Current browse context:
eess.IV
Change to browse by:
References & Citations
Electrical Engineering and Systems Science > Image and Video Processing
Title: ESRGAN+ : Further Improving Enhanced Super-Resolution Generative Adversarial Network
(Submitted on 21 Jan 2020 (v1), last revised 15 Jul 2020 (this version, v2))
Abstract: Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) is a perceptual-driven approach for single image super resolution that is able to produce photorealistic images. Despite the visual quality of these generated images, there is still room for improvement. In this fashion, the model is extended to further improve the perceptual quality of the images. We have designed a novel block to replace the one used by the original ESRGAN. Moreover, we introduce noise inputs to the generator network in order to exploit stochastic variation. The resulting images present more realistic textures. The code is available at this https URL .
Submission history
From: Nathanaël Carraz Rakotonirina [view email][v1] Tue, 21 Jan 2020 14:44:15 GMT (5674kb,D)
[v2] Wed, 15 Jul 2020 10:15:37 GMT (5675kb,D)
Link back to: arXiv, form interface, contact.