Intel Corporation

Intel Corproation

Anthony Rhodes

Manan Goel

Ke Ding

Swarnendu Kar

Techniques related to automatically segmenting video frames into per pixel fidelity object of interest and background regions are discussed. Such techniques include applying tessellation to a video frame to generate feature frames corresponding to the video frame and applying a segmentation network implementing context aware skip connections to an input volume including the feature frames and a context feature volume corresponding to the video frame to generate a segmentation for the video frame.

High fidelity interactive segmentation for video data with deep convolutional tessellations and context aware skip connections

Methods, systems, and apparatus for high-fidelity vision tasks using deep neural networks are disclosed. An example apparatus includes a feature extractor to extract low-level features and edge-enhanced features of an input image processed using a convolutional neural network, an eidetic memory block generator to generate an eidetic memory block using the extracted low-level features or the extracted edge-enhanced features, and an interactive segmentation network to perform image segmentation using the eidetic memory block, the eidetic memory block used to propagate domain-persistent features through the segmentation network.

Methods and apparatus for high-fidelity vision tasks using deep neural networks

Various embodiments are generally directed to techniques for image segmentation utilizing context, such as with a machine learning (ML) model that injects context into various training stages. Many embodiments utilize one or more of an encoder-decoder model topology and select criteria and parameters in hyper-parameter optimization (HPO) to conduct the best model neural architecture search (NAS). Some embodiments are particularly directed to resizing context frames to a resolution that corresponds with a particular stage of decoding. In several embodiments, the context frames are concatenated with one or more of data from a previous decoding stage and data from a corresponding encoding stage prior to being provided as input to a next decoding stage.

Techniques for interactive image segmentation networks

Methods, apparatus, systems, and articles of manufacture are disclosed for interactive image segmentation. An example apparatus includes an inception controller to execute an inception sublayer of a convolutional neural network (CNN) including two or more inception-atrous-collation (IAC) layers, the inception sublayer including two or more convolutions including respective kernels of varying sizes to generate multi-scale inception features, the inception sublayer to receive one or more context features indicative of user input; an atrous controller to execute an atrous sublayer of the CNN, the atrous sublayer including two or more atrous convolutions including respective kernels of varying sizes to generate multi-scale atrous features; and a collation controller to execute a collation sublayer of the CNN to collate the multi-scale inception features, the multi-scale atrous features, and eidetic memory features.

Methods, apparatus, and articles of manufacture for interactive image segmentation

Methods, systems and apparatuses may provide for technology that trains a neural network by inputting video data to the neural network, determining a boundary loss function for the neural network, and selecting weights for the neural network based at least in part on the boundary loss function, wherein the neural network outputs a pixel-level segmentation of one or more objects depicted in the video data. The technology may also operate the neural network by accepting video data and an initial feature set, conducting a tensor decomposition on the initial feature set to obtain a reduced feature set, and outputting a pixel-level segmentation of object(s) depicted in the video data based at least in part on the reduced feature set.

High resolution interactive video segmentation using latent diversity dense feature decomposition with boundary loss

Techniques related to automatically segmenting a video frame into fine grain object of interest and background regions using a ground truth segmentation of an object in a previous frame are discussed. Such techniques apply multiple levels of segmentation tracking and prediction based on color, shape, and motion of the segmentation to determine per-pixel object probabilities, and solve an energy summation model to generate a final segmentation for the video frame using the object probabilities.

Fine-grain object segmentation in video with deep features and multi-level graphical models

An apparatus, method, system and computer readable medium for video tracking. An exemplar crop is selected to be tracked in an initial frame of a video. Bayesian optimization is applied with each subsequent frame of the video by building a surrogate model of an objective function using Gaussian Process Regression (GPR) based on similarity scores of candidate crops collected from a search space in a current frame of the video. A next candidate crop in the search space is determined using an acquisition function. The next candidate crop is compared to the exemplar crop using a Siamese neural network. Comparisons of new candidate crops to the exemplar crop are made using the Siamese neural network until the exemplar crop has been found in the current frame. The new candidate crops are selected based on an updated surrogate model.

Video tracking with deep Siamese networks and Bayesian optimization

Techniques related to automatically segmenting video frames into per pixel dense object of interest and background regions are discussed. Such techniques include applying a segmentation convolutional neural network (CNN) to a CNN input including a current video frame, a previous video frame, an object of interest indicator frame, a motion frame, and multiple feature frames each including features compressed from feature layers of an object classification convolutional neural network as applied to the current video frame to generate candidate segmentations and selecting one of the candidate segmentations as a final segmentation of the current video frame.

Deep learning for dense semantic segmentation in video with automated interactivity and improved temporal coherence

Examples described herein relate to automatic identification and transformation of a color region. A user can identify a region of a video frame or image that corresponds to a color region that is to be segmented. A color region can include one or more colors that appear to be approximately a uniform color. For one or more video frames, gamma correction can be applied to frames of the video. One or more frames of a video can be mapped to two color spaces. For each pixel in an image, a determination is made if the pixel has the same color as that of the identified region based on each of the at least two color spaces identifying the pixel as the color. The color region can be identified throughout a video and transformed to another color to aid in video editing.

Technologies for automated screen segmentation

A mechanism is described for facilitating wind detection and wind noise reduction in computing environments according to one embodiment. An apparatus of embodiments, as described herein, includes wind detection logic to detect wind associated with the apparatus including a wearable computing device, wherein the wind is detected based on samples from multiple microphones and extraction and use of multiple features including spectral sub-band centroid (SSC) features and coherence features; and decision and execution logic to reduce wind noise associated with the detected wind.

Growing community of inventors

Portland, OR, United States of America

Anthony Rhodes