Exploring the Effectiveness of SRGAN As a Video Upsampler

Authors

  • Lakshay Vashisth, Aayush Bajaj, Abhishek Jha, Prof. Ajay Kaushik

Abstract

SRGAN, a generative adversarial network (GAN) has been one of the state-of-the-art techniques for image super-resolution. The proposed perceptual loss function, which is further composed of an adversarial loss and a content loss, is able to upscale images effectively by a factor of 4. While enough research and experiments have been carried out to evaluate the performance of SRGAN on images, its potential on videos is still obscure. In this paper, our main objective is to generate high-resolution counterparts of low-resolution videos via SRGAN and evaluate the network’s capabilities on the same. To quantify the model’s performance, we have used two image quality assessment metrics, Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM).  While PSNR has been in use for a long time and is widely considered to be a tried and tested approach, SSIM is a newer metric that is designed based on three factors i.e. luminance, contrast, and structure, in order to relate it more to the mechanics of the human visual system. Through these two metrics, we aim to gauge the effectiveness of SRGAN as a video upsampler, from the perspective of reconstructed pixel quality as well as human perception.

Downloads

Published

2022-07-30