This program is an improved version based on srgan, the based code is in srgan by zsdonghao
- You should download the the pretrained VGG16 model in here and put it in folder below in
config.py
path_vgg16="your_vgg16_folder\\"
- You can download the dataset in DIV2K,and put the dataset in folders below in
config.py
path_train_HR_orin = 'your_dir\\DIV2K_train_HR\\'
path_train_LR_orin = 'your_dir\\DIV2K_train_LR_bicubic\\X4\\'
path_valid_HR_orin = 'your_dir\\DIV2K_valid_HR\\'
path_valid_LR_orin = 'your_dir\\DIV2K_valid_LR_bicubic\\X4\\'
- You should change the parameters suitalbe for your GPU/CPU in
config.py
| parameters | note |
|---|---|
| batch_size_init | 初始化生成器使用的batchsize |
| n_epoch_init | 初始化生成器使用的epoch数目 |
| train_step_init | （初始化生成器）使用所有剪裁好的图片中的1/train_step_init 作为训练集，train_step_init=1表示使用全部图片 |
| lr_init | 初始化生成器使用的学习率 |
| batch_size_adv | 对抗训练使用的batchsize |
| n_epoch_adv | 对抗训练使用的epoch数目 |
| train_step_adv | （对抗训练）使用所有剪裁好的图片中的1/train_step_init 作为训练集，train_step_init=1表示使用全部图片 |
| lr_adv | 对抗训练使用的学习率 |
- To cut images in the dataset to the right size for training you should run the following code at least once.
config.py
- tensorflow
- tensorlayer
- numpy
- Start training
train.py
- Start Evaluating
evaluate.py
If you find this project useful, we would be grateful if you cite the TensorLayer paper：
@article{tensorlayer2017,
author = {Dong, Hao and Supratak, Akara and Mai, Luo and Liu, Fangde and Oehmichen, Axel and Yu, Simiao and Guo, Yike},
journal = {ACM Multimedia},
title = {{TensorLayer: A Versatile Library for Efficient Deep Learning Development}},
url = {http://tensorlayer.org},
year = {2017}
}