公开(公告)号：US20240422325A1

公开(公告)日：2024-12-19

申请号：US18747765

申请日：2024-06-19

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Xu ZHANG

IPC: H04N19/14 ,  H04N19/105 ,  H04N19/169 ,  H04N19/182

Abstract: A blocking artifact detection method is provided. The method, which is applied to a video encoder, includes: detecting, based on a sum of pixel differences between each coding unit in a current coding tree unit and its adjacent coding unit, whether the coding unit and the adjacent coding unit meet an edge detection condition; in response to determining that a coding unit and its adjacent coding unit meet the edge detection condition, determining, based on information about the coding unit, image regions of an original image corresponding to the coding unit and the adjacent coding unit; detecting whether the image regions meet the edge detection condition; and determining, in response to determining that the image regions do not meet the edge detection condition, that there is a blocking artifact between the coding unit and the adjacent coding unit. This implementation improves the accuracy of blocking artifact detection.

公开(公告)号：US20250054272A1

公开(公告)日：2025-02-13

申请号：US18749546

申请日：2024-06-20

Applicant: Beijing Baidu Netcom Science Technology Co., Ltd.

Inventor： Xu ZHANG

IPC: G06V10/44 ,  G06V20/70 ,  H04N19/176

Abstract: The present disclosure provides a method and apparatus for training a reference frame screening model, a method and apparatus for screening a reference frame, an electronic device, a storage medium, and a computer program product, relates to the field of artificial intelligence. An implementation scheme is: acquiring a training sample set, where training samples in the training sample set include a sequence of video frames and labels corresponding to video frames in the sequence of video frames, and the labels are used to represent whether the video frames corresponding to the labels are reference frames of other video frames in the sequence of video frames; and training, using a machine learning method, using the sequence of video frames as input, using the labels corresponding to the input sequence of video frames as desired output, to obtain a reference frame screening model.

公开(公告)号：US20230081957A1

公开(公告)日：2023-03-16

申请号：US17943025

申请日：2022-09-12

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Xu ZHANG ,  Wenpeng DING

IPC: H04N19/533 ,  H04N19/139 ,  H04N19/513

Abstract: A method includes: performing a first diamond search according to an initial point determined from points in a search window, a search step size being incremented by ith power of 2, i being a natural number, 0≤i≤N; and performing the following first processing: acquiring an updated initial point and an optimization range, the optimization range being less than 2N; performing a second diamond search according to the initial point, wherein, prior to a search with a search step size larger than the optimization range, if it is determined that an ending condition is met, the diamond search is ended and a corresponding second optimal point is determined; and determining a required optimal motion vector according to the second optimal point if the second optimal point meets a predetermined requirement, and otherwise, repeating the first processing.