Abstract: Video object and keypoint location detection techniques are presented. The system includes a detection system for generation locations of an object's keypoints along with probabilities associated with the locations, and a stability system for stabilizing keypoint locations of the detected objects. In some aspects, the generated probabilities are two-dimensional array correspond locations within input images, and stability system fits the generated probabilities to a two-dimensional probability distribution function.

Abstract: A video delivery system generates and stores reduced bandwidth videos from source video. The system may include a track generator that executes functionality of application(s) to be used at sink devices, in which the track generator generates tracks from execution of the application(s) on source video and generates tracks having a reduced data size as compared to the source video. The track generator may execute a first instance of application functionality on the source video, which identifies region(s) of interest from the source video. The track generator further may downsample the source video according to downsampling parameters, and execute a second instance of application functionality on the downsampled video. The track generator may determine, from a comparison of outputs from the first and second instances of the application, whether the output from the second instance of application functionality is within an error tolerance of the output from the first instance of application functionality.

Abstract: A video classification, indexing, and retrieval system is disclosed that classifies and retrieves video along multiple indexing dimensions. A search system may field queries identifying desired parameters of video, search an indexed database for videos that match the query parameters, and create clips extracted from responsive videos that are provided in response. In this manner, different queries may cause different clips to be created from a single video, each clip tailored to the parameters of the query that is received.

Abstract: Techniques are disclosed for coding video data predictively based on predictions made from spherical-domain projections of input pictures to be coded and reference pictures that are prediction candidates. Spherical projection of an input picture and the candidate reference pictures may be generated. Thereafter, a search may be conducted for a match between the spherical-domain representation of a pixel block to be coded and a spherical-domain representation of the reference picture. On a match, an offset may be determined between the spherical-domain representation of the pixel block to a matching portion of the of the reference picture in the spherical-domain representation. The spherical-domain offset may be transformed to a motion vector in a source-domain representation of the input picture, and the pixel block may be coded predictively with reference to a source-domain representation of the matching portion of the reference picture.

Abstract: Embodiments of the present disclosure provide systems and methods for perspective shifting in a video conferencing session. In one exemplary method, a video stream may be generated. A foreground element may be identified in a frame of the video stream and distinguished from a background element of the frame. Data may be received representing a viewing condition at a terminal that will display the generated video stream. The frame of the video stream may be modified based on the received data to shift of the foreground element relative to the background element. The modified video stream may be displayed at the displaying terminal.

Abstract: In communication applications, aggregate source image data at a transmitter exceeds the data that is needed to display a rendering of a viewport at a receiver. Improved streaming techniques that include estimating a location of a viewport at a future time. According to such techniques, the viewport may represent a portion of an image from a multi-directional video to be displayed at the future time, and tile(s) of the image may be identified in which the viewport is estimated to be located. In these techniques, the image data of tile(s) in which the viewport is estimated to be located may be requested at a first service tier, and the other tile in which the viewport is not estimated to be located may be requested at a second service tier, lower than the first service tier.

Abstract: In an example method, a system accesses first input data and a machine learning architecture. The machine learning architecture includes a first module having a first neural network, a second module having a second neural network, and a third module having a third neural network. The system generates a first feature set representing a first portion of the first input data using the first neural network, and a second feature set representing a second portion of the first input data using the second neural network. The system generates, using the third neural network, first output data based on the first feature set and the second feature set.

Abstract: In communication applications, aggregate source image data at a transmitter exceeds the data that is needed to display a rendering of a viewport at a receiver. Improved streaming techniques that include estimating a location of a viewport at a future time. According to such techniques, the viewport may represent a portion of an image from a multi-directional video to be displayed at the future time, and tile(s) of the image may be identified in which the viewport is estimated to be located. In these techniques, the image data of tile(s) in which the viewport is estimated to be located may be requested at a first service tier, and the other tile in which the viewport is not estimated to be located may be requested at a second service tier, lower than the first service tier.

Abstract: Improved neural-network-based image and video coding techniques are presented, including hybrid techniques that include both tools of a host codec and neural-network-based tools. In these improved techniques, the host coding tools may include conventional video coding standards such H.266 (VVC). In an aspects, source frames may be partitioned and either host or neural-network-based tools may be selected per partition. Coding parameter decisions for a partition may be constrained based on the partitioning and coding tool selection. Rate control for host and neural network tools may be combined. Multi-stage processing of neural network output may use a checkerboard prediction pattern.

Abstract: Embodiments of the present disclosure provide systems and methods for perspective shifting in a video conferencing session. In one exemplary method, a video stream may be generated. A foreground element may be identified in a frame of the video stream and distinguished from a background element of the frame. Data may be received representing a viewing condition at a terminal that will display the generated video stream. The frame of the video stream may be modified based on the received data to shift of the foreground element relative to the background element. The modified video stream may be displayed at the displaying terminal.

Abstract: Embodiments of the present disclosure provide systems and methods for perspective shifting in a video conferencing session. In one exemplary method, a video stream may be generated. A foreground element may be identified in a frame of the video stream and distinguished from a background element of the frame. Data may be received representing a viewing condition at a terminal that will display the generated video stream. The frame of the video stream may be modified based on the received data to shift of the foreground element relative to the background element. The modified video stream may be displayed at the displaying terminal.

Abstract: A method of operating a high dynamic range display device comprises the steps of: accessing an image signal; generating an intermediate backlighting driver signal for individual backlight elements for a backlighting unit responsive to the image signal; convoluting the intermediate backlighting driver signals with a point spread function of the backlighting unit; deriving at least one new backlighting driver signal responsive to the convoluting step; determining display error associated with a plurality of available light shutter signals of a front-end unit having individual light shutters and associated with the at least one new backlighting driver signal, the front-end unit having a higher resolution than the backlighting unit; driving the display device with a combination of shutter signals and new backlighting driver signals that causes a reduction in the display error with respect to other generated intermediate backlighting driver signals and other available light shutter signals.

Abstract: Systems and methods disclosed for video compression, utilizing neural networks for predictive video coding. Processes employed combine multiple banks of neural networks with codec system components to carry out the coding and decoding of video data.

Abstract: Techniques are presented for modifying images of an object in video, for example to correct for lens distortion, or to beautify a face. These techniques include extracting and validating features of an object from a source video frame, tracking those features over time, estimating a pose of the object, modifying a 3D model of the object based on the features, and rendering a modified video frame based on the modified 3D model and modified intrinsic and extrinsic matrices. These techniques may be applied in real-time to an object in a sequence of video frames.

Abstract: Methods and Systems disclosed to counteract spatial distortions introduced by imaging processes of multi-directional video frames, where objects may be projected to spherical or equirectangular representations. Techniques provided to invert the spatial distortions in video frames used as reference picture data in predictive coding, by spatially transforming the image content of the reference picture data before this image content is being used for the prediction of input video data in prediction-based coders and decoders.

Abstract: Systems and methods disclosed for video compression, utilizing neural networks for predictive video coding. Processes employed combine multiple banks of neural networks with codec system components to carry out the coding and decoding of video data.

Abstract: Video object and keypoint location detection techniques are presented. The system includes a detection system for generation locations of an object's keypoints along with probabilities associated with the locations, and a stability system for stabilizing keypoint locations of the detected objects. In some aspects, the generated probabilities are two-dimensional array correspond locations within input images, and stability system fits the generated probabilities to a two-dimensional probability distribution function.

Abstract: Techniques are disclosed for coding video data predictively based on predictions made from spherical-domain projections of input pictures to be coded and reference pictures that are prediction candidates. Spherical projection of an input picture and the candidate reference pictures may be generated. Thereafter, a search may be conducted for a match between the spherical-domain representation of a pixel block to be coded and a spherical-domain representation of the reference picture. On a match, an offset may be determined between the spherical-domain representation of the pixel block to a matching portion of the of the reference picture in the spherical-domain representation. The spherical-domain offset may be transformed to a motion vector in a source-domain representation of the input picture, and the pixel block may be coded predictively with reference to a source-domain representation of the matching portion of the reference picture.

Abstract: Chroma deblock filtering of reconstructed video samples may be performed to remove blockiness artifacts and reduce color artifacts without over-smoothing. In a first method, chroma deblocking may be performed for boundary samples of a smallest transform size, regardless of partitions and coding modes. In a second method, chroma deblocking may be performed when a boundary strength is greater than 0. In a third method, chroma deblocking may be performed regardless of boundary strengths. In a fourth method, the type of chroma deblocking to be performed may be signaled in a slice header by a flag. Furthermore, luma deblock filtering techniques may be applied to chroma deblock filtering.

Abstract: Techniques are disclosed for managing display of content from multi-view video data. According to these techniques, an object may be identified from content of the multi-view video. The object's location may be tracked across a sequence of multi-view video. The technique may extract a sub-set of video that is contained within a view window that is shifted in an image space of the multi-view video in correspondence to the tracked object's location. These techniques may be implemented either in an image source device or an image sink device.