US 12,105,770 B2
Estimation device, estimation method, and computer program product
Norihiro Nakamura, Kawasaki (JP); and Akihito Seki, Yokohama (JP)
Assigned to KABUSHIKI KAISHA TOSHIBA, Tokyo (JP)
Filed by KABUSHIKI KAISHA TOSHIBA, Tokyo (JP)
Filed on Mar. 4, 2021, as Appl. No. 17/191,771.
Claims priority of application No. 2020-149570 (JP), filed on Sep. 7, 2020.
Prior Publication US 2022/0076058 A1, Mar. 10, 2022
Int. Cl. G06F 18/211 (2023.01); G06F 18/214 (2023.01); G06F 18/25 (2023.01); G06N 3/08 (2023.01); G06T 15/00 (2011.01); G01S 17/66 (2006.01)
CPC G06F 18/211 (2023.01) [G06F 18/214 (2023.01); G06F 18/251 (2023.01); G06N 3/08 (2013.01); G06T 15/00 (2013.01); G01S 17/66 (2013.01); G06T 2210/52 (2013.01)] 9 Claims
OG exemplary drawing
 
1. An estimation device comprising:
one or more processors configured to:
acquire first point cloud data;
generate, from the first point cloud data, second point cloud data obtained by combining an attention point that gains attention as an estimation target of an attribute and at least one observation point;
estimate an attribute of the attention point by using an attribute indicated by an estimation result label having a higher belonging probability out of belonging probabilities output from an attribute estimation neural network, the attribute estimation neural network including an input layer to which the second point cloud data is input and an output layer that outputs the belonging probabilities for respective attributes of the attention point;
estimate reliability of the estimation result label by using a reliability estimation neural network including an input layer to which an intermediate feature of the attribute estimation neural network is input and an output layer that outputs either logit of an attribute being a correct answer or an estimated value of a belonging probability; and
display, on a display device, first display information generated by performing rendering on an object including an attention point whose attribute is estimated by an attribute of the estimation result label whose reliability is higher than a first threshold, and generated by not performing rendering on an object including an attention point whose attribute is estimated by an attribute of the estimation result label whose reliability is the first threshold or less.