Abstract: A method for training an image recognition model includes: obtaining training image sets; obtaining a first predicted probability, a second predicted probability, a third predicted probability, and a fourth predicted probability based on the training image sets by using an initial image recognition model; determining a target loss function according to the first predicted probability, the second predicted probability, the third predicted probability, and the fourth predicted probability; and training the initial image recognition model based on the target loss function, to obtain an image recognition model.

Abstract: A video encoding method includes: acquiring a coding block of a video image frame; determining a partitioning mode available to the coding block according to a size of the coding block and an allowable maximum sub-block size ratio in response to a determination that sub-block transform (SBT) is to be performed on the coding block; determining, based on the partitioning mode available to the coding block, a target partitioning mode used during the SBT performed on the coding block; and partitioning the coding block based on the target partitioning mode, and decoding a plurality of sub-blocks obtained through the partitioning.

Abstract: A data processing method for immersive media including acquiring point cloud data, encoding the point cloud data to obtain a point cloud code stream and a header information set, determining reuse indication information of the point cloud code stream based on a reuse state of the header information set, generating decoder configuration record information for the point cloud code stream according to the reuse state, and encapsulating the point cloud code stream and the decoder configuration record information as a media file of the immersive media.

Abstract: This disclosure relates to a method for vertical federated learning. In multiple participation nodes deployed in a multi-way tree topology, an upper-layer participation node corresponds to k lower-layer participation nodes. After the upper-layer participation node and the k lower-layer participation nodes exchange public keys with each other, the upper-layer participation node performs secure two-party joint computation with the lower-layer participation nodes with a first public key and second public keys as encryption parameters to obtain k two-party joint outputs of a federated model. Further, the upper-layer participation node aggregates the k two-party joint outputs to obtain a first joint model output corresponding to the federated model. As such, a multi-way tree topology deployment-based vertical federated learning architecture is provided, improving the equality of each participation node in a vertical federated learning process.

Abstract: This disclosure relates to an object account grouping method and apparatus. The method includes: displaying aggregation information corresponding to at least one object account, the aggregation information including at least two dimensions of identity association information, such as identity information of the object account, chat information of a local object account and the object account, and interaction information of the object account and the local object account in a social circle; determining, in response to a group selection operation triggered for the aggregation information on a selection control of a group tag, a target group to which a target object account in the at least one object account belongs; and adding the target object account to the target group.

Abstract: A data processing method for an electronic device includes: obtaining data rebalance information representing a new partition plan for a partition table, the partition table being a table for data distribution based on a partition key; creating, for a set specified in the partition plan, a corresponding partition space, and recording, in the corresponding partition space, a current data range of a set corresponding to the partition space, the data range comprising one of a value range of the partition key and a value list of the partition key; updating the data range recorded in the corresponding partition space based on the partition plan, and updating correspondence between the corresponding partition space and sets of the partition table by using the updated data range in the corresponding partition space; and updating data distribution in the sets of the partition table based on the correspondence.

Abstract: A network information exposure method, performed by a core network object, includes: receiving a congestion information exposure request; determining, based on the congestion information exposure request being received, a first congestion information exposure manner of a first network device based on congestion policy configuration information of a second network device; and transmitting a congestion information exposure subscription message of the first network device, wherein the congestion information exposure subscription message includes first information indicating the first congestion information exposure manner.

Abstract: A secure multi-party computation method and apparatus, a device, and a storage medium are provided, which belong to the field of data security technologies. The method includes: performing homomorphic encryption processing on first plaintext data to generate first ciphertext data; transmitting encrypted data containing the first ciphertext data to a second node device; receiving difference ciphertext data transmitted by the second node device; and decrypting the difference ciphertext data to obtain a positive and negative property of a difference between the first plaintext data and the second plaintext data. The foregoing method and apparatus, device, and storage medium are conducive to reducing the number of times of communications of secure multi-party computation, have low communication overhead and high computation efficiency, and enable magnitude comparison and equality testing to be performed simultaneously.

Abstract: In an information processing method, a first key frame is obtained. The first key frame includes initial posture data of a target virtual character. The initial posture data corresponds to an initial posture. Target posture data of the target virtual character is determined by inputting the initial posture data and a target task to a policy network trained by reinforcement learning. The target task is set for the target virtual character to perform. An output of the policy network indicates the target posture data that corresponds to a target posture of the target virtual character. At least one force to be acted on at least one first joint of the target virtual character is determined according to the initial posture data and the target posture data. The initial posture of the target virtual character is adjusted to the target posture by applying the at least one force.

Abstract: A target dangerous travel scene existing on a target road section is acquired. Reference information corresponding to the target road section is acquired. The reference information is determined according to a detection record corresponding to each of N dangerous travel scenes detected on the target road section within a target time period. N is an integer greater than 1. The detection record corresponds to each dangerous travel scene being used for reflecting time at which the corresponding dangerous travel scene is detected. An associated dangerous travel scene correlated to the target dangerous travel scene from the N dangerous travel scenes is determined according to the reference information. A prompt operation according to the target dangerous travel scene and the associated dangerous travel scene is performed.

Abstract: Embodiments of this disclosure include an information processing method. In the method, a picture of a virtual environment is displayed. The picture includes a target region located in the virtual environment where the target region is adjacent to a neighboring region in the virtual environment. In response to a first movement instruction, a virtual character is controlled to enter the target region from the neighboring region by processing circuitry of an information processing apparatus. A first movement speed of the virtual character in the neighboring region is higher than a second movement speed of the virtual character in the target region. The information processing apparatus and non-transitory computer-readable storage medium counterpart embodiments are also contemplated.

Abstract: This application discloses a face anti-spoofing recognition method and apparatus, a device, and a storage medium, and relates to the field of artificial intelligence (AI) technologies. The method is performed by a computer device, including: acquiring face video data including a to-be-detected face; detecting eye contours of the to-be-detected face from image frames of the face video data, and generating an eye contour sequence; performing eye movement probability prediction based on the eye contour sequence, and generating an eye movement state sequence; acquiring a first dataset and a second dataset in the eye movement state sequence; determining a probability of existence of an abnormal blink behavior of the to-be-detected face according to the first dataset and the second dataset; and determining that the to-be-detected face is a real face in a case that the probability is less than a predetermined threshold.

Abstract: A method and an apparatus for recognizing a subtitle region, a device, and a storage medium are provided, relating to the field of computer vision technologies of artificial intelligence. The method includes: recognizing a video to obtain n candidate subtitle regions, the candidate subtitle regions being regions in which text contents are displayed in the video, and n being a positive integer; and screening the n candidate subtitle regions according to a subtitle region screening policy to obtain the subtitle region, the subtitle region screening policy being used for determining a candidate subtitle region in which text contents have a repetition rate being lower than a repetition rate threshold and have a longest total display duration as the subtitle region. By using the method and apparatus, device, and system, labor resources required for subtitle region recognition can be saved.

Abstract: In a method of virtual prop allocation, a geographic location and state information of a terminal device in a first scene is obtained. The geographic location and the state information are transmitted to a server device, wherein the server device determines a first virtual prop in the first scene for the terminal device based on the geographic location and the state information of the terminal device. Geographic location information for releasing the first virtual prop transmitted by the server device is received. The first virtual prop is obtained according to the geographic location information for releasing the first virtual prop. Apparatus and non-transitory computer-readable storage medium counterpart embodiments are also contemplated.

Abstract: A vibration signal encoding processing method and apparatus, a computer device, and a storage medium, capable of obtaining a plurality of control elements in a target device, determining layout information of the plurality of control elements in the target device according to a device position of any control element of the plurality of control elements in the target device, determining scenario indication information corresponding to the plurality of control elements according to an expression form used by the target device to perform content output based on the plurality of control elements, determining orientation indication information of any control element based on the layout information, and performing vibration signal encoding on each of the plurality of control elements based on the scenario indication information and the orientation indication information to obtain vibration signal encoding information corresponding to each of the plurality of control elements.

Abstract: A point cloud processing method, performed by a computer device, includes: setting a limit parameter of a point cloud; obtaining attribute information of a first point of the point cloud to be coded; coding the attribute information of the first point; and transmitting the coded attribute information in a code stream, wherein the coding the attribute information is controlled according to the limit parameter, and wherein the limit parameter includes at least one of a cache limit parameter or a delay limit parameter.

Abstract: A data transmission method, performed by an access network element, includes: obtaining a first latency variation distribution characteristic of a service data packet indicating a transmission latency range of the service data packet; receiving one or more latency variation conditions of the service data packet determined by a core network element; and configuring a scheduling and transmission policy of the access network element based on the first latency variation distribution characteristic and the one or more latency variation conditions.

Abstract: A reminder method includes: displaying a group session interface of the group session, the group session being a group session in which a group member is allowed to use a group tool, the group tool being a program function that supports use by at least two group members together; displaying a group tool message on the group session interface, the group tool message including an entry link of the group tool; displaying, in response to that the group tool message is a message with a group attribute, a reminder control corresponding to the group tool in the group session interface, the reminder control being used as a reminder of an item in the group tool. The present disclosure provides a human-computer interaction solution for effectively transmitting the group tool message to the group member.

Abstract: A computing device obtains mini-program code of a to-be-classified mini-program. The computing device runs the mini-program code to obtain dynamic features of the to-be-classified mini-program during the running process. The device uses the dynamic features as inputs for a trained classifier model, to obtain classification information of the to-be-classified mini-program. The device stores the classification information of the to-be-classified mini-program.

Abstract: This application provides a method for acquiring a position of a virtual object in a virtual scene performed at an electronic device. The method includes: presenting an operation control of a target virtual prop in a virtual scene; in response to a virtual image generation operation on the operation control, controlling a first virtual object to generate a virtual image corresponding to the first virtual object in the virtual scene by using the target virtual prop; and when a second virtual object interacting with the virtual image exists in the virtual scene, presenting a position of the second virtual object in the virtual scene.