Abstract: Methods for augmenting a training image base representing a print on a background, for training parameters of a convolutional neural network, CNN, or for classification of an input image The present invention relates to a method for augmenting a training image base representing a print on a background, characterized in that it comprises the implementation, by data processing means (11) of a server (1), of steps of: (b) For at least a first image of said base, and a ridge map of a second print different from the print represented by said first image, generation by means of at least one generator sub-network (GB, GM, GLT) of a generative adversarial network, GAN, of a synthetic image presenting the background of said first image and representing the second print.

Abstract: The present disclosure provides an image processing method and apparatus, a storage medium and a terminal. The image processing method includes: acquiring a to-be-processed blurred image, wherein the to-be-processed blurred image is obtained by an under-screen camera through a device screen; inputting the to-be-processed blurred image to a trained generative adversarial network model to obtain a processed clear image, wherein the generative adversarial network model is trained using a preset training sample, the preset training sample includes a clear image sample and a blurred image sample corresponding to each other; and outputting the processed clear image. Embodiments of the present disclosure can improve image quality of an image captured by the under-screen camera.

Abstract: An image processing system includes: an image signal processor including a first neural network, and processing an input image by using the first neural network so as to generate a post-processed image; and a discriminator including a second neural network, and receiving a target image and the post-processed image, and discriminating the target image and the post-processed image into a real image and a fake image by using the second neural network, wherein the second neural network is trained to discriminate the target image as a real image and to discriminate the post-processed image as a fake image, and the first neural network is trained in such a manner that the post-processed image is discriminated as a real image by the second neural network.

Abstract: A computer-implemented method for implementing face recognition includes obtaining a face recognition model trained on labeled face data, separating, using a mixture of probability distributions, a plurality of unlabeled faces corresponding to unlabeled face data into a set of one or more overlapping unlabeled faces that include overlapping identities to those in the labeled face data and a set of one or more disjoint unlabeled faces that include disjoint identities to those in the labeled face data, clustering the one or more disjoint unlabeled faces using a graph convolutional network to generate one or more cluster assignments, generating a clustering uncertainty associated with the one or more cluster assignments, and retraining the face recognition model on the labeled face data and the unlabeled face data to improve face recognition performance by incorporating the clustering uncertainty.

Abstract: A method for jointly removing rolling shutter (RS) distortions and blur artifacts in a single input RS and blurred image is presented. The method includes generating a plurality of RS blurred images from a camera, synthesizing RS blurred images from a set of GS sharp images, corresponding GS sharp depth maps, and synthesized RS camera motions by employing a structure-and-motion-aware RS distortion and blur rendering module to generate training data to train a single-view joint RS correction and deblurring convolutional neural network (CNN), and predicting an RS rectified and deblurred image from the single input RS and blurred image by employing the single-view joint RS correction and deblurring CNN.

Abstract: Commercial interactions with non-discretized items such as liquids in carafes or other dispensers are detected and associated with actors using images captured by one or more digital cameras including the carafes or dispensers within their fields of view. The images are processed to detect body parts of actors and other aspects therein, and to not only determine that a commercial interaction has occurred but also identify an actor that performed the commercial interaction. Based on information or data determined from such images, movements of body parts associated with raising, lowering or rotating one or more carafes or other dispensers may be detected, and a commercial interaction involving such carafes or dispensers may be detected and associated with a specific actor accordingly.

Abstract: Methods, systems, an apparatus, including computer programs encoded on a storage device, for training an image classifier. A method includes receiving an image that includes a depiction of an object; generating a set of poorly localized bounding boxes; and generating a set of accurately localized bounding boxes. The method includes training, at a first learning rate and using the poorly localized bounding boxes, an object classifier to classify the object; and training, at a second learning rate that is lower than the first learning rate, and using the accurately localized bounding boxes, the object classifier to classify the object. The method includes receiving a second image that includes a depiction of an object; and providing, to the trained object classifier, the second image. The method includes receiving an indication that the object classifier classified the object in the second image; and performing one or more actions.

Abstract: Disclosed are a method and an apparatus for detecting abnormal objects in a video. The method for detecting abnormal objects in a video reconstructs a restored batch by applying each input batch to which an inpainting pattern is applied to a trained auto-encoder model, and fuses a time domain reconstruction error using time domain restored frames output by extracting and restoring a time domain feature point by applying a spatial domain reconstruction error and a plurality of successive frames using a restored frame output by combining the reconstructed restoring batch to a trained LSTM auto-encoder model to estimate an area where an abnormal object is positioned.

Abstract: A method for accelerating a convolution of a kernel matrix over an input matrix for computation of an output matrix using in-memory computation involves storing in different sets of cells, in an array of cells, respective combinations of elements of the kernel matrix or of multiple kernel matrices. To perform the convolution, a sequence of input vectors from an input matrix is applied to the array. Each of the input vectors is applied to the different sets of cells in parallel for computation during the same time interval. The outputs from each of the different sets of cells generated in response to each input vector are sensed to produce a set of data representing the contributions of that input vector to multiple elements of an output matrix. The sets of data generated across the input matrix are used to produce the output matrix.

Abstract: A system and method for extracting features from a 2D image of an object using a deep learning neural network and a vector field estimation process. The method includes extracting a plurality of possible feature points, generating a mask image that defines pixels in the 2D image where the object is located, and generating a vector field image for each extracted feature point that includes an arrow directed towards the extracted feature point. The method also includes generating a vector intersection image by identifying an intersection point where the arrows for every combination of two pixels in the 2D image intersect. The method assigns a score for each intersection point depending on the distance from each pixel for each combination of two pixels and the intersection point, and generates a point voting image that identifies a feature location from a number of clustered points.

Abstract: A method includes receiving an image by a deep neural network (DNN) and obtaining a first feature map based on the image while the DNN is in a trained state, wherein the DNN is configured to perform a task based on the image, and is trained with a training image by using a feature sparsification with smoothness regularization process and a back propagation and weight update process that updates the DNN based on an output of the feature sparsification with smoothness regularization process.

Abstract: There is provided with an image processing apparatus. A detection unit detects an object from a captured image. A generation unit generates a map representing a correspondence between objects detected in a plurality of captured images. A determination unit matches the objects detected in the plurality of captured images based on the generated map.

Abstract: A passenger vehicle optical communication system includes a source vehicle including a light source and an endpoint vehicle including a camera. The source vehicle transmits a series of patterns using the light source to communicate, as one example, state information to the endpoint vehicle.

Abstract: The present disclosure advantageously provides an Optical Hardware Accelerator (OHA) for an Artificial Neural Network (ANN) that includes a communication bus interface, a memory, a controller, and an optical computing engine (OCE). The OCE is configured to execute an ANN model with ANN weights. Each ANN weight includes a quantized phase shift value ?i and a phase shift value ?i. The OCE includes a digital-to-optical (D/O) converter configured to generate input optical signals based on the input data, an optical neural network (ONN) configured to generate output optical signals based on the input optical signals, and an optical-to-digital (O/D) converter configured to generate the output data based on the output optical signals. The ONN includes a plurality of optical units (OUs), and each OU includes an optical multiply and accumulate (OMAC) module.

Abstract: A machine-learning system includes a quaternion (QT) computation engine. Input data to the QT computation engine includes quaternion values, each comprising a real component and three imaginary components, represented as a set of real-valued tensors. A single quaternion value is represented as a 1-dimensional real-valued tensor having four real-valued components, wherein a first real-valued component represents the real component of the single quaternion value, and wherein a second, a third, and a fourth real-valued component each respectively represents one of the imaginary components. A quaternion-valued vector having a size N is represented as a 2-dimensional real-valued tensor comprising N 1-dimensional real-valued tensors. A quaternion-valued matrix having N×M dimensions is represented as a 3-dimensional real-valued tensor comprising M 2-dimensional real-valued tensors comprising N 1-dimensional real-valued tensors.

Abstract: Computer automated interactive activity recognition based on keypoint detection includes retrieving, by one or more processors, a temporal sequence of image frames from a video recording. The one or more processors identify first and second keypoints in each of the image frames in the temporal sequence using machine learning techniques. The first keypoints are associated with an object in the temporal sequence of image frames while the second keypoints are associated with an individual interacting with the object. The one or more processors combine the first keypoints with the second keypoints and extract spatial-temporal features from the combination that are used to train a classification model based on which interactive activities can be recognized.

Abstract: A method for training a classification model includes: performing training on the classification model using first and second sample sets, to calculate a classification loss; extracting a weight vector and a feature vector of each sample; calculating a mean weight vector and a mean feature vector of all samples in the first sample set; calculating a weight loss based on a difference of the weight vector of each sample in the second sample set from the mean weight vector, and calculating a feature loss based on a difference of a feature vector of each sample in the second sample set from the mean feature vector; calculating a total loss of the classification model based on the classification loss and at least one of the feature loss and the weight loss; and adjusting a parameter of the classification model until a predetermined condition is satisfied.

Abstract: A video encoder compresses video for real-time transmission to a video decoder of a remote teleoperator system that provides teleoperator support to the vehicle based on the real-time video. The video encoder recognizes one or more generic objects in captured video that can be removed from the video without affecting the ability of the teleoperator to control the vehicle. The video encoder removes regions of the video corresponding to the generic objects to compress the video, and generates a metadata stream encoding information about the removed objects. The video decoder generates replacement objects for the objects removed the compressed video. The video decoder inserts the rendered replacement objects into relevant regions of the compressed video to reconstruct the scene.

Abstract: The present disclosure provides a method, apparatus, medium, and electronic device for evaluating environmental noise of device. The method comprises obtaining original image data to be displayed; determining at least part of the original image data to be displayed as source data; obtaining comparison data according to the source data; obtaining a difference value according to the comparison data and the source data; and evaluating environmental noise of device according to the difference value.

Abstract: A convolutional neural network (CNN) processing method and apparatus. The apparatus may select, based on at least one of a characteristic of at least one kernel of a convolution layer or a characteristic of an input of the convolution layer, one operation mode from a first operation mode reusing a kernel, of the at least one kernel, and a second operation mode reusing the input, and perform a convolution operation based on the selected operation mode.