Generative-Model
Generative-Model copied to clipboard
Repository for implementation of generative models with Tensorflow 1.x
Generative Models
-
This repository is for implementation of
Generative Models
using Tensorflow 1.12 -
The structure of the code is based on the Hwalsuk Lee's Generative Model github repository
Contributors
MMC Lab GAN Study Group members
Implemented Paper List (20 Papers)
GAN
- [GAN] Generative Adversarial Networks
- [DCGAN] Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
- [LSGAN] Least Squares Generative Adversarial Networks
- [WGAN] Wasserstein GAN
- [WGAN_GP] Improved Training of Wasserstein GANs
- [CGAN] Conditional Generative Adversarial Nets
- [InfoGAN] Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets
- [HoloGAN] Unsupervised Learning of 3D Representations From Natural Images
- [SinGAN] Learning a Generative Model from a Single Natural Image
- [PGGAN] Progressive Growing of GANs for Improved Quality, Stability, and Variation
- [StyleGAN] A Style-Based Generator Architecture for Generative Adversarial Networks
Image-to-Image Translation
- [CycleGAN] Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
- [AGGAN] Attention-Guided Generative Adversarial Networks for Unsupervised Image-to-Image Translation
- [StarGAN] Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation
- [DMIT] Multi-mapping Image-to-Image Translation via Learning Disentanglement
Interpretable GAN Latent
VAE
- Auto-Encoding Variational Bayes (VAE)
- Beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework
- Neural Discrete Representation Learning(VQ-VAE)
Application
Our Results
GAN Results
1. GAN
MNIST
![](https://github.com/Kyushik/Generative-Model/raw/master/image/GAN_MNIST.png)
2. DCGAN
MNIST | CelebA |
---|---|
![]() |
![]() |
3. LSGAN
MNIST | CelebA |
---|---|
![]() |
![]() |
4. WGAN
MNIST | CelebA |
---|---|
![]() |
![]() |
5. WGAN-GP
MNIST | CelebA |
---|---|
![]() |
![]() |
6. Conditional GAN
MNIST
![](https://github.com/Kyushik/Generative-Model/raw/master/image/ConditionalGAN_MNIST.png)
7. InfoGAN
MNIST
![](https://github.com/Kyushik/Generative-Model/raw/master/image/InfoGAN_MNIST.png)
8. HoloGAN
CelebA
![](https://github.com/Kyushik/Generative-Model/raw/master/image/HoloGAN_CelebA.png)
9. SinGAN
Balloon
![](https://github.com/Kyushik/Generative-Model/raw/master/image/SinGAN_Ballon_1.png)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/SinGAN_Ballon_0.png)
Mountain
![](https://github.com/Kyushik/Generative-Model/raw/master/image/SinGAN_Mountain_1.png)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/SinGAN_Mountain_0.png)
Starry Night
![](https://github.com/Kyushik/Generative-Model/raw/master/image/SinGAN_Starry_0.png)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/SinGAN_Starry_1.png)
10. PGGAN
It took about 2 weeks on TITAN RTX and trained 600k images per stage.
1024x1024 images
Cherry picked images
Latent interpolation
![](https://github.com/Kyushik/Generative-Model/raw/master/image/PGGAN_inter.png)
Fixed latent
![](https://github.com/Kyushik/Generative-Model/raw/master/image/PGGAN_latent.gif)
No cherry picked images
![](https://github.com/Kyushik/Generative-Model/raw/master/image/PGGAN_NC.png)
11. StyleGAN
CelebA HQ (512x512 images)
Selected images
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StyleGAN_selected_result.png)
Style Mixing with Latent Codes
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StyleGAN_selected_style.png)
Random Images
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StyleGAN_random_result.png)
AFHQ (512x512 images)
Selected images
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StyleGAN_AFHQ_selected_result.png)
Style Mixing with Latent Codes
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StyleGAN_AFHQ_selected_style.png)
Random Images
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StyleGAN_AFHQ_random_result.png)
Image-to-Image Translation Results
1. CycleGAN
Monet to Photo | Photo to Monet |
---|---|
![]() |
![]() |
Horse to Zebra | Zebra to Horse |
---|---|
![]() |
![]() |
2. AGGAN
Horse to Zebra | Zebra to Horse |
---|---|
![]() |
![]() |
3. StarGAN
CelebA
![](https://github.com/Kyushik/Generative-Model/raw/master/image/StarGAN_CelebA.png)
4. DMIT
Summer2Winter
![](https://github.com/Kyushik/Generative-Model/raw/master/image/DMIT.png)
Interpretable GAN Latent
1. Unsupervised Discovery of Interpretable Directions in the GAN Latent Space
1) MNIST
![](https://github.com/Kyushik/Generative-Model/raw/master/image/unsupervised_interpretable_latent.png)
VAE Results
1. VAE
Reconstruction
MNIST | CelebA |
---|---|
![]() |
![]() |
Latent Space Interpolation (MNIST)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/VAE_Latent_MNIST.png)
Latent Space Interpolation (CelebA)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/VAE_Latent_CelebA.png)
2. Beta-VAE
Latent Space Interpolation: Beta = 10 (CelebA)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/betaVAE_10.png)
Latent Space Interpolation: Beta = 200 (CelebA)
![](https://github.com/Kyushik/Generative-Model/raw/master/image/betaVAE_200.png)
3. VQ-VAE
Reconstruction (MNIST)
Input | Reconstruction |
---|---|
![]() |
![]() |
Reconstruction (CelebA)
Input | Reconstruction |
---|---|
![]() |
![]() |
PixelCNN Trained Latent Decoding
MNIST | CelebA |
---|---|
![]() |
![]() |
Application Results
1. Raindrop Removal
![](https://github.com/Kyushik/Generative-Model/raw/master/image/Raindrop_Removal.png)