US 11,908,141 B2
Deep neural network for segmentation of road scenes and animate object instances for autonomous driving applications
Ke Chen, Sunnyvale, CA (US); Nikolai Smolyanskiy, Seattle, WA (US); Alexey Kamenev, Bellevue, WA (US); Ryan Oldja, Redmond, WA (US); Tilman Wekel, Sunnyvale, CA (US); David Nister, Bellevue, WA (US); Joachim Pehserl, Lynnwood, WA (US); Ibrahim Eden, Redmond, WA (US); Sangmin Oh, San Jose, CA (US); and Ruchi Bhargava, Redmond, WA (US)
Assigned to NVIDIA Corporation, Santa Clara, CA (US)
Filed by NVIDIA Corporation, Santa Clara, CA (US)
Filed on Jul. 24, 2020, as Appl. No. 16/938,706.
Claims priority of provisional application 62/878,659, filed on Jul. 25, 2019.
Prior Publication US 2021/0026355 A1, Jan. 28, 2021
Int. Cl. G06T 7/00 (2017.01); G06F 18/00 (2023.01); G06T 7/11 (2017.01); G05D 1/00 (2006.01); G06T 7/10 (2017.01); G06T 5/50 (2006.01); G06V 20/58 (2022.01); G06F 18/22 (2023.01); G06F 18/23 (2023.01); G06V 10/82 (2022.01); G06V 20/56 (2022.01); G06V 10/44 (2022.01)
CPC G06T 7/11 (2017.01) [G05D 1/0088 (2013.01); G06F 18/22 (2023.01); G06F 18/23 (2023.01); G06T 5/50 (2013.01); G06T 7/10 (2017.01); G06V 10/82 (2022.01); G06V 20/56 (2022.01); G06V 20/58 (2022.01); G06T 2207/10028 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30252 (2013.01); G06V 10/454 (2022.01)] 27 Claims
OG exemplary drawing
 
1. A method comprising:
applying, to a neural network, data representative of one or more images of an environment;
generating, using the neural network and based at least in part on the data:
a first output representative of, for each pixel of a plurality of pixels, one or more classifications corresponding to one or more detected objects in the environment; and
a second output representative of, for each pixel of the plurality of pixels, one or more values representative of an association between the pixel and one or more instances of the one or more detected objects;
generating, based at least in part on the second output, one or more bounding shapes corresponding to the one or more instances of the one or more detected objects;
associating the one or more classifications with the one or more bounding shapes; and
performing one or more operations by an autonomous machine based at least in part on the one or more classifications associated with the one or more bounding shapes.