p-geon/GANSurvey

ガンガンサーベイしていこうな

GAN Survey

データセット / Dataset

GANでよく使われるデータセットまとめ

• noise -> image
  • MNIST
  • cifar-10
  • ImageNet
• image -> image
  • apple2orange
  • facade
• text
• audio
• textures
  • DTD

評価指標 / Metrics

• Inception Score
• Frechet Inception Distance
• Sliced Wasserstein distance
• LPIPS distance
• Perceptual Path Length -> StyleGANv2
• Human Study (どっちのほうがよく生成できているか判定)

サーベイ論文 / Survey Paper

• Generative Adversarial Networks in Computer Vision: A Survey and Taxonomy, (GitHub)
• A Survey on Generative Adversarial Networks: Variants, Applications, and Training
• A Mathematical Introduction to Generative Adversarial Nets (GAN)

基礎GAN

• GAN
• DCGAN
• cGAN
• Adversarial Autoencoders

GAN-variant

• Adversarially Learned Inference
• BiGAN
• InfoGAN, NIPS2016
  • ss-InfoGAN, CVPR201?
• f-GAN
• UnrolledGAN, ICLR2017
• DRAGAN
• Dirac-GAN

energy-base

• EBGAN
• Generative Adversarial Networks as Variational Training of Energy Based Models

ロス設計

• LSGAN
  • pros/cons
• LSGAN-gp
• WGAN
• WGAN-gp

出力構造

• ACGAN
• cGANs with Projection Discriminator

Discriminator

• Relativistic Discriminator
• U-Net GAN, CVPR2020

理論寄りの論文 / GAN Logic, Convergence...

• Localized GAN
• Are GANs Created Equal? A Large-Scale Study
• Discriminator Rejection Sampling
• How good is my GAN?
• GANs May Have No Nash Equilibria

正規化 / Normalization

• Batch Normalization
• Instance Normalization
• Layer Normalization
• Spectral Normalization
• AdaIN
• SPADE

学習安定性 / Training Stability

• Stabilizing Training of Generative Adversarial Networks through Regularization
• Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients
• Mode Seeking Generative Adversarial Networks for Diverse Image Synthes
• Smoothness and Stability in GANs, 19.09

ロバストなGAN / Robustness

• RoCGAN, ICLR2019
• RcGAN, NeurIPS2018

データ拡張 / Data Augmentation

GANで使うデータ拡張 / DA for GAN

• Data Augmentation Generative Adversarial Networks
• Image Augmentations for GAN Training

GANをデータ拡張に使う / GAN for NNs

• Data Augmentation Using GANs

少量データによるGANの学習

• SinGAN, ICCV2019bestpaper
• Differentiable Augmentation for Data-Efficient GAN Training
• Towards Faster and Stabilized GAN Training for High-fidelity Few-shot Image Synthesis, ICLR2021(poster)
• Training Generative Adversarial Networks with Limited Data, NeurIPS2020

高解像度系 / High-Resolution Image Synthesis

• LAPGAN, CVPR
• PGGAN
• BigGAN
  • BigBiGAN, TFHub
• StyleGAN
• StyleGANv2, (GitHub)

TGAN系列

• TGAN

画像変換 / Image2Image Translation

• pix2pix
  • pix2pix HD
• CycleGAN
  • BiCycleGAN, paper/poster
• UGATIT, メルアイコン変換器を作った話
• AnimeGAN, git, press
• CountRGAN

AdaIN系

• InstaGAN
• GauGAN(Spade)
• High-DT, paper/git

UNIT系列

• UNIT
• MUNIT
• FUNIT

穴埋めタスク / Filling

• Context Encoders

クロスドメイン / Cross Domain

• DiscoGAN
• StarGAN
  • StarGAN v2
• XFork
• SelectionGAN

超解像 / Single Image Super-Resolution(SISR)

• SRGAN
• ESRGAN, ECCV2018workshop
• SFTGAN, CVPR2018

色付け・色塗り / Colarization

• Style2Paints
• SketchyGAN, CVPR2018

動画生成 / Video Generation

• ReCycleGAN, paper
• Vid2Vid
• few-shot Vid2Vid

テキスト生成系 / Text Generation

• Text Generation Based on Generative Adversarial Nets with Latent Variable

Text2Image

• GAN-GLS
• StackGAN
• StackGAN++
• Obj-GANs

声質変換 / Voice Conversion

• MelGAN

• MBMelGAN

• StyleMelGAN

• HiFi-GAN

• CycleGAN-VC

  • CycleGAN-VC2
  • CycleGAN-VC3

• StarGAN-VC

  • StarGAN-VC2

• Parallel Wave-GAN

音声合成 / Text-to-Speech(TTS)

• GANTTS

3D

• 3DGAN

異常検知 / Anomaly Detection

• AnoGAN
• Dual Auto-Encoders GAN
• Anomaly Detection by One Class Latent Regularized Networks

デコーディング / Brain-Decoding

EEG

• CS-GAN

fMRI

• Generative adversarial networks for reconstructing natural images from brain activity
• End-to-End Deep Image Reconstruction From Human Brain Activity
• DCNN-GAN: Reconstructing Realistic Image from fMRI
• Reconstructing faces from fMRI patterns using deep generative neural networks
• Reconstructing Natural Scenes from fMRI Patterns using BigBiGAN
• BigGAN-based Bayesian reconstruction of natural images from human brain activity

マテリアル系 / Materials Engineering

• https://arxiv.org/pdf/1805.07615.pdf
• Generative Adversarial Networks for Crystal Structure Prediction
• Designing complex architectured materials with generative adversarial networks

Neural Architecture Search (NAS)

• AutoGAN

ポーズ系 / Pose Estimation, Pose Generation

• MoCoGAN
• Liquid Warping GAN, ICCV2019

music2dance

• Dancing to Music
• DeepDance
• Learning to dance: A graph convolutional adversarial network to generate realistic dance motions from audio

Privacy & Fairness

• VGAN-Based Image Representation Learning for Privacy-Preserving Facial Expression Recognition

強化学習 / Reinforcement Learning

• Generative Adversarial Imitation Learning
• Task-Relevant Adversarial Imitation Learning
• Triple-GAIL

VAE

• VAE-GAN
• cVAE-GAN

Flow

• Flow-GAN

---

---

Reading List

• Semi-Supervised GAN, paper
• Efficient Conditional GAN Transfer with Knowledge Propagation across Classes, paper
• Few-Shot Adversarial Learning of Realistic Neural Talking Head Models, paper
• Dual-Agent GAN, paper/poster
• Triple Generative Adversarial Nets, paper
• GraphGAN, paper, (AAAI2018)
• PriorGAN, paper

Quantum

• Quantum GAN, paper
• Quantum generative adversarial learning in a superconducting quantum circuit, paper

軽量化

• GAN Slimming: All-in-One GAN Compression by A Unified Optimization Framework, paper

Transformer

• TransGAN

OCR

• TLGAN

saliency map

• SalGAN

バイオ系/Bioligy

• Gene Expression Imputation with Generative Adversarial Imputation Nets

資料 / Appendix

• KL-Divergence
• JS-Divergence
• Checker Board Archifact
  • Deconvolution and Checkerboard Artifacts (medium)