Image parameters for SRGAN

In some implementations of SRGAN I've noticed, that datasets consist of the high-resolution images and the low-resolution images are created later by, e.g. resizing (decreasing the size) hr-images. However, e.g. I have a dataset of both hr-images and lr-images, all of them are of the same size but lr-images are of lower quality (less details, blurred). Is it proper to use SRGAN in such case or should I use another way of image enhancing?

Topic image generative-models dataset

Category Data Science

1
answered at

Actually yes !

The most important thing is that your dataset contains high resolution images. You can recreate low resolution images with a home made algorithm or you can use the low resolution images (or crops of these) provided in the dataset. In this video : Image Super Resolution: SRResNet and SRGAN, someone is creating a super resolution gan from the div2k dataset which have both high resolution images and low resolution images.

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.