US 12,437,364 B2
Region-of-interest (ROI) guided sampling for AI super resolution transfer learning feature adaptation
Wai Mun Wong, Singapore (SG); Chia-Da Lee, Hsinchu (TW); Cheng Lung Jen, Hsinchu (TW); Chun Chen Lin, Hsinchu (TW); Shih-Che Chen, Hsinchu (TW); and Pei-Kuei Tsung, Hsinchu (TW)
Assigned to MediaTek Singapore Pte. Ltd., Singapore (SG)
Filed by MediaTek Singapore Pte. Ltd., Singapore (SG)
Filed on Jul. 21, 2022, as Appl. No. 17/870,261.
Claims priority of provisional application 63/234,728, filed on Aug. 19, 2021.
Prior Publication US 2023/0053776 A1, Feb. 23, 2023
Int. Cl. G06T 3/4053 (2024.01); G06T 3/4046 (2024.01)
CPC G06T 3/4053 (2013.01) [G06T 3/4046 (2013.01); G06T 2200/24 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A method for collecting a training dataset for training an artificial intelligence (AI) model, comprising:
receiving a plurality of high-resolution (HR) images and information of one or more regions-of-interest (ROIs) in the HR images;
mapping a stride distribution to the ROIs, wherein the ROIs are mapped to one or more stride values that are lower than a stride value or stride values outside the ROIs;
sampling the HR images with non-uniform strides according to the ROIs and the stride distribution to generate corresponding low-resolution (LR) images; and
training the AI model to perform super-resolution (SR) operations using training pairs formed by the HR images and respective corresponding LR images.