SRGAN Explained| Super-Resolution Generative Adversarial Network

SRGAN up sample the images by a factor of 4 and produce high resolution images.
An input image of size (172 x 208 pixels) will be of size (688 x 832 pixels) after being upscaled using SRGAN. 

************************************************************************************
For queries: You can comment in comment section or you can mail me at [email protected]
************************************************************************************

Research paper on generating Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network proposes a loss called perceptual loss. 
This loss evaluates the image quality based on its perceptual quality. An interesting way to do this is by comparing the high level features of the generated image and the ground truth image. We can obtain these high level features by passing both of these images through a pre-trained image classification network (such as a VGG-Net or a ResNet).

The generator architecture is basically a fully convolutional SRRESNET model which is utilized for generating high-quality super-resolution images. The addition of the discriminator model, which acts as an image classifier, is constructed to ensure that the overall architecture adjusts accordingly to the quality of the images and the resulting images are much more optimal. The SRGAN architecture generates plausible-looking natural images with high perceptual quality.

#srgan #superresolution #gans #generativeadversarialnetworks #neuralnetworks #ai #deeplearning #computervision #deeplearning #ml #machinelearning #pifordtechnologies #generativeai #generativemodels