Abstract: An improved machine learning architecture is proposed that is adapted to generate mouth regions corresponding to a target audio track that can be used, for example, in lip dubbing a base video in a first language to match a second language in the target audio track. The proposed machine learning architecture specifically includes modifications to resolve an internal mouth ambiguity problem. A number of variants are proposed along with corresponding methods and computer program products/computer readable media.

Abstract: An image processing system comprising: a computer readable medium and at least one processor configured to provide a machine learning architecture for image processing. In particular, one or more modified keyframes are used for training the machine learning architecture. The modifications are then automatically propagated to remaining frames requiring modification through interpolation or extrapolation through processing remaining frames through the trained machine learning architecture. The generated modified frames or frame portions can then be inserted into an original video to generate a modified video where the modifications have been propagated. Example usages include automatic computational approaches for aging/de-aging and addition/removal of tattoos or other visual effects.

Abstract: A computer implemented system is described in various embodiments herein, the system includes a processor, a memory coupled to the processor and storing processor-executable instructions that, when executed, configure the processor to: receive an initial video data object V comprising a plurality of image frame data objects I; receive a set of time-synchronized viseme parameters corresponding to target audio data object A?; and process the initial video data object V and the set of time-synchronized viseme parameters using a machine learning network to generate an output video data object V?, wherein initial mouth regions in the initial video data object V have been replaced with replacement mouth regions generated based on the set of time-synchronized viseme parameters.

Abstract: An image processing system comprising: a computer readable medium and at least one processor configured to provide a machine learning architecture for image processing. In particular, modified keyframes are used for training the machine learning architecture. The modifications are then automatically propagated to remaining frames requiring modification through interpolation or extrapolation through processing remaining frames through the trained machine learning architecture. The generated modified frames or frame portions can then be inserted into an original video to generate a modified video where the modifications have been propagated. Example usages include automatic computational approaches for aging/de-aging and addition/removal of tattoos or other visual effects.

Abstract: A computer implemented system is described in various embodiments herein, the system includes a processor, a memory coupled to the processor and storing processor-executable instructions that, when executed, configure the processor to: receive an initial video data object V comprising a plurality of image frame data objects I; receive a set of time-synchronized viseme parameters corresponding to target audio data object A?; and process the initial video data object V and the set of time-synchronized viseme parameters using a machine learning network to generate an output video data object V?, wherein initial mouth regions in the initial video data object V have been replaced with replacement mouth regions generated based on the set of time-synchronized viseme parameters.

Abstract: An image processing system comprising: a computer readable medium and at least one processor configured to provide a machine learning architecture for image processing. In particular, modified keyframes are used for training the machine learning architecture. The modifications are then automatically propagated to remaining frames requiring modification through interpolation or extrapolation through processing remaining frames through the trained machine learning architecture. The generated modified frames or frame portions can then be inserted into an original video to generate a modified video where the modifications have been propagated. Example usages include automatic computational approaches for aging/de-aging and addition/removal of tattoos or other visual effects.

Abstract: An image processing system comprising: a computer readable medium and at least one processor configured to provide a machine learning architecture for image processing. In particular, keyframes are selected for modification by a visual artist, and the modifications are used for training the machine learning architecture. The modifications are then automatically propagated to remaining frames requiring modification through interpolation or extrapolation through processing remaining frames through the trained machine learning architecture. The generated modified frames or frame portions can then be inserted into an original video to generate a modified video where the modifications have been propagated. Example usages include automatic computational approaches for aging/de-aging and addition/removal of tattoos or other visual effects.

Abstract: In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes. First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.

Abstract: An image processing system comprising: a computer readable medium and at least one processor configured to provide a machine learning architecture for image processing. In particular, keyframes are selected for modification by a visual artist, and the modifications are used for training the machine learning architecture. The modifications are then automatically propagated to remaining frames requiring modification through interpolation or extrapolation through processing remaining frames through the trained machine learning architecture. The generated modified frames or frame portions can then be inserted into an original video to generate a modified video where the modifications have been propagated. Example usages include automatic computational approaches for aging/de-aging and addition/removal of tattoos or other visual effects.

Abstract: In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes. First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.

Abstract: In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes, First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.

Abstract: Presented herein are techniques for a low-complexity process of generating an artificial frame that can be used for prediction. At least a first reference frame and a second reference frame of a video signal are obtained. A synthetic reference frame is generated from the first reference frame and the second reference frame. Reference blocks from each of the first reference frame and the second reference frame are combined to derive an interpolated block of the synthetic reference frame.

Abstract: A system, method, and computer-readable media for performing a fine-grained encrypted search of data stored in encrypted form in an encrypted search database. The system includes at least one processor and at least one memory having computer-readable instructions for performing the method. The method includes performing an encrypted search of the encrypted search database based on one or more search terms, presenting a result of the encrypted search on an interface, and performing a further search of an encrypted data storage based on a response to the result.

Abstract: Presented herein are techniques for a low-complexity process of generating an artificial frame that can be used for prediction. At least a first reference frame and a second reference frame of a video signal are obtained. A synthetic reference frame is generated from the first reference frame and the second reference frame. Reference blocks from each of the first reference frame and the second reference frame are combined to derive an interpolated block of the synthetic reference frame.

Abstract: A low-complexity process of generating an artificial frame that can be used for prediction. At least a first reference frame and a second reference frame of a video signal are obtained. A synthetic reference frame is generated from the first reference frame and the second reference frame. Reference blocks from each of the first reference frame and the second reference frame are combined to derive an interpolated block of the synthetic reference frame.

Abstract: A first endpoint device has access to common video data including common video frames and encoded common video data having the common video frames encoded therein. The encoded common video data is downloaded to a second endpoint device. After, or during, the downloading of the encoded common video data, live video frames are played in a play order. The live video frames are encoded in the play order into encoded live video frames. To encode the live video frames, each live video frame is predicted based on a previous live video frame that has been encoded and a common video frame from the common video data that has been downloaded in the encoded common video data. The encoded live video frames include indications of the previous live video frame and the common video frame used to encode each encoded live video frame are transmitted to the second endpoint device.

Abstract: In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes, First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.

Abstract: In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes. First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.

Abstract: In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes. First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.

Abstract: In video coding, where differences between input picture values and picture prediction values are transformed in a block based transform, the differences are formed in a series of parallel steps. A first step conducted in parallel upon a first subset of pixels uses prediction values based wholly on previously processed blocks. This first subset can include anchor pixels which are not contiguous with any previously processed block. A second step conducted in parallel upon a second subset includes pixels which are predicted from pixels of the first subset.