US 12,468,940 B2
System and method for training deep learning classification networks
Xianzhi Du, College Park, MD (US); Mostafa El-Khamy, Durham, NC (US); and Jungwon Lee, San Diego, CA (US)
Assigned to Samsung Electronics Co., Ltd., Seoul (KR)
Filed by Samsung Electronics Co., Ltd., Gyeonggi-do (KR)
Filed on Nov. 22, 2021, as Appl. No. 17/532,323.
Application 17/532,323 is a continuation of application No. 15/588,223, filed on May 5, 2017, granted, now 11,205,120.
Claims priority of provisional application 62/438,795, filed on Dec. 23, 2016.
Claims priority of provisional application 62/438,177, filed on Dec. 22, 2016.
Prior Publication US 2022/0083861 A1, Mar. 17, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 3/08 (2023.01); G06F 18/214 (2023.01); G06N 3/045 (2023.01); G06N 3/084 (2023.01); G06N 7/01 (2023.01); G06N 20/00 (2019.01); G06V 10/70 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06N 3/08 (2013.01) [G06F 18/214 (2023.01); G06N 3/045 (2023.01); G06N 3/084 (2013.01); G06N 7/01 (2023.01); G06N 20/00 (2019.01); G06V 10/70 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)] 19 Claims
OG exemplary drawing
 
1. A method for deep learning training, comprising:
receiving a candidate unit for classification, the candidate unit including an intersection area between a ground-truth bounding box and a detection box;
determining whether a value based on the intersection area is between two threshold values;
assigning, to the candidate unit, a label that is based on the determining, wherein the assigned label is configured as a probability value that a given feature related to the ground-truth bounding box is observed in the detection box based on the intersection area, if the value based on the intersection area is between the two threshold values, and the assigned label is configured as a hard label, if the value based on the intersection area is not between the two threshold values; and
performing deep learning training using the assigned label of the candidate unit.