Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media related to multi-stream video encoding for screen sharing a communications session. The system may determine an active pixel area and a remaining pixel area of a video region. of a video region. A first video stream of the active pixel area is generated at a first frame rate. A second video stream of the remaining pixel area is generated at second frame rate, where the second frame rate is a frame rate lower than the first frame rate. A client device may transmit the first video stream and the second video stream to a second client device.

Abstract: Provided are a method for automatic parallelization of a mixture of experts model, an apparatus for automatic parallelization of a mixture of experts model, a device, a medium, and a program. The method includes acquiring a computational graph of the mixture of experts model; determining a process mesh of an expert weight tensor of an expert model, where the processes are supported to execute by computing devices in the distributed system; splitting a global data tensor into sub-data tensors required by corresponding expert models, and configuring a process mesh of a sub-data tensor to be the same as a process mesh of a corresponding expert model; performing, by the expert model, processing based on an input sub-data tensor and the expert weight tensor to output a sub-result tensor; and determining a result tensor of the mixture of experts model based on at least one sub-result tensor.

Abstract: Noise can be reduced in video frames by performing one or more techniques described herein. For example, a system can divide a video frame into a set of blocks, determine a variance associated with a selected block, generate first frequency residuals associated with the selected block, and determine whether the variance is below a predefined threshold. If the variance is below the predefined threshold, the system can adjust an offset to be used during a quantization operation from a first offset value to a second offset value. The system can then apply the quantization operation to the first frequency residuals using the second offset value, to thereby generate second frequency residuals. Using the second offset value during the quantization operation can result in the second frequency residuals having a larger number of zero-valued frequency residuals, which in turn can reduce noise with respect to the selected block.

Abstract: Techniques for method for filtering artifacts from digital video are disclosed. In an example, a method includes accessing a video stream that includes a block with sub-blocks. The method includes, for each candidate filter strength parameter of a set of candidate filter strength parameters associated with the block, filtering each sub-block using a filter having the candidate filter strength parameter, creating a filtered block by combining the filtered sub-blocks, and calculating, for the filtered block, a distortion error measurement between a corresponding source block and the filtered block. The method includes selecting the filter strength parameters associated with a lowest distortion error measurement of the distortion error measurements. The method includes applying the selected filter strength parameters to an in-loop deringing filter to generate a filtered block of video.

Abstract: Techniques for selection of transform kernels for AV1 encoding are disclosed. In an example method, a computing system selects a transform kernel by iteratively evaluating each transform kernel in a candidate list until an early termination condition is satisfied. The iterative evaluation involves: generating transform coefficients for the transform kernel by applying a forward transform on a residual block of the transform kernel; generating quantized transform coefficients and dequantized transform coefficients; determining a transform-domain distortion using the quantized transform coefficients and the dequantized transform coefficients; determining a code rate based on the quantized transform coefficients; determining a rate-distortion cost using the code rate and a final distortion based on the transform-domain distortion; and selecting the transform kernel by determining that an early termination condition is satisfied. The computing system encodes the block using the selected transform kernel.

Abstract: The present invention relates to production method for nanoparticles used for light valves of different colors, light valves of different colors and application thereof. The light values comprise: a first transparent substrate, a first transparent conductive layer, a light modulating layer, a second transparent conductive layer and a second transparent substrate, the light modulating layer comprises a dispersion liquid and a nanoparticle dispersed-phase dispersed in the dispersion liquid; The tinted states of the light valve are displayed as multiple different colors in the visible spectral region, the bright state is colorless and transparent, and the multiple different colors of the light valve are realized by mixing nanoparticles of same or different color to form a mixture in the form of the nanoparticle dispersed-phase.

Abstract: The present invention relates to the technical field of path planning, and provides a deep reinforcement learning-based path exploration parameter optimization system. The system comprises: a variable parameter path planning module, configured to perform node exploration based on a deep reinforcement learning network, conduct collision detection on child nodes in a child node set, calculate cost values for all child nodes, and finally generate a loading and parking path using a Reeds-Shepp curve; an environmental state space modeling module, configured to perform regional division of obstacles around a current node and conduct environmental state space modeling; and a deep learning parameter optimization module, configured to construct a deep learning network to compute an optimal step size and an optimal steering angle, build a reward function to optimize the deep learning network, and simultaneously execute a training process of the deep learning network.

Abstract: In one example, a video encoder divides a superblock of a video frame into subblocks. The dividing includes dividing the superblock according to a first and second partitioning schemes to generate a first and second combination of subblocks, respectively. The number of non-zero residuals in each subblock in the first and second combinations of subblocks is less than a corresponding threshold set for a size of the subblock. The encoder determines the subblocks for the superblock as the first combination of subblocks based on a first cost value associated with the first combination of subblocks being lower than a second cost value associated with the second combination of subblocks. The encoder further encodes the superblock into a video bitstream representing the video by encoding the subblocks.

Abstract: A computer assisted method comprising: storing a training dataset including a plurality of geotagged candidate images and a plurality of query images, each query image having at least one corresponding candidate image having the same geolocation; applying a quasi-random or random azimuth rotation to each of the plurality of query images, and storing the azimuth rotation for each of the plurality of rotated query images; training a machine learning model, including: extracting features from the plurality of rotated query images; estimating the azimuth rotation of the rotated query image based on an inference of the extracted features of the rotated query image and extracted features from the candidate images, and using an objective function including a first loss function based on a weighted soft-margin triplet loss, and a second loss function based on an absolute angle error between the stored azimuth rotation and the estimated azimuth rotation for the stored dataset.

Abstract: Fast parameter search for chroma from luma (CFL) intra prediction in video encoding is provided. A video encoder determines a value of a parameter of a linear CFL predictor for a coding block of a frame of a video. The determining includes dividing a search range within a value range of the parameter into three portions by a starting value: the starting value and values on two sides of the starting value. The encoder determines cost values of parameter values from the three portions and select the value of the parameter that corresponds to the lowest cost value. The encoder further encodes the coding block of the frame into a bitstream representing the video using the CFL predictor with the selected value of the parameter.

Abstract: Noise can be reduced in video frames by performing one or more techniques described herein. For example, a system can divide a video frame into a set of blocks, determine a variance associated with a selected block, generate first frequency residuals associated with the selected block, and determine whether the variance is below a predefined threshold. If the variance is below the predefined threshold, the system can adjust an offset to be used during a quantization operation from a first offset value to a second offset value. The system can then apply the quantization operation to the first frequency residuals using the second offset value, to thereby generate second frequency residuals. Using the second offset value during the quantization operation can result in the second frequency residuals having a larger number of zero-valued frequency residuals, which in turn can reduce noise with respect to the selected block.

Abstract: Deriving in-loop filter parameters via training for video encoding is provided. A video encoder performs inter prediction for a frame in a set of frames of the video to generate prediction residuals for the frame. The inter prediction for the frame is performed based on a reconstructed frame in the set of frames filtered using an in-loop filter. The value of a parameter of the in-loop filter is determined by determining, for each candidate in-loop filter parameter value, a visual quality metric for a set of training frames in training video sequences filtered by the in-loop filter. The candidate in-loop filter parameter value that corresponds the highest visual quality metric can be selected as the value of the parameter of the in-loop filter. The video encoder further encodes the prediction residues of the frame and the parameter of the in-loop filter into a bitstream representing the video.

Abstract: Provided is a tensor processing method, an electronic device, and a storage medium, relating to the fields of deep learning and artificial intelligence. The method includes: determining relevant information of a conversion function corresponding to each of one or more target input tensors of a first operator in a target computation graph based on computation logic of the first operator and source split states of at least part of source input tensors of the first operator; splitting each source input tensor of the first operator based on the relevant information of the conversion function corresponding to each target input tensor to obtain each target input tensor; and sending each target input tensor to a plurality of computing devices. The plurality of computing devices are configured to perform distributed parallel communication based on each target input tensor and the first operator, to obtain an output tensor of the first operator.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media for fast search of transform kernel selection. A set of transform kernels are selected. Multiple images are encoded via the AV1 coded using the selected transform kernels. The encoded images are transmitted to another computing device.

Abstract: In one example, a video encoder divides a superblock of a video frame into subblocks. The dividing includes dividing the superblock according to a first and second partitioning schemes to generate a first and second combination of subblocks, respectively. The number of non-zero residuals in each subblock in the first and second combinations of subblocks is less than a corresponding threshold set for a size of the subblock. The encoder determines the subblocks for the superblock as the first combination of subblocks based on a first cost value associated with the first combination of subblocks being lower than a second cost value associated with the second combination of subblocks. The encoder further encodes the superblock into a video bitstream representing the video by encoding the subblocks.

Abstract: Constructing a motion vector candidate list for inter-prediction in video coding is provided. A video encoder performs inter prediction for a frame of the video to generate prediction residuals. The inter prediction is performed using motion vectors (MVs) for blocks of the frame. To determine the motion vector of a block, the encoder determines an integer-valued MV for the block based on a reference frame and selects, based on the integer-valued MV, a subset of candidate MVs from a list of MVs including MVs of neighboring blocks. The encoder identifies a MV from the subset of candidate MVs for the block and searches in a neighborhood of the identified MV for a refined MV. The encoder calculates the prediction residuals for the block based on a reference block in the reference frame pointed by the refined MV and encodes the prediction residuals into a bitstream representing the video.

Abstract: Fast parameter search for chroma from luma (CFL) intra prediction in video encoding is provided. A video encoder determines a value of a parameter of a linear CFL predictor for a coding block of a frame of a video. The determining includes dividing a search range within a value range of the parameter into three portions by a starting value: the starting value and values on two sides of the starting value. The encoder determines cost values of parameter values from the three portions and select the value of the parameter that corresponds to the lowest cost value. The encoder further encodes the coding block of the frame into a bitstream representing the video using the CFL predictor with the selected value of the parameter.

Abstract: A method and system are provided for identifying time-varying suspension characteristics of heavy-load vehicles. The method includes collecting sequential control state data of a mining truck using sensors, predicting parameter-related factors through a deep learning network, estimating suspension stiffness and damping coefficients via a linear dynamic model considering longitudinal-vertical coupling, and predicting future system states through a nonlinear dynamic model based on the estimated parameters and learned factors.

Abstract: Deriving in-loop filter parameters via training for video encoding is provided. A video encoder performs inter prediction for a frame in a set of frames of the video to generate prediction residuals for the frame. The inter prediction for the frame is performed based on a reconstructed frame in the set of frames filtered using an in-loop filter. The value of a parameter of the in-loop filter is determined by determining, for each candidate in-loop filter parameter value, a visual quality metric for a set of training frames in training video sequences filtered by the in-loop filter. The candidate in-loop filter parameter value that corresponds the highest visual quality metric can be selected as the value of the parameter of the in-loop filter. The video encoder further encodes the prediction residues of the frame and the parameter of the in-loop filter into a bitstream representing the video.

Abstract: Constructing a motion vector candidate list for inter-prediction in video coding is provided. A video encoder performs inter prediction for a frame of the video to generate prediction residuals. The inter prediction is performed using motion vectors (MVs) for blocks of the frame. To determine the motion vector of a block, the encoder determines an integer-valued MV for the block based on a reference frame and selects, based on the integer-valued MV, a subset of candidate MVs from a list of MVs including MVs of neighboring blocks. The encoder identifies a MV from the subset of candidate MVs for the block and searches in a neighborhood of the identified MV for a refined MV. The encoder calculates the prediction residuals for the block based on a reference block in the reference frame pointed by the refined MV and encodes the prediction residuals into a bitstream representing the video.