US 11,704,894 B2
Semantic image segmentation using gated dense pyramid blocks
Libin Wang, Beijing (CN); Anbang Yao, Beijing (CN); Jianguo Li, Beijing (CN); and Yurong Chen, Beijing (CN)
Assigned to Intel Corporation, Santa Clara, CA (US)
Filed by Intel Corporation, Santa Clara, CA (US)
Filed on Oct. 25, 2021, as Appl. No. 17/510,013.
Application 17/510,013 is a continuation of application No. 16/489,084, granted, now 11,157,764, previously published as PCT/US2017/078256, filed on Mar. 27, 2017.
Prior Publication US 2022/0044053 A1, Feb. 10, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06V 10/44 (2022.01); G06F 18/214 (2023.01); G06F 18/2413 (2023.01); G06N 3/04 (2023.01)
CPC G06V 10/454 (2022.01) [G06F 18/2148 (2023.01); G06F 18/24143 (2023.01); G06N 3/04 (2013.01)] 25 Claims
OG exemplary drawing
 
1. At least one computer readable memory or storage device comprising computer readable instructions that, when executed, cause at least one processor to at least:
process an input image with a trained, gated dense pyramid (GDP) network to generate semantic labels for respective pixels in the input image, the GDP network including a plurality of GDP blocks, a first one of the GDP blocks to include at least one of (i) a plurality of middle layers coupled by dense connections including a forward dense connection and a backward dense connection, (ii) a forward dense connection with a middle layer having a decreased scale to generate down-sampled feature maps, (iii) a global contextual feature to smooth prediction results, or (iv) a backward dense connection with a middle layer to generate up-sampled feature maps; and
generate, based on the generated semantic labels, a segmented image corresponding to the input image.