公开(公告)号：US11557123B2

公开(公告)日：2023-01-17

申请号：US17344629

申请日：2021-06-10

Applicant: SHANDONG UNIVERSITY

Inventor： Yang Yang ,  Chenguan Li ,  Peng Xu ,  Yunxia Liu ,  Man Guo ,  Yujun Li

IPC: G06V40/10 ,  G06V20/40

Abstract: A scene change method and system combining instance segmentation and cycle generative adversarial networks are provided. The method includes: processing a video of a target scene and then inputting the video into an instance segmentation network to obtain segmented scene components, that is, obtain mask cut images of the target scene; and processing targets in the mask cut images of the target scene by using cycle generative adversarial networks according to the requirements of temporal attributes to generate data in a style-migrated state, and generating style-migrated targets with unfixed spatial attributes into a style-migrated static scene according to a specific spatial trajectory to achieve a scene change effect.

公开(公告)号：US11763540B2

公开(公告)日：2023-09-19

申请号：US17341855

申请日：2021-06-08

Applicant: SHANDONG UNIVERSITY

Inventor： Yang Yang ,  Junjie Wang ,  Yunxia Liu

IPC: G06V10/25 ,  G06N3/08 ,  G06F18/214 ,  G06V10/774 ,  G06V20/40

CPC classification number: G06V10/25 ,  G06F18/214 ,  G06N3/08 ,  G06V10/7747 ,  G06V20/46

Abstract: A data enhancement expansion method, recognition method and system for deep learning, the data enhancement expansion method includes the following steps: collecting original video data of a target to be recognized, and extracting original images of several recognized targets from the original video data; extracting seed images of RoI outlines of the recognized targets from the original images; performing an image enhancement operation on the seed images of the RoI outlines of the recognized targets, and randomly extracting the seed images subjected the image enhancement operation for image aliasing enhancement to obtain several composite images; and generating a data set based on the original images of the several recognized targets and the several composite images. Original data materials are easy to obtain with extremely low cost and high authenticity, and can be really put to a deep learning network to achieve good recognition results.