EnhanceNet

Tensorflow implementation of EnhanceNet for a magnification ratio of 4.

We slightly changed the procedure for training Enet as followings

• Discriminator has been changed like DCGAN.
• We only used pool5_4 and conv3_1 features from VGG-19. See losses.py
• So, we changed hyper-parameters for loss combination.

Results

Input	Enet-E	Enet-PAT

How to train?

1. Download COCO_train_DB for training ENet and unzip train2017.zip

wget http://images.cocodataset.org/zips/train2017.zip
unzip train2017.zip

2. Download VGG-19 slim model and untar

wget http://download.tensorflow.org/models/vgg_19_2016_08_28.tar.gz
tar xvzf vgg_19_2016_08_28.tar.gz

3. Do train!

# ENet-E
python3 train_SR.py --model=enhancenet --upsample=nearest \
--recon_type=residual --SR_scale=4 --run_gpu=0 \
--batch_size=32 --num_readers=4 --input_size=32 \
--losses='mse' \
--learning_rate=0.0001 \
--save_path=/your/models/will/be/saved \
--image_path=/where/is/your/COCODB/train2017/*.jpg

# ENet-PAT
python3 train_SR.py --model=enhancenet --upsample=nearest \
--recon_type=residual --SR_scale=4 --run_gpu=0 \
--batch_size=32 --num_readers=4 --input_size=32 \
--losses='perceptual,texture,adv' --adv_ver=ver2 \
--adv_gen_w=0.003 --learning_rate=0.0001 \
--save_path=/your/models/will/be/saved \
--image_path=/where/is/your/COCODB/train2017/*.jpg \
--vgg_path=/where/is/your/vgg19/vgg_19.ckpt

How to test?

python3 test_SR.py --model_path=/your/pretrained/model/folder \
--image_path=/your/image/folder \
--save_path=/generated_image/will/be/saved/here \
--run_gpu=0

Reference

@inproceedings{enhancenet,
  title={{EnhanceNet: Single Image Super-Resolution through Automated Texture Synthesis}},
  author={Sajjadi, Mehdi S. M. and Sch{\"o}lkopf, Bernhard and Hirsch, Michael},
  booktitle={Computer Vision (ICCV), 2017 IEEE International Conference on},
  pages={4501--4510},
  year={2017},
  organization={IEEE},
  url={https://arxiv.org/abs/1612.07919/}
}