US 12,243,334 B2
Instance segmentation using sensor data having different dimensionalities
Kevin Sheu, Fremont, CA (US); and Jie Mao, Santa Clara, CA (US)
Assigned to Pony AI Inc., Grand Cayman (KY)
Filed by Pony AI Inc., Grand Cayman (KY)
Filed on Feb. 15, 2022, as Appl. No. 17/672,118.
Application 17/672,118 is a continuation of application No. 16/939,546, filed on Jul. 27, 2020, granted, now 11,250,240.
Prior Publication US 2022/0172495 A1, Jun. 2, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06V 20/64 (2022.01); G01S 17/894 (2020.01); G01S 17/931 (2020.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01); G06F 18/25 (2023.01); G06V 10/50 (2022.01)
CPC G06V 20/647 (2022.01) [G01S 17/894 (2020.01); G01S 17/931 (2020.01); G06F 18/2148 (2023.01); G06F 18/217 (2023.01); G06F 18/251 (2023.01); G06V 10/50 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for training an instance segmentation model using sensor data having different dimensionalities, the method comprising:
capturing first sensor data having a first dimensionality and second sensor data having a second dimensionality;
assigning a first set of labels to the first sensor data to obtain labeled first sensor data and a second set of labels to the second sensor data to obtain labeled second sensor data;
projecting the first sensor data onto the second sensor data to obtain training data;
generating, based at least in part on the training data and the first set of labels, a set of sparse instance segmentation masks having the second dimensionality;
performing loss propagation using the set of sparse instance segmentation masks;
providing the training data as input to the instance segmentation model; and
training the instance segmentation model using the training data.