US 12,136,262 B2
Segmenting objects by refining shape priors
Weicheng Kuo, Berkeley, CA (US); Anelia Angelova, Sunnyvale, CA (US); and Tsung-Yi Lin, Sunnyvale, CA (US)
Assigned to Google LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Oct. 12, 2023, as Appl. No. 18/379,532.
Application 18/379,532 is a continuation of application No. 17/290,814, granted, now 11,823,443, previously published as PCT/US2019/046447, filed on Aug. 14, 2019.
Claims priority of provisional application 62/768,481, filed on Nov. 16, 2018.
Prior Publication US 2024/0037926 A1, Feb. 1, 2024
Int. Cl. G06V 10/82 (2022.01); G06T 7/10 (2017.01); G06V 10/25 (2022.01); G06V 10/26 (2022.01); G06V 10/44 (2022.01); G06V 10/764 (2022.01); G06V 10/77 (2022.01); G06V 10/774 (2022.01); G06V 20/10 (2022.01)
CPC G06V 10/82 (2022.01) [G06T 7/10 (2017.01); G06V 10/25 (2022.01); G06V 10/26 (2022.01); G06V 10/454 (2022.01); G06V 10/764 (2022.01); G06V 10/7715 (2022.01); G06V 10/774 (2022.01); G06V 20/10 (2022.01); G06T 2207/20081 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method performed by one or more data processing apparatus, the method comprising:
processing an image to generate data identifying a region of the image that depicts a particular object;
obtaining data defining a plurality of example object segmentations from a set of training object segmentations, wherein each example object segmentation defines an example segmentation of a respective object;
generating a shape distribution that defines a respective weight value for each of the example object segmentations using a prior neural network, wherein the weight value for an example object segmentation characterizes an estimated similarity between: (i) a segmentation of the particular object depicted in the region of the image, and (ii) the example object segmentation;
generating a shape descriptor that characterizes an estimated shape of the particular object depicted in the region of the image by combining the example object segmentations in accordance with the weight values for the example object segmentations; and
processing the shape descriptor that characterizes the estimated shape of the particular object depicted in the region of the image, by one or more neural network layers, to generate an output segmentation of the particular object depicted in the region of the image, wherein the output segmentation of the particular object defines, for each pixel in the region of the image, whether the pixel is included in the particular object depicted in the region, comprising:
processing the shape descriptor, by one or more first neural network layers, to generate a coarse segmentation of the object; and
processing the coarse segmentation of the object depicted in the region of the image, by one or more second neural network layers, to generate the output segmentation as a refinement of the coarse segmentation of the object.