Abstract: Embodiments of the present invention provide a system of interconnected neural networks capable audio-visual simulation generation by interpreting and processing a first image and, utilizing a given reference image or training set, modifying the first image such that the new image possesses parameters found within the reference image or training set. The images used and generated may include video. The system includes an autoposer, an automasker, an encoder, a generator, an improver, a discriminator, styler, and at least one training set of images or video. The system can also generate training sets for use within.

Abstract: A method can include classifying, using a compressed and specialized convolutional neural network (CNN), an object of a video frame into classes, clustering the object based on a distance of a feature vector of the object to a feature vector of a centroid object of the cluster, storing top-k classes, a centroid identification, and a cluster identification, in response to receiving a query for objects of class X from a specific video stream, retrieving image data for each centroid of each cluster that includes the class X as one of the top-k classes, classifying, using a ground truth CNN (GT-CNN), the retrieved image data for each centroid, and for each centroid determined to be classified as a member of the class X providing image data for each object in each cluster associated with the centroid.

Abstract: There is a region of interest of a synthetic image depicting an object from a class of objects. A trained neural image generator, having been trained to map embeddings from a latent space to photorealistic images of objects in the class, is accessed. A first embedding is computed from the latent space, the first embedding corresponding to an image which is similar to the region of interest while maintaining photorealistic appearance. A second embedding is computed from the latent space, the second embedding corresponding to an image which matches the synthetic image. Blending of the first embedding and the second embedding is done to form a blended embedding. At least one output image is generated from the blended embedding, the output image being more photorealistic than the synthetic image.

Abstract: In a method and a computer for determining a training function in order to generate annotated training images, a training image and training-image information are provided to a computer, together with an isolated item of image information that is independent of the training image. A first calculation is made in the computer by applying an image-information-processing first function to the isolated item of image information, and a second calculation is made by applying an image-information-processing second function to the training image. Adjustments to the first and second functions are made based on these calculation results, from which a determination of a training function is then made in the computer.

Abstract: A method is for obtaining at least one feature of interest, especially a biomarker, from an input image acquired by a medical imaging device. The at least one feature of interest is the output of a respective node of a machine learning network, in particular a deep learning network. The machine learning network processes at least part of the input image as input data. The used machine learning network is trained by machine learning using at least one constraint for the output of at least one inner node of the machine learning network during the machine learning.

Abstract: Apparatuses and methods for image processing are provided. The image processing apparatus performs area classification and object detection in an image, and includes a feature map generator configured to generate the feature map of the input image using the neural network, and an image processor configured to classify the areas and to detect the objects in the image using the generated feature map.

Abstract: In some embodiments, a system includes an artificial intelligence (AI) chip and a processor coupled to the AI chip and configured to receive an input image, crop the input image into a plurality of cropped images, and execute the AI chip to produce a plurality of feature maps based on at least a subset of the plurality of cropped images. The system may further merge at least a subset of the plurality of feature maps to form a merged feature map, and produce an output image based on the merged feature map. The cropping and merging operations may be performed according to a same pattern. The system may also include a training network configured to train weights of the CNN model in the AI chip in a gradient descent network. Cropping and merging may be performed over the training sample images in the training work in a similar manner.

Abstract: A computer-implemented method for domain analysis comprises: obtaining, by a computing device, a domain; and inputting, by the computing device, the obtained domain to a trained detection model to determine if the obtained domain was generated by one or more domain generation algorithms. The detection model comprises a neural network model, a n-gram-based machine learning model, and an ensemble layer. Inputting the obtained domain to the detection model comprises inputting the obtained domain to each of the neural network model and the n-gram-based machine learning model. The neural network model and the n-gram-based machine learning model both output to the ensemble layer. The ensemble layer outputs a probability that the obtained domain was generated by the domain generation algorithms.

Abstract: The disclosure relates to system and method for detecting an instance of lie using a Machine Learning (ML) model. In one example, the method may include extracting a set of features from an input data received from a plurality of data sources at predefined time intervals and combining the set of features from each of the plurality of data sources to obtain a multimodal data. The method may further include processing the multimodal data through an ML model to generate a label for the multimodal data. The label is generated based on a confidence score of the ML model. The label is one of a true value that corresponds to an instance of truth or a false value that corresponds to an instance of lie.

Abstract: The disclosure provides method of training a machine-learning model employing a procedurally synthesized training dataset, a machine that includes a trained machine-learning model, and a method of operating a machine. In one example, the method of training includes: (1) generating training image definitions in accordance with variations in content of training images to be included in a training dataset, (2) rendering the training images corresponding to the training image definitions, (3) generating, at least partially in parallel with the rendering, ground truth data corresponding to the training images, the training images and the ground truth comprising the training dataset, and (4) training a machine-learning model using the training dataset and the ground truth data.

Abstract: This application provides an image aesthetic processing method and an electronic device. A method for generating an image aesthetic scoring model includes: constructing a first neural network based on a preset convolutional structure set; obtaining an image classification neural network, where the image classification neural network is used to classify image scenarios; obtaining a second neural network based on the first neural network and the image classification neural network, where the second neural network is a neural network containing scenario information; and determining an image aesthetic scoring model based on the second neural network, where output information of the image aesthetic scoring model includes image scenario classification information. In this method, scenario information is integrated into a backbone neural network, so that a resulting image aesthetic scoring model is interpretable.

Abstract: Remote distribution of multiple neural network models to various client devices over a network can be implemented by identifying a native neural network and remotely converting the native neural network to a target neural network based on a given client device operating environment. The native neural network can be configured for execution using efficient parameters, and the target neural network can use less efficient but more precise parameters.

Abstract: An image classification system is provided for determining a likely classification of an image using multiple machine learning models that share a base machine learning model. The image classification system may be a browser-based system on a user computing device that obtains multiple machine learning models over a network from a remote system once, stores the models locally in the image classification system, and uses the models multiple times without needing to subsequently request the machine learning models again from the remote system. The image classification system may therefore determine likely a classification associated with an image by running the machine learning models on a user computing device.

Abstract: A data generation device includes one or more processors. The processors input input data into a neural network and obtain an inference result of the neural network The processors calculate a first loss and a second loss. The first loss becomes smaller in value as a degree of matching between the inference result and a target label becomes larger. The target label indicates a correct answer of the inference. The second loss is a loss based on a contribution degree to the inference result of a plurality of elements included in the input data and the target label. The processors update the input data based on the first loss and the second loss.

Abstract: A computer, including a processor and a memory, the memory including instructions to be executed by the processor to determine six degree of freedom (DoF) data for a first object in a first video image and generate a synthetic video image corresponding to the first video image including a synthetic object and a synthetic object label based on the six DoF data. The instructions can include further instructions to train a generative adversarial network (GAN) based on a paired first video image and a synthetic video image to generate a modified synthetic image and train a deep neural network to locate the synthetic object in the modified synthetic video image based on the synthetic object. The instructions can include further instructions to download the trained deep neural network to a computing device in a vehicle.

Abstract: In an approach for detecting key messages for a video, a processor builds a role model based on data from one or more data sources, with an identification feature of each role in a video. A processor samples a plurality of frames of the video. A processor identifies a key object presented in the plurality of frames. The key object is a role in the video. A processor recognizes a movement scenario associated with the role. A processor dynamically updates the role model based on the movement scenario. A processor identifies a role name based on the movement scenario. A processor generates a description script associated with the movement scenario for the role. A processor outputs the description script.

Abstract: An image processing apparatus applies an image to a first learning network model to optimize the edges of the image, applies the image to a second learning network model to optimize the texture of the image, and applies a first weight to the first image and a second weight to the second image based on information on the edge areas and the texture areas of the image to acquire an output image.

Abstract: The invention includes systems and methods, including computer programs encoded on computer storage media, for classifying inputs as belonging to a known or unknown class as well as for updating the system to improve is performance. In one system, there is a desired feature representation for unknown inputs, e.g., a zero vector, and the system includes transforming input data to produce a feature representation, using that to compute dissimilarity with the desired feature representation for unknown inputs and combining dissimilarity with other transformations of the feature representation to determine if the input is from a specific known class or if it is unknown. In one embodiment, the system transforms the magnitude of the feature representation into a confidence score.

Abstract: A method is disclosed. The method models a plurality of visual cortex neurons, models one or more connections between at least two visual cortex neurons in the plurality of visual cortex neurons, assigns synaptic weight value to at least one of the one or more connections, simulates application of one or more electrical signals to at least one visual cortex neuron in the plurality of visual cortex neurons, adjusts the synaptic weight value assigned to at least one of the one or more connection based on the one or more electrical signals, and generates an orientation map of the plurality of visual cortex neurons based on the adjusted synaptic weight values.

Abstract: The present subject matter refers a method for training image-alignment procedures in a computing environment. The method comprises communicating one or more images of an object to a user and receiving a plurality of user-selected zones within said one or more through a user-interface. An augmented data-set is generated based on said one or more images comprising the user-selected zones, wherein such augmented data set comprises a plurality of additional images defining variants of said one or more communicated images. Thereafter, a machine-learning based image alignment is trained based on at-least one of the augmented data set and the communicated images.

Abstract: Disclosed herein are system, method, and computer program product embodiments for querying document terms and identifying target data from documents. In an embodiment, a document processing system may receive a document and a query string. The document processing system may perform optical character recognition to obtain character information and positioning information for the characters of the document. The document processing system may generate a two-dimensional character grid for the document. The document processing system may apply a convolutional neural network to the character grid and the query string to identify target data from the document corresponding to the query string. The convolutional neural network may then produce a segmentation mask and/or bounding boxes to identify the targeted data.

Abstract: A trained model is described in the form of a Bayesian neural network (BNN) which provides a quantification of its inference uncertainty during use and which is trained using marginal likelihood maximization. A Probably Approximately Correct (PAC) bound may be used in the training to incorporate prior knowledge and to improve training stability even when the network architecture is deep.

Abstract: An image analysis method for analyzing an image of a tissue collected from a subject using a deep learning algorithm of a neural network structure. The image analysis method includes generating analysis data from the analysis target image that includes the tissue to be analyzed, inputting the analysis data to a deep learning algorithm, and generating data indicating a layer structure configuring a tissue in the analysis target image by the deep learning algorithm.

Abstract: A method and system for extracting information from a drawing. The method includes classifying nodes in the drawing, extracting attributes from the nodes, determining whether there are errors in the node attributes, and removing the nodes from the drawing. The method also includes identifying edges in the drawing, extracting attributes from the edges, and determining whether there are errors in the edge attributes. The system includes at least one processing component, at least one memory component, an identification component, an extraction component, and a correction component. The identification component is configured to classify nodes in the drawing, remove the nodes from the drawing, and identify edges in the drawing. The extraction component is configured to extract attributes from the nodes and edges. The correction component is configured to determine whether there are errors in the extracted attributes.

Abstract: There is provided is a method and an apparatus for training a neural network capable of improving the performance of the neural network by performing intelligent normalization according to a target task of the neural network. The method according to some embodiments of the present disclosure includes transforming the output data into first normalized data using a first normalization technique, transforming the output data into second normalized data using a second normalization technique and generating target normalized data by aggregating the first normalized data and the second normalized data based on a learnable parameter. At this time, a rate at which the first normalization data is applied in the target normalization data is adjusted by the learnable parameter so that the intelligent normalization according to the target task can be performed, and the performance of the neural network can be improved.

Abstract: A method for image decoding performed by a decoding apparatus, according to the present disclosure, comprises the steps of: obtaining residual information for a current block from a bitstream; deriving a prediction sample for the current block; deriving a residual sample for the current block on the basis of the residual information; deriving a reconstructed picture on the basis of the prediction sample and the residual sample; and performing filtering on the reconstructed picture on the basis of a convolution neural network (CNN).

Abstract: An example method includes generating, using a multi-scale block of a convolutional neural network (CNN), a first output image based on an optical coherence tomography (OCT) reflectance image of a retina and an OCT angiography (OCTA) image of the retina. The method further includes generating, using an encoder of the CNN, at least one second output image based on the first output image and generating, using a decoder of the CNN, a third output image based on the at least one second output image. An avascular map is generated based on the third output image. The avascular map indicates at least one avascular area of the retina depicted in the OCTA image.

Abstract: In part, the disclosure relates to methods, and systems suitable for evaluating image data from a patient on a real time or substantially real time basis using machine learning (ML) methods and systems. Systems and methods for improving diagnostic tools for end users such as cardiologists and imaging specialists using machine learning techniques applied to specific problems associated with intravascular images that have polar representations. Further, given the use of rotating probes to obtain image data for OCT, IVUS, and other imaging data, dealing with the two coordinate systems associated therewith creates challenges. The present disclosure addresses these and numerous other challenges relating to solving the problem of quickly imaging and diagnosis a patient such that stenting and other procedures may be applied during a single session in the cath lab.

Abstract: The present disclosure describes methods, devices, and storage medium for recognizing a target object in a target image. The method including obtaining, by a device, an image recognition instruction, the image recognition instruction carrying object identification information used for indicating a target object in a target image. The device includes a memory storing instructions and a processor in communication with the memory. The method includes obtaining, by the device, an instruction feature vector matching the image recognition instruction; obtaining, by the device, an image feature vector set matching the target image, the image feature vector set comprising an ith image feature vector for indicating an image feature of the target image in an ith scale, and i being a positive integer; and recognizing, by the device, the target object from the target image according to the instruction feature vector and the image feature vector set.

Abstract: An automatic target recognizer system including: a database that stores target recognition data including multiple reference features associated with each of multiple reference targets; a pre-selector that selects a portion of the target recognition data based on a reference gating feature of the multiple reference features; a preprocessor that processes an image received from an image acquisition system which is associated with an acquired target and determines an acquired gating feature of the acquired target; a feature extractor and processor that discriminates the acquired gating feature with the reference gating feature and, if there is a match, extracts multiple segments of the image and detects the presence, absence, probability or likelihood of one of multiple features of each of the multiple reference targets; a classifier that generates a classification decision report based on a determined classification of the acquired target; and a user interface that displays the classification decision report.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for selecting locations in an environment of a vehicle where objects are likely centered and determining properties of those objects. One of the methods includes receiving an input characterizing an environment external to a vehicle. For each of a plurality of locations in the environment, a respective first object score that represents a likelihood that a center of an object is located at the location is determined. Based on the first object scores, one or more locations from the plurality of locations are selected as locations in the environment at which respective objects are likely centered. Object properties of the objects that are likely centered at the selected locations are also determined.

Abstract: Real-time scene understanding system employing an object detection module with an algorithm for localization and classification of objects in an image, and a semantic segmentation module with an algorithm for classification of individual pixels in the image, wherein the system comprises an encoder module operable on an input image for the extraction of notable features in the input image, one or more attention modules to attribute among the notable features in the input image as provided by the encoder a relative contribution of each of such notable features in an output image to be reconstructed from the input image, and a decoder module for reconstructing the output image using the notable features, wherein the reconstructed output image is made available to the object detection module with the algorithm for localization and classification of objects in the image, and to the semantic segmentation module with the algorithm for classification of individual pixels in the image.

Abstract: Methods and systems are provided for augmenting ultrasound image training data, which may be used to train one or more machine learning models. One example method for augmenting ultrasound training data comprises, selecting an ultrasound image and a ground truth output corresponding to the ultrasound image, determining a first modification to apply to the ultrasound image, applying the first modification to the ultrasound image to produce an augmented ultrasound image, modifying the ground truth output based on the first modification to produce an augmented ground truth output corresponding to the augmented ultrasound image, and training a machine learning model using the augmented ultrasound image and the augmented ground truth output. In this way, a machine learning model may learn a more robust mapping from ultrasound image features to expected output, with less probability of overfitting, and with increased generalizability to noisy ultrasound images, or ultrasound images containing artifacts.

Abstract: A method of training a heterogeneous convolutional neural network (HCNN) system includes identifying batch sizes for a first task and a second task, defining images for a first batch, a second batch, and a batch x for the first task, defining images for a first batch, a second batch, and a batch y for the second task, training the HCNN using the first batch for the first task, training the HCNN using the first batch for the second task, training the HCNN using the second batch for the first task, training the HCNN using the second batch for the second task. The sequential training continues for each of the batches and each of the tasks until the end of an epoch. When the epoch is complete, the images for each batch and each task are reshuffled.

Abstract: Multiple trained AI models are tested using known genuine samples of respective multiple modalities of multimedia to generate versions of the multiple modalities of a given multimedia sample. Data for the multimedia and the multimedia sample are divided into the multiple modalities. Respective differences are computed between respective components of the multiple trained AI models to produce respective multiple difference vector, which are compared with corresponding baseline difference vectors determined in order to train the multiple trained AI models. The given multimedia sample is classified as genuine or altered using at least the comparison.

Abstract: Methods and systems regarding importance sampling for the modification of a training procedure used to train a segmentation network are disclosed herein. A disclosed method includes segmenting an image using a trainable directed graph to generate a segmentation, displaying the segmentation, receiving a first selection directed to the segmentation, and modifying a training procedure for the trainable directed graph using the first selection. In a more specific method, the training procedure alters a set of trainable values associated with the trainable directed graph based on a delta between the segmentation and a ground truth segmentation, the first selection is spatially indicative with respect to the segmentation, and the delta is calculated based on the first selection.

Abstract: A system and method for automated generation of control signals for sexual stimulation devices from videos of sexual activity. The system and method use annotation data indicating comprising indications of movements in the video corresponding to sexual activity to generate control signals for compatible sexual stimulation devices. The annotation data may be generated manually or automatically. Device control signals may be generated directly from the annotation data and synchronized with a video, or may be processed through a series of machine learning algorithms to generate models of “typical” sexual activity represented in the videos, which models are then used to generate signals for the sexual stimulation device.

Abstract: An image processing method for recognising characters included in an image. A first character recognition unit performs recognition of a first group of characters corresponding to a first region of the image. A measuring unit calculates a confidence measure of the first group of characters. A determination unit determines whether further recognition is to be performed based on the confidence measure. A selection unit selects a second region of the image that includes the first region, if it is determined that further recognition is to be performed. A second character recognition unit performs further recognition of a second group of characters corresponding to the second region of the image.

Abstract: Techniques for training a machine-learning (ML) model for captioning images are disclosed. A plurality of feature vectors and a plurality of visual attention maps are generated by a visual model of the ML model based on an input image. Each of the plurality of feature vectors correspond to different regions of the input image. A plurality of caption attention maps are generated by an attention model of the ML model based on the plurality of feature vectors. An attention penalty is calculated based on a comparison between the caption attention maps and the visual attention maps. A loss function is calculated based on the attention penalty. One or both of the visual model and the attention model are trained using the loss function.

Abstract: A deep machine-learning approach is used for medical image fusion by a medical imaging system. This one approach may be used for different applications. For a given application, the same deep learning is used but with different application-specific training data. The resulting deep-learnt classifier provides a reduced feature vector in response to input of intensities of one image and displacement vectors for patches of the one image relative to another image. The output feature vector is used to determine the deformation for medical image fusion.

Abstract: Edge-Loss-based image construction is enabled by a method including generating a reconstructed image from a first edge image with a generator, extracting a second edge image from the reconstructed image with an edge extractor, smoothing the first edge image and the second edge image, discriminating between the reconstructed image and an original image corresponding to the first edge image with a discriminator to obtain an adversarial loss, and training the generator by using an edge loss and the adversarial loss, the edge loss being calculated from the smoothed first edge image and the smoothed second edge image.

Abstract: Disclosed herein are system, method, and computer program product embodiments for analyzing contextual symbol information for document processing. In an embodiment, a language model system may generate a vector grid that incorporates contextual document information. The language model system may receive a document file and identify symbols of the document file to generate a symbol grid. The language model system may also identify position parameters corresponding to each of the symbols. The language model system may then analyze the symbols using an embedding function and neighboring symbols to determine contextual vector values corresponding to each of the symbols. The language model system may then generate a vector grid mapping the contextual vector values using the position parameters. The contextual information from the vector grid may provide increase document processing accuracy as well as faster processing convergence.

Abstract: A computer, including a processor and a memory, the memory including instructions to be executed by the processor to generate two or more stereo pairs of synthetic images and generate two or more stereo pairs of real images based on the two or more stereo pairs of synthetic images using a generative adversarial network (GAN), wherein the GAN is trained using a six-axis degree of freedom (DoF) pose determined based on the two or more pairs of real images. The instructions can further include instructions to train a deep neural network based on a sequence of real images and operate a vehicle using the deep neural network to process a sequence of video images acquired by a vehicle sensor.

Abstract: An object recognition method and apparatus for a deformed image are provided. The method includes: inputting an image into a preset localization network to obtain a plurality of localization parameters for the image, wherein the preset localization network comprises a preset number of convolutional layers, and wherein the plurality of localization parameters are obtained by regressing image features in a feature map that is generated from a convolution operation on the image; performing a spatial transformation on the image based on the plurality of localization parameters to obtain a corrected image; and inputting the corrected image into a preset recognition network to obtain an object classification result for the image. In the process of the neural network based object recognition, the embodiment of the present application first transforms the deformed image that has deformation, and then performs the object recognition on the transformed image.

Abstract: Methods and systems are provided to generate an uncorrupted version of an image given an observed image that is a corrupted version of the image. In some embodiments, a corruption mimicking (“CM”) system iteratively trains a corruption mimicking network (“CMN”) to generate corrupted images given modeled images, updates latent vectors based on differences between the corrupted images and observed images, and applies a generator to the latent vectors to generate modeled images. The training, updating, and applying are performed until modeled images that are input to the CMN result in corrupted images that approximate the observed images. Because the CMN is trained to mimic the corruption of the observed images, the final modeled images represented the uncorrupted version of the observed images.

Abstract: A neural network is provided. The neural network includes 2n number of sampling units sequentially connected; and a plurality of processing units. A respective one of the plurality of processing units is between two adjacent sampling units of the 2n number of sampling units. A first sampling unit to an n-th sample unit of the 2n number of sampling units are DeMux units. A respective one of the DeMux units is configured to rearrange pixels in a respective input image to the respective one of the DeMux units following a first scrambling rule to obtain a respective rearranged image. An (n+1)-th sample unit to a (2n)-th sample unit of the 2n number of sampling units are Mux units. A respective one of the Mux units is configured to combing respective m? number of input images to the respective one of the Mux units to obtain a respective combined image.

Abstract: This disclosure relates to methods, non-transitory computer readable media, and systems that can initially train a machine-learning-logo classifier using synthetic training images and incrementally apply the machine-learning-logo classifier to identify logo images to replace the synthetic training images as training data. By incrementally applying the machine-learning-logo classifier to determine one or both of logo scores and positions for logos within candidate logo images, the disclosed systems can select logo images and corresponding annotations indicating positions for ground-truth logos. In some embodiments, the disclosed systems can further augment the iterative training of a machine-learning-logo classifier to include user curation and removal of incorrectly detected logos from candidate images, thereby avoiding the risk of model drift across training iterations.

Abstract: A method includes receiving a user object specified by a user. A similarity score is computed using a similarity function between the user object and one or more candidate objects in a database based on respective feature vectors. A first subset of the one or more candidate objects is presented to the user based on the respective computed similarity scores. First feedback is received from the user about the first subset of candidate objects. The similarity function is adjusted based on the received first feedback.

Abstract: The present disclosure relates to systems, devices and methods for identifying objects and scenarios that have not been trained or are unidentifiable to vehicle perception sensors or vehicle assistive driving systems. Embodiments are directed to using a trained vehicle data set to identify target objects in vehicle sensor data. In one embodiment, a process is provided that includes running a scene detection operation on vehicle to derive a vector of target object attributes of the vehicle sensor data and generating a vector representation for the scene detection operation and the attributes of the vehicle sensor data. The vector representation compared to a familiarity vector to represent effectiveness of the scene detection operation. In addition, the vector representation can be scored to identify one or more target objects or significant scenarios, including unidentifiable objects and/or driving scenes, scenarios for reporting.

Abstract: Disclosed are methods, apparatus and systems for gesture recognition based on neural network processing. One exemplary method for identifying a gesture communicated by a subject includes receiving a plurality of images associated with the gesture, providing the plurality of images to a first 3-dimensional convolutional neural network (3D CNN) and a second 3D CNN, where the first 3D CNN is operable to produce motion information, where the second 3D CNN is operable to produce pose and color information, and where the first 3D CNN is operable to implement an optical flow algorithm to detect the gesture, fusing the motion information and the pose and color information to produce an identification of the gesture, and determining whether the identification corresponds to a singular gesture across the plurality of images using a recurrent neural network that comprises one or more long short-term memory units.