This is an implementation of the SRGAN model proposed in the paper (Edge Enhanced GAN For Remote Sensing Image Superresolution) with TensorFlow.

• Python 3
• TensorFlow 2.1(GPU)
• OpenCV
• tqdm
• requests
• Pillow
• Matplotlib

Download the ImageNet dataset and preprocess them with:

$ cd vgg19/imagenet
$ python get_urls.py
$ python create_db.py
$ python download_images.py
$ python preprocess.py

Train with VGG-19:

$ cd vgg19
$ python train.py

Or you can download the pretrained model file: vgg19_model.tar.gz

Train with EEGAN:

Download the satellite image from Kaggle Open Source Draper_Satellite_Dataset and preprocess them to obetain the training samples with

$ cd src/lfw
$ python preprocess.py

Then train the model EEGAN with:

$ cd src
$ python train.py

The evaluation results will be stored in "src/result".

Test with EEGAN:

The pretrained weight can be downloaded in: Pretrained_Weight.

Put the test images into "src/test/test30" or your own folder, then

$ cd src/test
$ python test.py

The tes results will be stored in "src/test/result30".

This implementation adopts the least squares loss function instead of the sigmoid cross entropy loss function for the discriminator. See the details: Least Squares Generative Adversarial Networks

The paper says VGG54 is the perceptually most convincing results. But this implemetation uses all the feature maps generated by every layer (i.e. phi12, phi22, phi34, phi44, phi54) within the VGG19 network.

@ARTICLE{8677274, author={K. {Jiang} and Z. {Wang} and P. {Yi} and G. {Wang} and T. {Lu} and J. {Jiang}}, journal={IEEE Transactions on Geoscience and Remote Sensing}, title={Edge-Enhanced GAN for Remote Sensing Image Superresolution}, year={2019}, volume={}, number={}, pages={1-14}, keywords={Image edge detection;Image reconstruction;Satellites;Feature extraction;Image resolution;Gallium nitride;Generative adversarial networks;Adversarial learning;dense connection;edge enhancement;remote sensing imagery;superresolution.}, doi={10.1109/TGRS.2019.2902431}, ISSN={0196-2892}, month={},}