US 11,756,166 B2
Image enhancement via iterative refinement based on machine learning models
Chitwan Saharia, Toronto (CA); Jonathan Ho, Berkeley, CA (US); William Chan, Toronto (CA); Tim Salimans, Utrecht (NL); David Fleet, Toronto (CA); and Mohammad Norouzi, Toronto (CA)
Assigned to Google LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Jan. 17, 2023, as Appl. No. 18/155,420.
Application 18/155,420 is a continuation of application No. 17/391,150, filed on Aug. 2, 2021.
Prior Publication US 2023/0153959 A1, May 18, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06T 5/00 (2006.01); G06N 3/08 (2023.01); G06T 5/50 (2006.01); G06N 3/045 (2023.01); G06T 3/40 (2006.01)
CPC G06T 5/002 (2013.01) [G06N 3/045 (2023.01); G06N 3/08 (2013.01); G06T 3/4007 (2013.01); G06T 5/50 (2013.01); G06T 2207/20016 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 21 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
receiving training data from an image database;
training, based on the training data, a neural network to predict a high-resolution version of a low-resolution input image, wherein the training comprises downsampling the low-resolution input image using bicubic interpolation, and wherein the neural network is trained based on a diffusion process comprising:
an image corruption process that iteratively adds noise to a high-resolution image until a noise content is above a predefined threshold, and
an image denoising process that learns to reverse the image corruption process by starting from an initial image with an initial noise content above the predefined threshold, and iteratively removing noise from the initial image to achieve a target distribution; and outputting the trained neural network.