Abstract: An image inpainting method is provided. The image inpainting method includes determining a missing region in an original image, generating an input image to be reconstructed from the original image, based on the missing region, obtaining a mask image indicating the missing region, determining whether to extract a structural feature of the missing region, based on an attribute of the missing region, obtaining structure vectors each consisting of one or more lines and one or more junctions by applying the input image and the mask image to a first model for extracting a structural feature of the input image, and obtaining an inpainted image in which the missing region in the input image is reconstructed by applying the input image, the mask image, and a structure vector image converted from the structure vectors to a second model for reconstructing the input image.

Abstract: Methods, systems, and storage media are described for motion estimation in video frame interpolation. Disclosed embodiments use feature pyramids as image representations for motion estimation and seamlessly integrates them into a deep neural network for frame interpolation. A feature pyramid is extracted for each of two input frames. These feature pyramids are wrapped together with the input frames to the target temporal position according to the inter-frame motion estimated via optical flow. A frame synthesis network is used to predict interpolation results from the pre-warped feature pyramids and input frames. The feature pyramid extractor and the frame synthesis network are jointly trained for the task of frame interpolation. An extensive quantitative and qualitative evaluation demonstrates that the described embodiments utilizing feature pyramids enables robust, high-quality video frame interpolation. Other embodiments may be described and/or claimed.

Abstract: An image classification system defends against physically realizable attacks. A training dataset of input images is retrieved and an adversarial image is generated based on one of the input images that is selected. The adversarial image is created by occluding a portion of the selected image by superimposing a predetermined shape (e.g., a rectangle) containing noise on the selected image. A defense against occlusion attacks (DOA) classifier is trained using the training dataset and the adversarial image. The DOA classifier is utilized to classify captured images of items (e.g., street signs) that may have been attacked (e.g., sticker placement, vandalism).

Abstract: Technology is provided for generating conversation features for recorded conversations. The technology includes, receiving videos depicting a multiple-user interaction, segmenting the videos into multiple utterances based on identifying utterances from individual users, receiving label data for the utterance segments specifying conversation features, and storing the label data in association with the utterance segments.

Abstract: A self-contained, low-cost, low-weight guidance system for vehicles is provided. The guidance system can include an optical camera, a case, a processor, a connection between the processor and an on-board control system, and computer algorithms running on the processor. The guidance system can be integrated with a vehicle control system through “plug and play” functionality or a more open Software Development Kit. The computer algorithms re-create 3D structures as the vehicle travels and continuously updates a 3D model of the environment. The guidance system continuously identifies and tracks terrain, static objects, and dynamic objects through real-time camera images. The guidance system can receive inputs from the camera and the onboard control system. The guidance system can be used to assist vehicle navigation and to avoid possible collisions. The guidance system can communicate with the control system and provide navigational direction to the control system.

Abstract: A method of testing an impedance-sensitive system with a switching device, wherein the method comprises: switching the disconnecting device into the on-state configured to permit transmission of energy via the coil; implementing a first measurement with the impedance-sensitive system; switching the disconnecting device into the off-state configured to permit damping of the external positioning signal that couples into the coil so as to reduce the undesirable oscillations of the coil; implementing a second measurement with the impedance-sensitive system; performing a comparison of the first measurement and the second measurement; performing a verification of the comparison with a target specification; and displaying a correct function and/or a malfunction depending on the verification.

Abstract: A method of training and using a machine learning model that controls for consideration of undesired factors which might otherwise be considered by the trained model during its subsequent analyses of new data. For example, the model may be a neural network trained on a set of training images to evaluate an insurance applicant based upon an image or audio data of the insurance applicant as part of an underwriting process to determine an appropriate life or health insurance premium. The model is trained to probabilistically correlate an aspect of the applicant's appearance with a personal and/or health-related characteristic. Any undesired factors, such as age, sex, ethnicity, and/or race, are identified for exclusion. The trained model receives the image (e.g., a “selfie”) of the insurance applicant, analyzes the image without considering the identified undesired factors, and suggests the appropriate insurance premium based only on the remaining desired factors.

Abstract: An apparatus for inspecting a pharmaceutical cylindrical container made of glass or of a polymer is provided. A corresponding method for inspecting the pharmaceutical cylindrical container made of glass or of a polymer and to a bundle of pharmaceutical cylindrical containers made of glass or of a polymer is also provided.

Abstract: A method for using a single-pixel camera to reconstruct images of objects obscured by fog or other dynamic scattering media. A pseudo-random phase or intensity pattern is imposed on illumination beams directed at a target. The beam with the imposed pattern forms a pseudo random pattern on the target. Information regarding the pattern imposed on each pulse is entered into a data processor/controller. The illumination beams with the pseudo random patterns are reflected off the target, collected by receiving optics and a bucket detector and converted into electronic signals fed into the data processor/controller. The data processor/controller applies a high-pass filter to remove slower signal variations produced by dynamic changes in the scattering medium over time. The filtered bucket values are then used together with their corresponding speckle patterns to generate the images using any appropriate reconstruction algorithm such as CGI or CSI.

Abstract: The technology disclosed also initializes a new hand that enters the field of view of a gesture recognition system using a parallax detection module. The parallax detection module determines candidate regions of interest (ROI) for a given input hand image and computes depth, rotation and position information for the candidate ROI. Then, for each of the candidate ROI, an ImagePatch, which includes the hand, is extracted from the original input hand image to minimize processing of low-information pixels. Further, a hand classifier neural network is used to determine which ImagePatch most resembles a hand. For the qualified, most-hand like ImagePatch, a 3D virtual hand is initialized with depth, rotation and position matching that of the qualified ImagePatch.

Abstract: Aspects of the present disclosure provide systems, methods, and computer-readable storage media that support detection and mitigation of defects occurring in outputs of a process. A modelling engine creates and trains models to identify defects occurring in outputs of a process or system. A set of causation rules may be created and trained to identify causes of different defects identifiable via the models and a set of control rules may be created and trained to generate control data that may mitigate the causes of identified defects. A process monitoring device incorporating the trained models and rules may then be used to monitor the process and detect defects using the models. Once a defect is detected, the rules may be applied to determine the cause of the defect and then control data may be generated to modify the process such that further occurrences of the defect are reduced or eliminated.

Abstract: Embodiments include methods, systems, and computer program products for communication control. One such embodiment receives a telephone call from a plain old telephone service (POTS) device. The telephone call includes a telephone number corresponding to a destination device. Next, a ringing tone is provided to the POTS device and the telephone number is transmitted to a proxy agent. A message is received from the proxy agent. The message indicates that a telephonic connection has been established between the proxy agent and the destination device. Responsive to receiving the message from the proxy agent, a real-time communications connection between the POTS device and the destination device is created by providing answer signaling for the received telephone call to the POTS device.

Abstract: One embodiment of the invention provides a method for video recognition. The method comprises receiving an input video comprising a sequence of video segments over a plurality of data modalities. The method further comprises, for a video segment of the sequence, selecting one or more data modalities based on data representing the video segment. Each data modality selected is optimal for video recognition of the video segment. The method further comprises, for each data modality selected, providing at least one data input representing the video segment over the data modality selected to a machine learning model corresponding to the data modality selected, and generating a first type of prediction representative of the video segment via the machine learning model. The method further comprises determining a second type of prediction representative of the entire input video by aggregating all first type of predictions generated.

Abstract: A device may receive tracking data identifying trajectories of objects, and may annotate the tracking data to identify object categories for the tracking data. The device may identify objects based on the object categories, and may determine object trajectories for the objects. The device may transform the object trajectories to overhead planar trajectories, and may generate simplified object trajectories. The device may convert a continuous space associated with the simplified object trajectories into a discrete space, and may convert the discrete space to a lower dimensional space. The device may transform the object trajectories to binary feature vectors, and may process the binary feature vectors, with a clustering model, to determine trajectory clusters. The device may train a classification model with the binary feature vectors and the trajectory clusters, and may cause the trained classification model to be implemented.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network to predict optical flow. One of the methods includes obtaining a batch of one or more training image pairs; for each of the pairs: processing the first training image and the second training image using the neural network to generate a final optical flow estimate; generating a cropped final optical flow estimate from the final optical flow estimate; and training the neural network using the cropped optical flow estimate.

Abstract: A system includes an electronic display, a computer processor, one or more memory units, and a module stored in the one or more memory units. The module is configured to access a source image stored in the one or more memory units and determine depth data for each pixel of a plurality of pixels of the source image. The module is further configured to map, using the plurality of pixels and the determined depth data for each of the plurality of pixels, the source image to a four-dimensional light field. The module is further configured to send instructions to the electronic display to display the mapped four-dimensional light field.

Abstract: Disclosed is an automatic image analysis method that can be used to automatically recognise at least one rare characteristic in an image to be analysed. The method comprises a learning phase during which at least one convolutional neural network is trained to recognise characteristics, a parameter space of dimension n, in which n?2, is constructed from at least one intermediate layer of the network, a presence probability function is determined for each characteristic in the parameter space from a projection of reference images in the parameter space. During a phase of analysing the image to be analysed, the method comprises a step of recognising the at least one rare characteristic in the image to be analysed on the basis of the presence probability function determined for the at least one rare characteristic.

Abstract: An information processing device includes: a storage unit that stores first information indicating at least one of a structure or a physical property of a first object obtained by capturing a real physical body into a virtual space; and a specifying unit that specifies an arrangement of a second object indicating a virtual object in the virtual space so that the arrangement is capable of expressing an interaction with the first object, based on an arrangement condition of the second object and on the first information.

Abstract: Methods, apparatus, systems and articles of manufacture are disclosed. An example apparatus includes an audio detector to determine a first audience count based on signatures of audio data captured in the media environment, a thermal image detector to determine a heat blob count based on a frame of thermal image data captured in the media environment, and an audience image detector to identify at least one audience member based on a comparison of a frame of audience image data with a library of reference audience images, the audience image detector to perform the comparison in response to the first audience count not matching the heat blob count.

Abstract: A system for identifying handwritten characters on an image using a classification model that employs a neural network. The system includes a computer having a processor and a memory device that stores data and executable code that, when executed, causes the processor to read and convert typed text on the image to machine encoded text to identify locations of the typed text on the image; identify a location on the image that includes handwritten text based on the location of predetermined typed text on the image; identify clusters of non-white pixels in the image at the location having the handwritten text; generate an individual and separate cluster image for each identified cluster; classify each cluster image using machine learning and at least one neural network to determine the likelihood that the cluster is a certain character; and determine the accuracy of the characters by comparing to a secondary database.