US 11,790,489 B2
Systems and method of training networks for real-world super resolution with unknown degradations
Haoyu Ren, San Diego, CA (US); Amin Kheradmand, Mountain View, CA (US); Mostafa El-Khamy, San Diego, CA (US); Shuangquan Wang, San Diego, CA (US); Dongwoon Bai, San Diego, CA (US); and Jungwon Lee, San Diego, CA (US)
Assigned to Samsung Electronics Co., Ltd.
Filed by Samsung Electronics Co., Ltd., Gyeonggi-do (KR)
Filed on Dec. 24, 2020, as Appl. No. 17/133,785.
Claims priority of provisional application 63/006,390, filed on Apr. 7, 2020.
Prior Publication US 2021/0312591 A1, Oct. 7, 2021
Int. Cl. G06T 3/40 (2006.01); G06N 3/088 (2023.01); G06N 3/045 (2023.01)
CPC G06T 3/4046 (2013.01) [G06N 3/045 (2023.01); G06N 3/088 (2013.01); G06T 3/4053 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A method, comprising:
generating a dataset for real-world super resolution (SR);
training a first generative adversarial network (GAN) to perform real-world super resolution (SR) using a standard discriminator based on the generated dataset;
training a second GAN to perform real-world SR using a relativistic discriminator based on the generated dataset; and
fusing an output of the first GAN and an output of the second GAN according to an illumination threshold.