Methods › General › Normalization › Instance Normalization

Instance Normalization

572 papers tagged archive 2025-07-28

Introduced by Dmitry Ulyanov et al. in Instance Normalization: The Missing Ingredient for Fast Stylization

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

Instance Normalization (also known as contrast normalization) is a normalization layer where:

yₜᵢⱼₖ = (xₜᵢⱼₖ - μₜᵢ)/(√(σₜᵢ² + ϵ)), μₜᵢ = 1/HW∑ₗ₌₁ᵂ ∑ₘ₌₁ᴴ xₜᵢₗₘ, σₜᵢ² = 1/HW∑ₗ₌₁ᵂ ∑ₘ₌₁ᴴ (xₜᵢₗₘ - μₜᵢ)².

This prevents instance-specific mean and covariance shift simplifying the learning process. Intuitively, the normalization process allows to remove instance-specific contrast information from the content image in a task like image stylization, which simplifies generation.

PaperSourceSee Code · pytorch/pytorch

Papers archive 2025-07-28

30 shown of 572, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

20 shown of 340 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Translation116
Image-to-Image Translation99
Image Generation61
Style Transfer61
Domain Adaptation56
Semantic Segmentation53
Generative Adversarial Network45
Data Augmentation39
Segmentation38
Unsupervised Domain Adaptation26
Object Detection25
SSIM24
Domain Generalization23
object-detection23
Image Segmentation22
Denoising19
Computed Tomography (CT)18
Diagnostic18
Disentanglement16
Image Classification16

Usage over time archive 2025-07-28

Papers per year tagged with Instance Normalization: 2016 to 2025, peak 92 92 0 2016: 1 paper 2016 2017: 14 papers 2017 2018: 35 papers 2018 2019: 68 papers 2019 2020: 92 papers 2020 2021: 91 papers 2021 2022: 81 papers 2022 2023: 77 papers 2023 2024: 81 papers 2024 2025: 32 papers 2025
Papers per year the archive tags with this method, by the paper's archive date (572 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Normalization

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections