morristech / gan-papers

GAN papers

Recommendation

• Our GAN based work for facial attribute editing - https://github.com/LynnHo/AttGAN-Tensorflow.

---

---

Frameworks

1. GAN, DCGAN
2. cGAN - Label-conditioning generation
3. AcGAN - Supervised generation (auxiliary classifier with labels)
4. SGAN - D outputs [CLASS-1, CLASS-2, . . . CLASS-N, FAKE]
5. InfoGAN - Unsupervised lantern space disentangling
6. ALI, BiGAN - match p(G(z),z) and q(x,E(x))
7. GMAN - Multiple discriminator models
8. AdaGAN - Multiple generative models
9. MGGAN - Multiple generative models

---

High Resolution

1. LAPGAN - Coarse-to-fine generation
2. StackGAN - Two-step generation
3. PGGAN

---

Function ('~' means 'match')

g(Z) ~ X

1. GAN, DCGAN

g(Z|Y) ~ X|Y, where Y is label or lantern variable

1. cGAN
2. AcGAN
3. InfoGAN
4. CFGAN

g(Z),Z ~ X,E(X)

1. ALI, BiGAN

---

Objective Function

f-Divergence

1. GAN, DCGAN - JS divergence
2. LSGAN (Least Squares GAN) - Pearson χ2 divergence
3. f-GAN - Variational divergence minimization
4. f-GANs

Integral Probability Metrics (IPM)

1. WGAN - Wasserstein distance
   • WGAN-GP - Gradient penalty, less capacity compromise
2. McGAN - Mean and covariance matching
3. Geometric GAN
4. Fisher GAN - Chi-squared distance

Maximum Mean Discrepancy (MMD)

1. GMMN
2. MMD nets

Others

1. CatGAN - Entropy of P(Y|X)
2. EBGAN - D(x) = ||Dec(Enc(x)) − x||
3. LS-GAN (Loss-Sensitive GAN)
4. BEGAN - Wasserstein distance between the distribution of real/fake auto-encoder loss
5. AGE - $max_emin_g\Delta(e(g(Z))||Y)-\Delta(e(X)||Y)$ and encoder-generator reciprocity (bidirectional mapping)
6. Softmax GAN
7. Cramér GAN - Cramér Distance
8. LDGAN

---

Regularized GAN

1. DRGAN - Vanilla GAN with gradient penalty
2. Cramér GAN - Cramér Distance
3. Regularized GAN
4. CT-GAN
5. Varying k-Lipschitz Constraint for Generative Adversarial Networks

---

Representation Learning

Lantern Space Disentangling

1. InfoGAN - Unsupervised lantern space disentangling
2. AcGAN - Supervised lantern space disentangling

Specifying Lantern Space Distribution

1. AAE
2. ALI, BiGAN - Match p(G(z),z) and q(x,E(x)), simultaneously learn an encoder and decoder
3. AGE

---

Semi-supervised Learning

1. CatGAN
2. SGAN - D outputs [CLASS-1, CLASS-2, . . . CLASS-N, FAKE]
3. Improved GAN
4. Triple-GAN
5. CT-GAN

---

Evaluations of GANs

1. Inception Score - Improved GAN
2. FCN Score - pix2pix
3. AMT Perceptual Studies - pix2pix
4. Semantic Segmentation Metrics - CycleGAN
5. FID, Precision, Recall and F1 Score - Are GANs Created Equal? A Large-Scale Study

---

Applications

Data Augmentation

1. Adversarial Generation of Training Examples for Vehicle License Plate Recognition

Domain Adaptation

1. Unsupervised Pixel-Level Domain Adaptation with Generative Adversarial Networks
2. DTN
3. UNIT
4. CoGAN

Image Denoising

1. Deep Semantic Face Deblurring

Image-to-Image Translation

Unpaired

1. CoGAN
2. DTN
3. CycleGAN, DiscoGAN, DualGAN
4. UNIT
5. Face Transfer with Generative Adversarial Network
6. XGAN - Semantic consistency

Paired

1. pix2pix/PatchGAN
2. Scribbler
3. PAN - Perceptual adversarial loss
4. Cross-View Image Synthesis using Conditional GANs

Inpainting

1. Context Encoder
2. PAN

Super-Resolution

1. SRGAN

Text-to-Image Synthesis

1. GAN-INT-CLS
2. GAWWN
3. StackGAN

Face Editing

1. VAE/GAN - Visual attribute vectors
2. IcGAN
3. DIAT
4. Learning Residual Images for Face Attribute Manipulation
5. DistanceGAN
6. CFGAN
7. Age-cGAN
8. UNIT
9. SL-GAN
10. IAN
11. Neural Face Editing with Intrinsic Image Disentangling
12. GeneGAN - Object transfiguration
13. Fader Networks
14. Unsupervised Visual Attribute Transfer with Reconfigurable Generative Adversarial Networks
15. ExprGAN
16. StarGAN
17. GLCA-GAN

Face Frontalization/Profiling

1. DR-GAN
2. TP-GAN

Others

1. iGAN - Image manipulation
2. TVSN - 3D view synthesis
3. ID-CGAN - Image de-raining
4. Perceptual GAN - Small object detection

---

Survey

1. Generative Adversarial Networks: An Overview
2. How Generative Adversarial Nets and its variants Work
3. Comparative Study on Generative Adversarial Networks
4. An Introduction to Image Synthesis with Generative Adversarial Nets

---

Unclassified

1. DeePSiM
2. Unrooled GAN
3. SGAN
4. AM-GAN
5. DeLiGAN - Mixture Gaussian prior distribution
6. TTUR

---

Datasets

1. MNIST
2. CIFAR10/CIFAR100
3. SVHN
4. CelebA