US 12,230,052 B1
System for mapping images to a canonical space
Igor Kviatkovsky, Haifa (IL); Nadav Israel Bhonker, Haifa (IL); Yevgeni Nogin, Tirat Carmel (IL); Roman Goldenberg, Haifa (IL); Manoj Aggarwal, Seattle, WA (US); and Gerard Guy Medioni, Seattle, WA (US)
Assigned to AMAZON TECHNOLOGIES, INC., Seattle, WA (US)
Filed by AMAZON TECHNOLOGIES, INC., Seattle, WA (US)
Filed on Dec. 12, 2019, as Appl. No. 16/712,655.
Int. Cl. G06K 9/00 (2022.01); G06F 18/2131 (2023.01); G06F 18/214 (2023.01); G06K 9/62 (2022.01); G06T 3/00 (2006.01); G06T 3/04 (2024.01); G06T 7/70 (2017.01); G06V 40/12 (2022.01)
CPC G06V 40/1347 (2022.01) [G06F 18/2131 (2023.01); G06F 18/214 (2023.01); G06T 3/04 (2024.01); G06T 7/70 (2017.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30196 (2013.01)] 21 Claims
OG exemplary drawing
 
1. A system comprising:
a memory storing first computer-executable instructions; and
a hardware processor to execute the first computer-executable instructions to:
determine a first model, wherein the first model comprises data representative of a three-dimensional human hand;
determine a first training texture, wherein the first training texture is of a first human hand as obtained using a first camera;
determine, based on the first model and the first training texture, first synthetic image data;
determine first synthetic map data that relates first individual pixels in the first synthetic image data with corresponding horizontal segments of a first set of horizontal segments that are associated with the first training texture;
determine second synthetic map data that relates second individual pixels in the first synthetic image data with corresponding horizontal segments of a second set of vertical segments that are associated with the first training texture;
train a first neural network using the first synthetic image data and the first synthetic map data;
train a second neural network using the first synthetic image data and the second synthetic map data;
determine a first input image of a second human hand;
determine a first feature map of the first input image;
determine superposition map data associated with the first input image, using the first neural network and the second neural network; and
determine a first canonical image based on the first input image and the superposition map data.