Image Super-Resolution with Deep Variational Autoencoders

Image super-resolution (SR) techniques are used to generate a high-resolution image from low-resolution image. Until now, deep generative models such as autoregressive and Generative Adversarial Networks (GANs) have proven be effective at modelling images. VAE-based often been criticised for their feeble performance, but with new advancements VDVAE, there is now strong evidence that VAEs the potential outperform current state-of-the-art generation. In this paper, we introduce VDVAE-SR, model aims exploit most recent VAE methodologies improve upon results of similar models. VDVAE-SR tackles using transfer learning on pretrained VDVAEs. The presented competitive other models, having comparable quality metrics.