US 11,055,554 B2
Dynamic image region selection for visual inference
Myung Hwangbo, Palo Alto, CA (US); Hyunggi Cho, Palo Alto, CA (US); Gordon Huang, Campbell, CA (US); Filip Dziwulski, Foster City, CA (US); and Jaehyung Choi, San Mateo, CA (US)
Assigned to PHANTOM AI, INC., Burlingame, CA (US)
Filed by Phantom AI, Inc., Burlingame, CA (US)
Filed on Apr. 1, 2019, as Appl. No. 16/372,217.
Claims priority of provisional application 62/651,654, filed on Apr. 2, 2018.
Prior Publication US 2019/0303698 A1, Oct. 3, 2019
Int. Cl. G06K 9/62 (2006.01); G06K 9/20 (2006.01); G06T 3/40 (2006.01); G06K 9/00 (2006.01); G06K 9/72 (2006.01)
CPC G06K 9/2054 (2013.01) [G06K 9/00791 (2013.01); G06K 9/726 (2013.01); G06T 3/40 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for automated vision inference of an environment in an image, comprising:
receiving a full-size image of an image of an environment;
identifying a set of environmental characteristics of the image, the set of environmental characteristics describing the environment in which the full-size image was captured;
selecting a portion of the full-size image for automated vision inference based on the set of environmental characteristics of the image; and
determining semantic information about the environment in the image by applying the selected portion of the image to a vision inference model.