Abstract: Techniques are described for reducing latency in computer game network streaming using a machine learning (ML) model to determine an optimal bite rate/frame rate/resolution for encoding the video of the computer game to satisfy a just noticeable difference (JND) threshold while minimizing the amount of data being sent.

Abstract: Video stream data is selectively scaled so that sections within regions of interest (ROI) maintain high resolution while areas not within the region of interest are down-scaled to reduce bandwidth cost of transmission. A low compression encoder compresses sections of a video frame corresponding to one or more ROI without motion search or prediction mode decision to generate low-compression section data. The video frame is downscaled and a high compression encoder compresses the resulting downscaled video frame with prediction mode decision to generate high-compression frame data.

Abstract: Training a video decoder system may include masking one of at least two sets of video encoding parameters with invalid values to generate an invalid set. The at least two sets of video encoding parameters are provided to one or more neural networks to train them to predict valid values that correspond to values of the invalid set using an iterative training algorithm. Encoding parameters to encode are determined based on a prediction error of the one or more neural networks. Encoding parameters which are determined to be accurately predicted are dropped from the encoded data. A new video stream is encoded without the dropped encoding parameters.

Abstract: Training a video decoder system may include masking one of at least two sets of video encoding parameters with invalid values to generate an invalid set. The at least two sets of video encoding parameters are provided to one or more neural networks to train them to predict valid values that correspond to values of the invalid set using an iterative training algorithm. Encoding parameters to encode are determined based on a prediction error of the one or more neural networks. Encoding parameters which are determined to be accurately predicted are dropped from the encoded data. A new video stream is encoded without the dropped encoding parameters.

Abstract: Video stream data is selectively scaled so that sections within regions of interest (ROI) maintain high resolution while areas not within the region of interest are down-scaled to reduce bandwidth cost of transmission. A low compression encoder compresses sections of a video frame corresponding to one or more ROI without motion search or prediction mode decision to generate low-compression section data. The video frame is downscaled and a high compression encoder compresses the resulting downscaled video frame with prediction mode decision to generate high-compression frame data.

Abstract: Training a video decoder system may include generating at least two valid sets of video encoding parameters and masking one of the sets with invalid values to generate an invalid set. The valid sets of video encoding parameters may be provided to one or more neural networks to train them to predict valid values that correspond to values of the invalid set using an iterative training algorithm. A prediction error of the predicted valid values is determined from the results of the training of the neural networks and the valid video encoding parameters. The prediction error is inserted into encoding data and encoded parameters determined to be accurately predicted with the addition of the prediction error are dropped. A new video stream is encoded with the predication error and without the dropped encoding parameters.

Abstract: Gaze tracking data is analyzed to determine one or more regions of interest within an image of a video stream. The video stream data is selectively scaled so that sections within the regions of interest maintain high resolution while areas not within the region of interest are down-scaled to reduce bandwidth cost of transmission. A scheme for reduction of motion sickness by reducing the size of the high resolution area is also claimed.

Abstract: Training a video decoder system may include generating at least two valid sets of video encoding parameters and masking one of the sets with invalid values to generate an invalid set. The valid sets of video encoding parameters may be provided to one or more neural networks to train them to predict valid values that correspond to values of the invalid set using an iterative training algorithm. A prediction error of the predicted valid values is determined from the results of the training of the neural networks and the valid video encoding parameters. The prediction error is inserted into encoding data and encoded parameters determined to be accurately predicted with the addition of the prediction error are dropped. A new video stream is encoded with the predication error and without the dropped encoding parameters.

Abstract: Training a video decoder system comprising, generating at least two sets of video encoding parameters wherein the at least two sets of video encoding parameters are valid, masking a set of the at least two sets of encoding parameters with invalid values to generate an invalid set of video encoding parameters, providing a set of the at least two sets of video encoding parameters to a neural network, training the neural network to predict valid video encoding parameter values for the invalid set using an iterative training algorithm, determining which encoding parameters need to be encoded based on analysis of a prediction error of the trained recurrent neural network, dropping the encoding parameters from the encoded data which are determined to be accurately predicted by the trained recurrent neural network, encoding a new video stream without the dropped encoding parameters. A coder and decoder system with neural network is also disclosed.

Abstract: Training a video decoder system comprising, generating at least two sets of video encoding parameters wherein the at least two sets of video encoding parameters are valid, masking a set of the at least two sets of encoding parameters with invalid values to generate an invalid set of video encoding parameters, providing a set of the at least two sets of video encoding parameters to a neural network, training the neural network to predict valid video encoding parameter values for the invalid set using an iterative training algorithm, determining which encoding parameters need to be encoded based on analysis of a prediction error of the trained recurrent neural network, dropping the encoding parameters from the encoded data which are determined to be accurately predicted by the trained recurrent neural network, encoding a new video stream without the dropped encoding parameters. A coder and decoder system with neural network is also disclosed.

Abstract: Gaze tracking data is analyzed to determine one or more regions of interest within an image of a video stream. The video stream data is selectively scaled so that sections within the regions of interest maintain high resolution while areas not within the region of interest are down-scaled to reduce bandwidth cost of transmission. A scheme for reduction of motion sickness by reducing the size of the high resolution area is also claimed.

Abstract: Lost frame reconstruction is described. A previous good or reconstructed frame may be analyzed to determine a category for the lost frame. A percentage Pi may be associated with the determined category of the lost frame. A top Pi percent magnitude samples may be zeroed out in an excitation of the previous good or reconstructed frame to produce a reconstruction excitation. The reconstruction excitation may be applied to one or more linear prediction coefficients for the previous good or reconstructed frame to generate a reconstructed frame.

Abstract: Lost frame reconstruction is described. A previous good or reconstructed frame may be analyzed to determine a category for the lost frame. A percentage Pi may be associated with the determined category of the lost frame. A top Pi percent magnitude samples may be zeroed out in an excitation of the previous good or reconstructed frame to produce a reconstruction excitation. The reconstruction excitation may be applied to one or more linear prediction coefficients for the previous good or reconstructed frame to generate a reconstructed frame.