Abstract: Techniques are provided for generation of secure video and tamper detection of the secure video. A methodology implementing the techniques according to an embodiment includes selecting a subset of macroblocks from a video frame to be transmitted and calculating a low frequency metric on each of the selected macroblocks. The method also includes performing a hash calculation on the low frequency metrics to generate a frame signature, encrypting the frame signature (using a private key) to generate an encrypted watermark, and modifying pixels of each of the selected macroblocks to generate the secured video frame, the modifications based on bits of the encrypted watermark that are associated with the selected macroblock. The method further includes authenticating a received video frame by comparing a calculated frame signature to an authenticated frame signature, the authenticated frame signature decrypted (using a public key) from an extracted watermark of the received video frame.

Abstract: Camera tuning process is a time consuming and labor-intensive process. To address this issue, camera tuning system including a multi-modal large language model and a retrieval-augmented generation system can be implemented to intelligently and efficiently handle a camera tuning task in real-time. The multi-modal large language model can evaluate image quality and can be finetuned using high-quality labeled data and synthetically generated labeled data. The retrieval-augmented generation system can incorporate camera configuration knowledge into a vector database and can leverage a retrieved context to generate a configuration solution that addresses image quality issues identified by the multi-modal large language model. The resulting camera tuning system is a unified process that can identify image quality issues and provide configuration solutions that address both technical and aesthetic image quality concerns.

Abstract: Systems, apparatus, articles of manufacture, and methods are disclosed to estimate a pose of a head of a user of an electronic device. An example apparatus to estimate a head pose includes at least one processor circuit to be programmed by instructions to: identify a plurality of facial landmarks in a plurality of images; identify initial image data based on the plurality of facial landmarks; augment the initial image data with a transformation operation; and train a neural network based on the initial image data and the augmented image data to: infer three-dimensional model parameters; and infer a confidence metric.

Abstract: Example methods, apparatus, systems, and articles directed to real-time super-resolution image processing using neural networks are disclosed. Example apparatus disclosed herein cause a neural network to process an input frame of input video, the input video having a first resolution, the neural network trained to upscale the input frame to a second resolution, the neural network trained to reduce a presence of one or more types of image imperfections in the input frame. Disclosed example apparatus also obtain, from the neural network, an output frame at the second resolution. Disclosed example apparatus further cause the output frame to be presented as part of an output video at the second resolution.

Abstract: Systems, apparatus, articles of manufacture, and methods are disclosed to process images using segmentation. An example apparatus includes interface circuitry, machine readable instructions, and programmable circuitry to at least one of instantiate or execute the machine readable instructions to generate a scaled frame from an input video frame, segment, with a neural network, the scaled frame to generate a scaled segmentation map based on the scaled frame, the scaled segmentation map to associate pixels of the scaled frame with ones of a plurality of segments in the scaled frame, and generate an output video frame based on the input video frame and an upscaled version of the scaled segmentation map.

Abstract: Video segmentation predictions can be temporally unstable. Some techniques can be implemented to mitigate temporal instability, but the techniques can be computationally complex. Some techniques only account for changes in the output and do not account for changes in the input. To address some of these shortcomings, a lightweight technique can be implemented to compute a temporal consistency loss. The temporal consistency loss can be higher when a pixel-wise intensity change is small, and a pixel-wise prediction change is large. The temporal consistency loss can be lower otherwise. The temporal consistency loss can be used with one or more other losses as a part of a loss function for training a segmentation network to improve temporal stability in output segmentation maps.

Abstract: A high-level understanding of the scene captured by a camera allows for the use of scene-level understanding in the processing of the captured image. A downscaled image of a captured scene is generated and used as a basis for artificial intelligence analysis before the full image of the captured scene is processed. The downscaled image is generated concurrently with the capturing of the raw image at the image sensor and before full image signal processor (ISP) processing. Neural networks and other AI algorithms can be applied directly to the downscaled image to perform high-level understanding using minimal resources. The processing of the full scale captured image can be adapted to specific scenarios based on the understanding rather than undergoing all-purpose processing. The high-level understanding is provided to the full image processing pipe for enhancements in image quality, video conferencing, face detection, and other user experiences.

Abstract: Techniques are provided for generation of secure video and tamper detection of the secure video. A methodology implementing the techniques according to an embodiment includes selecting a subset of macroblocks from a video frame to be transmitted and calculating a low frequency metric on each of the selected macroblocks. The method also includes performing a hash calculation on the low frequency metrics to generate a frame signature, encrypting the frame signature (using a private key) to generate an encrypted watermark, and modifying pixels of each of the selected macroblocks to generate the secured video frame, the modifications based on bits of the encrypted watermark that are associated with the selected macroblock. The method further includes authenticating a received video frame by comparing a calculated frame signature to an authenticated frame signature, the authenticated frame signature decrypted (using a public key) from an extracted watermark of the received video frame.

Abstract: Systems and methods for real-time, efficient, monocular gaze position determination that can be performed in real-time on a consumer-grade laptop. Gaze tracking can be used for human-computer interactions, such as window selection, user attention on screen information, gaming, augmented reality, and virtual reality. Gaze position estimation from a monocular camera involves estimating the line-of-sight of a user and intersecting the line-of-sight with a two-dimensional (2D) screen. The system uses a neural network to determine gaze position within about four degrees of accuracy while maintaining very low computational complexity. The system can be used to determine gaze position across multiple screens, determining which screen a user is viewing as well as a gaze target area on the screen.

Abstract: Techniques are provided for generation of secure video and tamper detection of the secure video. A methodology implementing the techniques according to an embodiment includes selecting a subset of macroblocks from a video frame to be transmitted and calculating a low frequency metric on each of the selected macroblocks. The method also includes performing a hash calculation on the low frequency metrics to generate a frame signature; encrypting the frame signature (using a private key) to generate an encrypted watermark; and modifying pixels of each of the selected macroblocks to generate the secured video frame, the modifications based on bits of the encrypted watermark that are associated with the selected macroblock. The method further includes authenticating a received video frame by comparing a calculated frame signature to an authenticated frame signature, the authenticated frame signature decrypted (using a public key) from an extracted watermark of the received video frame.

Abstract: An example apparatus for video frame segmentation includes a receiver to receive a current video frame to be segmented. The apparatus also includes a segmenting neural network to receive a previous mask including a segmentation mask corresponding to a previous frame and generate a segmentation mask for the current frame based on the previous mask and the video frame.

Abstract: Methods, apparatus, systems, and articles of manufacture are disclosed to calibrate a stereo camera. An example apparatus includes means for determining a motion grid between a first image and a second image captured by the stereo camera; means for determining a calibration value to calibrate the stereo camera based on a prior calibration value, a relative orientation between the first image and the second image based on the motion grid, and a metric indicative of calibration improvement; and means for estimating a depth based on the calibration value.

Abstract: A method, system, and article of highly efficient neural network video image processing uses temporal correlations.

Abstract: A method, system, and article is directed to real-time super-resolution image processing using neural networks.

Abstract: An example apparatus for video frame segmentation includes a receiver to receive a current video frame to be segmented. The apparatus also includes a segmenting neural network to receive a previous mask including a segmentation mask corresponding to a previous frame and generate a segmentation mask for the current frame based on the previous mask and the video frame.

Abstract: Techniques are provided for generation of secure video and tamper detection of the secure video. A methodology implementing the techniques according to an embodiment includes selecting a subset of macroblocks from a video frame to be transmitted and calculating a low frequency metric on each of the selected macroblocks. The method also includes performing a hash calculation on the low frequency metrics to generate a frame signature; encrypting the frame signature (using a private key) to generate an encrypted watermark; and modifying pixels of each of the selected macroblocks to generate the secured video frame, the modifications based on bits of the encrypted watermark that are associated with the selected macroblock. The method further includes authenticating a received video frame by comparing a calculated frame signature to an authenticated frame signature, the authenticated frame signature decrypted (using a public key) from an extracted watermark of the received video frame.

Abstract: Techniques related to temporal noise reduction in captured video are discussed. Such techniques include performing motion estimation on a portion of a downsampled current frame performed during the downsampling of the current frame, replacing one or more of the resultant motion vectors based on confidence scores of the resultant motion vector, and blending the current frame and a temporally previous frame to generate a temporally filtered current frame. The temporally filtered current frame may be displayed to a user and/or encoded to generate a bitstream.

Abstract: Embodiments are directed towards enabling digital cameras to create a 3D view, which can be re-rendered onto any object within a scene, so that it is both in focus and a center of perspective, based on capturing a single set of multiple 2D images of the scene. From capturing a single set of 2D images for a scene, a depth map of the scene may be generated, and used to calculate principal depths, which are then used to capture an image focused at each of the principal depths. A correspondence between a 2D image of the scene and the principal depths are determined that corresponds to a specific principal depth. For different coordinates of the 2D image, different 3D views of the scene are created that are each focused at a principal dept that corresponds to the given coordinate.

Abstract: Techniques related to temporal noise reduction in captured video are discussed. Such techniques include performing motion estimation on a portion of a downsampled current frame performed during the downsampling of the current frame, replacing one or more of the resultant motion vectors based on confidence scores of the resultant motion vector, and blending the current frame and a temporally previous frame to generate a temporally filtered current frame. The temporally filtered current frame may be displayed to a user and/or encoded to generate a bitstream.

Abstract: Methods, apparatuses and systems may provide for operating a machine learning device by obtaining training image data, conducting an offline prediction analysis of the training image data with respect to one or more real-time parameters of an image capture device, and generating one or more parameter detection models based on the offline prediction analysis. Additionally, methods, apparatuses and systems may provide for operating the image capture device by obtaining a candidate image associated with the image capture device, determining that the candidate image corresponds to a particular type of scene represented in a parameter prediction model, and adjusting one or more real-time parameters of the image capture device based at least in part on one or more parameter values associated with the particular type of scene.