オムライスの備忘録

数学・統計学・機械学習・プログラミングに関することを記す

トップ > データサイエンス > 【深層学習】Generative Adversarial Network / GAN #まとめ編 #00

【深層学習】Generative Adversarial Network / GAN #まとめ編 #00

データサイエンス データサイエンス-深層学習

Index

Generative Adversarial Network / GAN

深層学習を用いた生成モデルの一つ.

基本アルゴリズム / 2014

応用手法

DC GAN / 2015 ★

Conditional GAN / 2014 ★



Semi Supervised GAN / 2016

Info GAN / 2016

  • InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets

  • 【論文メモ：InfoGAN】InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets

Auxiliary Classifier GAN / AC GAN / 2016

Context Conditional GAN / 2016

  • Semi-Supervised Learning with Context-Conditional Generative Adversarial Networks

Age C GAN / 2017

  • Face Aging With Conditional Generative Adversarial Networks

Coupled GAN / Co GAN / 2016 ★

  • Coupled Generative Adversarial Networks

Energy Based GAN / EB GAN / 2016

目的関数の工夫.

  • Energy-based Generative Adversarial Network

Least Squares GAN / LS GAN / 2016

目的関数の工夫.

  • Least Squares Generative Adversarial Networks

Super Resolution GAN / SR GAN / 2016

  • Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

  • Super-Resolution Generative Adversarial Networks (SRGAN)

Enhanced Super Resolution GAN / ESR GAN / 2018

  • ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks

Wasserstein GAN / W GAN / 2017 - ★

Boundary Equilibrium GAN / BE GAN / 2017

Discriminator を Autoencoder にする.

  • BEGAN: Boundary Equilibrium Generative Adversarial Networks

  • 【論文メモ:BEGAN】BEGAN: Boundary Equilibrium Generative Adversarial Networks

Boundary Seeking GAN / B GAN / 2017

  • Boundary-Seeking Generative Adversarial Networks

Softmax GAN / 2017

目的関数の工夫.

Progressive Growing GAN / Pro GAN / PG GAN / 2017 ★

Relativistic GAN / RGAN / 2018

目的関数の工夫.

  • The relativistic discriminator: a key element missing from standard GAN

Spectral Normalization GAN / SN GAN / 2018

Cluster GAN / 2018 ★

Self Attention GAN / SA GAN / 2018 ★

Big GAN / 2018 ★

Style GAN / 2018 - ★

  • Style GAN #まとめ編
    • Style GAN 2 / 2019
    • Style Space / 2020
    • Style GAN 3 / Alias Free GAN / 2021
    • StyleGAN-NADA / 2021
    • StylEx / 2021
    • 応用
      • pSp / 2020
      • Style CLIP / 2021
    • yhayato1320.hatenablog.com

DeSIGN / 2018

  • DeSIGN: Design Inspiration from Generative Networks

  • コンピュータによる創造性の評価を提案！GANによる創造的なファッションデザインの生成

SphereGAN / 2019

GauGAN / 2019

SinGAN / 2019

一枚の画像から学習.

  • SinGAN: Learning a Generative Model from a Single Natural Image

ICR GAN / 2020

  • Improved Consistency Regularization for GANs

  • GANへの新しい正則化「ICR」が期待大な件&解説

Trans GAN / 2021

Adaptive Pseudo Augmentation / APA / 2021

  • Deceive D: Adaptive Pseudo Augmentation for GAN Training with Limited Data

  • GANで限られたデータから高精度画像を生成

VQ GAN / 2021

Fourier Space Loss / 2021

高解像度 (Super Resolution) 画像 の生成における工夫.

  • Fourier Space Losses for Efficient Perceptual Image Super-Resolution

Projected GAN / 2021

TsT GAN / 2022

  • Time-series Transformer Generative Adversarial Networks

  • Researchers From Imperial College London introduce TsT-GAN: A Novel Framework For Training Time-Series Generative Models

CoordGAN / 2022

  • CoordGAN: Self-Supervised Dense Correspondences Emerge from GANs

  • UCSD and NVIDIA AI Researchers Propose ‘CoordGAN’: a Novel Disentangled GAN Mode That Produces Dense Correspondence Maps Represented by a Novel Coordinate Space

Autoencoder Base

Autoencoder をアーキテクチャのベースにして、Adversarial Loss を使って学習する.

Adversarial Autoencoder / 2015

Context Encoder / 2016

  • Context Encoders: Feature Learning by Inpainting

工夫・テクニック

データ分野・タスク

画像処理

スタイル変換 / Style Transfer

異常検知 / Anomaly Detection

AnoGAN / 2017

  • Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery

  • GANはただの画家ではない！鑑定士への転身を果たすAnoGAN

3D データ

時系列解析

音響処理

BigVGAN / 2022

  • BigVGAN: A Universal Neural Vocoder with Large-Scale Training

MultiModal

性能評価

Inception Score / IS

Frechet Inception Distance / FID

Inception-v3モデルを使って、本物と生成された画像それぞれの埋め込み表現を計算し、それらの平均と共分散を比較し、その距離が近い、つまり値が小さいほど良いスコア



  • GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

  • Frechet Inception Distance(FID)を理解する

参考

  • Self-supervised Learning: Generative or Contrastive

    • [2020]
    • 6 Theory behind Self-supervised Learning
      • 6.1 GAN
    • arxiv.org

  • A Survey on Generative Adversarial Networks: Variants, Applications, and Training

  • A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications

書籍

Web サイト

  • GAN(敵対的生成ネットワーク)について説明します!

  • GAN（敵対的生成ネットワーク）とは｜意味・仕組み・応用例

survey

動画

  • Generative Adversarial Networks and TF-GAN (ML Tech Talks)