    Questions about this term

    Does the resolution always remain?

    No. Dilation increases the context; the output size additionally depends on stride and padding.

    What does the dilation rate mean?

    It determines the distance between the scanning positions of the filter.

    All details and background

    A dilated convolution increases the distances between the scanning positions of a filter. With the same filter size, a larger context is thereby detected without additional filter parameters. Whether the output resolution remains the same also depends on stride and padding.

    ELI5 – Explained simply

    Like a magnifying glass with measuring points further apart: it sees more surface area without becoming heavier.

    Why it matters in practice

    Relevant for image segmentation and the processing of larger spatial or temporal contexts.

    Example

    Possible example: A segmentation model takes a larger image context into account when examining surface images.

    Pitfalls and limits

    • Unfavorable sampling gaps can create lattice artifacts; confuse dilation, stride and padding; do not test the utility on your own data.

    Origin & history

    Yu and Koltun described an approach to multi-scale context aggregation with dilated convolutions in 2015; published as an ICLR conference contribution in 2016.

    Use in mid-sized companies

    • Possible building block for image segmentation or audio analysis. Whether it is suitable for an operational application must be checked on the basis of suitable test data.

    Related terms

    • CNN
    • Receptive field
    • Semantic segmentation
    • WaveNet
    • DeepLab

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI Potential Analysis & Roadmap – Evaluating the Basics of an Image Analysis Application

    Matching tool / use case

    View use casesUse Case: Enlarge the captured context of a segmentation model.

    Sources

    • arxiv.orgYu & Koltun (2015), Multi-Scale Context Aggregation by Dilated Convolutions

    Quellen geprüft am 09.10.2026.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases