# pix2pixBEGAN.pytorch

**Repository Path**: shuanghaochen/pix2pixBEGAN.pytorch

## Basic Information

- **Project Name**: pix2pixBEGAN.pytorch
- **Description**: A pytorch implementation of pix2pix + BEGAN (Boundary Equilibrium Generative Adversarial Networks)
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2019-12-14
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# pix2pix + BEGAN
- [Image-to-Image Translation with Conditional Adversarial Nets](https://phillipi.github.io/pix2pix/)
- [BEGAN: Boundary Equilibrium Generative Adversarial Networks](https://arxiv.org/abs/1703.10717)

# Install
- install [pytorch](https://github.com/pytorch/pytorch) and [pytorch.vision](https://github.com/pytorch/vision)

# Dataset
- [Download images from author's implementation](https://github.com/phillipi/pix2pix)
- Suppose you downloaded "facades" dataset in ```/path/to/facades```

# Train
- **pix2pixGAN**
- ```CUDA_VISIBLE_DEVICES=x python main_pix2pixgan.py --dataroot /path/to/facades/train --valDataroot /path/to/facades/val --exp /path/to/a/directory/for/checkpoints```
- **pix2pixBEGAN**
- ```CUDA_VISIBLE_DEVICES=x python main_pix2pixBEGAN.py --dataroot /path/to/facades/train --valDataroot /path/to/facades/val --exp /path/to/a/directory/for/checkpoints```
- [Most of the parameters are the same](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/main_pix2pixBEGAN.py#L37-L48) for a fair comparision.
- The original pix2pix is modelled as a conditional GAN, however we didn't. Input samples are not given in D([Only target samples are given](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/main_pix2pixBEGAN.py#L175))
- We used the [image-buffer](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/main_pix2pixBEGAN.py#L178)(analogyous to replay-buffer in DQN) in training D.
- Try other datasets as your need. Similar results will be found.

# Training Curve(pix2pixBEGAN)
- **L_D and L_G \w BEGAN**

![loss](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/BEGAN_loss_niter500.png)

- We found out both L_D and L_G are balanced consistently(equilibrium parameter, gamma=0.7) and converged, even thought [network D](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/models/pix2pixBEGAN.py#L37) and [G](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/models/pix2pixBEGAN.py#L84) are different in terms of model capacity and detailed layer specification.

- **M_global**

![Mglobal](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/BEGAN_Mglobal_niter500.png)

- As the author said, M_global is a good indicator for monitoring convergence.
- [Parsing log](http://htmlpreview.github.io/?https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/pix2pixBEGAN.html): train-log file will be saved in the driectory, you specified, named as ```train.log```

- **L_D and L_G \w GAN**

![BEGAN_loss](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/GAN_loss_niter400.png)

# Comparison
- **pix2pixGAN vs. pix2pixBEGAN**
- ```CUDA_VISIBLE_DEVICES=x python compare.py --netG_GAN /path/to/netG.pth --netG_BEGAN /path/to/netG.pth --exp /path/to/a/dir/for/saving --tstDataroot /path/to/facades/test/```
![failure](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/failure.png)
![GANvsBEGAN](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/pix2pixGAN_vs_pix2pixBEGAN.png)
- [Checkout more results](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/comparison.jpg)(order in input, real-target, fake(pix2pixBEGAN), fake(pix2pixGAN))
- Interpolation on the input-space.
- ```CUDA_VISIBLE_DEVICES=x python interpolateInput.py --tstDataroot ~/path/to/your/facades/test/ --interval 14 --exp /path/to/resulting/dir --tstBatchSize 4 --netG /path/to/your/netG_epoch_xxx.pth```
- Upper rows: pix2pixGAN, Lower rows: pix2pixBEGAN
![interpolation](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/interpolated_pix2pixGAN_pix2pixBEGAN.png)

# Showing reconstruction from D and generation from G
- (order in input, real-target, reconstructed-real, fake, reconstructed-fake) 
![reconDandGenG](https://github.com/taey16/pix2pixBEGAN.pytorch/blob/master/imgs/generated_epoch_00000998_iter00399500.jpg)


# Reference
- [pix2pix.pytorch](https://github.com/taey16/pix2pix.pytorch)
- [BEGAN in pytorch](https://github.com/sunshineatnoon/Paper-Implementations/tree/master/BEGAN)
- [A simple conditional version of BEGAN](https://github.com/taey16/CBEGAN)
- fantastic [pytorch](http://pytorch.org/docs/)

# misc.
- We apologize for your inconvenience when cloning this project. Size of resulting images are huge. please be patient.(Downloading zip file seems to need less time.)