Lg Electronics Inc.

Jinseok Im

Sanghoon Kim

Jongchan Kim

Seungkyun Oh

Seung-Kyun Oh

Naeri Park

Hyunsoo Kim

Jeonghyu Yang

Seungjong Choi

Jungeun Lim

Sungho Bae

Cheoljoo Lee

Jaewon Ha

Ingu Hwang

Sungwook Shin

Seongmock Yoo

Disclosed is a batch normalization layer training method, which may be used in a neural network learning apparatus having limited operational processing capability and storage space. A batch normalization layer training method according to an embodiment of the present disclosure may perform batch normalization transform by setting the gradients of the standard deviation and the mean of the loss function to zero, and applying a normalized statistic value obtained from an initial neural network or a previous neural network to the gradient of the loss function. The neural network learning apparatus of the present disclosure may be connected or converged with an Artificial Intelligence module, an Unmanned Aerial Vehicle (UAV), a robot, an Augmented Reality (AR) apparatus, a Virtual Reality (VR), a 5G network service-related apparatus, etc.

Batch normalization layer training method

An automatic labeling apparatus for object recognition and a method therefor are provided. The automatic labeling apparatus for object recognition is configured to apply an object recognition algorithm to each of a plurality of image frames so as to recognize an object, and in response to a determination that an object recognition result in at least one first image frame among the image frames corresponds to a predetermined error condition, automatically generate a data set on an object which is a target of object recognition by using an object recognition result of a second image frame other than the first image frame among the image frames and an object image of the first image frame. The object recognition algorithm, which is a neural network model generated through machine learning, may be stored in a memory or provided through a server in an artificial intelligence environment through a 5G network.

Automatic labeling apparatus and method for object recognition

An artificial intelligence moving agent is provided. The artificial intelligence moving agent includes: a camera configured to photograph an image, and a processor configured to photograph an object, acquire type information of the object by providing an image of the photographed object to an artificial intelligence model, acquire correction type information designated by a user with respect to the image of the photographed object, and train the artificial intelligence model by using the correction type information.

Artificial intelligence moving agent

In a method for processing a video signal, according to the present invention, a first video decoder can extract motion information for inter prediction on a current half frame from a bitstream, a second video decoder can perform the inter prediction on the current half frame by using the extracted motion information, and the first video decoder can restore the inter-predicted current half frame. Accordingly, the coding time of a video sequence can be reduced.

Method and device for processing video signal by performing inter-prediction in parallel

Provided are a device and method for detecting watermarks in content carrying watermarks. One method for watermark detection according to the present invention comprises the steps of: extracting the label of a frequency component of the content; generating a bit sequence by making bit values correspond on the basis of changes in the label being extracted; checking correlation while shifting the phase of the band-spreading code (pn sequence) with respect to the generated bit sequence; and checking the amount of phase shift in the band-spreading code at times when the correlation checked in this way falls into the category of autocorrelation, and then determining a bit group having a value corresponding to the amount of phase shift checked in this manner. The bit group determined in this way constitutes part of watermark data.

Device and method for embedding watermarks in contents and for detecting embedded watermarks

A method and device for converting 3D images are disclosed herein. The method include the steps of analyzing a 2-dimensional (2D) image and generating 2D image information, generating a global depth and a local depth by using the generated 2D image information, the global depth corresponding to 3D effect pattern information of the 2D image and the local depth corresponding to 3D effect information of an object included in the 2D image, and generating a depth map by using the global depth and the local depth, wherein the depth map corresponds to 3D effect information of the 2D image, and generating a 3D image configured of a left-eye image and a right-eye image by using the 2D image and the depth map.

Growing community of inventors

Seoul, South Korea

Jinseok Im