US 11,941,081 B2
Systems and methods for training a style model
Kareem Metwaly, State College, PA (US); Rui Guo, San Jose, CA (US); Xuewei Qi, Dublin, CA (US); and Kentaro Oguchi, Mountain View, CA (US)
Assigned to Toyota Motor Engineering & Manufacturing North America, Inc., Plano, TX (US)
Filed by Toyota Motor Engineering & Manufacturing North America, Inc., Plano, TX (US)
Filed on Jun. 18, 2021, as Appl. No. 17/351,755.
Prior Publication US 2022/0405530 A1, Dec. 22, 2022
Int. Cl. G06F 18/214 (2023.01); G06F 18/21 (2023.01); G06N 20/00 (2019.01); G06V 20/56 (2022.01)
CPC G06F 18/214 (2023.01) [G06F 18/217 (2023.01); G06N 20/00 (2019.01); G06V 20/56 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A training system for improving object perception comprising:
a processor; and
a memory storing instructions that, when executed by the processor, cause the processor to:
encode, by a style model that processes data, an input image to identify first content information and identify style information, wherein the first content information is separated from the style information that identifies style features about objects within the input image;
decode, by the style model separate from the style features, the first content information into an albedo component and a shading component;
generate, by the style model, a synthetic image using the albedo component and the shading component; and
train the style model according to computed losses between the input image and the synthetic image, the computed losses including a synthetic loss from synthetic style generated with the synthetic image.