US 12,444,056 B2
Strip decoder and inside-outside loss for high quality image segmentation
Xibeijia Guan, Singapore (SG); Tiecheng Wu, Singapore (SG); and Bo Li, Singapore (SG)
Assigned to Black Sesame Technologies Inc., San Jose, CA (US)
Filed by Black Sesame Technologies Inc., San Jose, CA (US)
Filed on Nov. 1, 2022, as Appl. No. 17/978,430.
Prior Publication US 2024/0153099 A1, May 9, 2024
Int. Cl. G06T 7/194 (2017.01); G06V 10/25 (2022.01); G06V 10/26 (2022.01)
CPC G06T 7/194 (2017.01) [G06V 10/25 (2022.01); G06V 10/267 (2022.01); G06T 2207/20132 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A system to generate an enhanced image, comprising:
a strip pooling decoder, wherein the strip pooling decoder decodes an image by covering a larger portion of the image in a number of directions and predicts a bounding box around an object to be separated from a background of the image, wherein a ground truth mask is marked around the bounding box covering the background;
an inside-outside loss unit, wherein the inside-outside loss unit calculates an inside-outside loss in the image to separate the object from the background of the image;
an output mask, wherein the output mask is marked in accordance with the inside-outside loss calculated to convert the image into a masked background and an unmasked object;
a segmentation unit, wherein the segmentation unit performs high quality segmentation of the unmasked object by cropping the unmasked object from the masked background to generate the enhanced image; and
wherein the inside-outside loss is based on an object recall ratio and a background recall ratio.