Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images or features of images using an image classification system that includes a batch normalization layer. One of the systems includes a convolutional neural network configured to receive an input comprising an image or image features of the image and to generate a network output that includes respective scores for each object category in a set of object categories, the score for each object category representing a likelihood that that the image contains an image of an object belonging to the category, and the convolutional neural network comprising: a plurality of neural network layers, the plurality of neural network layers comprising a first convolutional neural network layer and a second neural network layer; and a batch normalization layer between the first convolutional neural network layer and the second neural network layer.

Abstract: An individual identification apparatus that identifies an individual product having a pattern of irregularities randomly formed on a surface thereof includes an imaging unit and an extracting unit. The imaging unit is configured to acquire an image obtained by capturing a date mark formed on the product. The extracting unit is configured to extract a feature value related to the pattern of the irregularities from the image as data identifying the individual product.

Abstract: There are provided systems and methods of obtaining a segmented image of a semiconductor specimen, the image comprising first structural elements, obtaining a reference image of the semiconductor specimen, the reference image being based on design data and comprising second structural elements, determining, for at least one pair of elements including a first structural element and a corresponding second structural element, data Dspat informative of a spatial transformation required in order to match the elements of the pair in accordance with a matching criterion, and determining at least one of data informative of a defect in the first structural element and data informative of edge roughness of the first structural element using at least Dspat.

Abstract: A method includes obtaining at least one image and a ground truth map associated with the at least one image. The method also includes generating multiple optical maps using multiple algorithms and the at least one image. The method further includes, for each algorithm, identifying at least one score for the algorithm using one or more of the optical maps generated using the algorithm and the ground truth map. The ground truth map identifies one or more boundaries associated with one or more foreground objects in the at least one image. The scores identify how well the optical maps generated using the algorithms separate the one or more foreground objects from a background in the at least one image.

Abstract: An apparatus includes a first acquisition unit that acquires a plurality of likelihood maps by setting a plurality of different weight parameters in a trained model that outputs, with an image feature extracted from the input image as an input, a likelihood map including, in association with a position in the input image, a likelihood indicating a possibility that the object is present, and a detection unit that detects, based on the acquired plurality of the likelihood maps, the position of the object included in the input image, wherein the trained model is a model that has learned the weight parameters based on loss values at least acquired using a first loss function for reducing a likelihood around a position of interest in the likelihood map, and a second loss function for increasing a likelihood acquired at the position of the object in the input image.

Abstract: A feature surface projection system includes a projector, a camera, and a controller. The projector projects a structured light and an image beam to a feature surface at different times. The structured light forms a structured light pattern on the feature surface, and the image beam forms an image on the feature surface. The camera photographs the structured light pattern and the image at different times. The controller is electrically connected to the projector and the camera, calculates a position of a feature point of the feature surface, and determines a difference between a position of a feature point of the image and the position of the feature point of the feature surface to determine whether to adjust the image beam. A projection method for a feature surface is also provided. The feature surface projection system provides an automatic alignment function between a projected content and the feature surface.

Abstract: In order to enable, even when an increased number of workers become analysis targets, efficient analysis on work efficiency statuses of workers without putting a greater burden on an administrator to perform setting operations for the analysis, analytical information is produced by performing a joint position estimation based on respective frames of video recordings to estimate joint positions of a worker, performing a posture match determination based on results of the joint position estimation to determine whether or not to achieve a posture match between a posture of the worker in each frame of the video recordings and a predetermined reference body posture, and measuring, based on results of the posture match determination, a working time and a presence time of the worker used as analytical information on a work efficiency status of the worker.

Abstract: A system, method, and computer-program product includes detecting, via a localization machine learning model, a target object within a scene based on downsampled image data of the scene, identifying a likely position of the target object within original image data of the scene, extracting, from the original image data of the scene, a target sub-image containing the target object, classifying, via an object classification machine learning model, the target object to a probable object class of a plurality of distinct object classes, routing the target image resolution of the target sub-image to a target object-condition machine learning classification model of a plurality of distinct object-condition machine learning classification models, classifying, via the target object-condition machine learning classification model, the target object to a probable object-condition class, and displaying, via a graphical user interface, a representation of the target object in association with the probable object-condition

Abstract: A technology is described for spatially estimating thermal emissivity. A method can include obtaining spectral emissivity data and satellite imaging data for a geographic area. A weighted emissivity model of emissivity values may be generated for surfaces included in the geographic area from the spectral emissivity data and the satellite imaging data, wherein the spectral emissivity data is mapped to the satellite imaging data to generate the weighted emissivity model. Thermal imaging data for the geographic area may be received from an airborne thermal imaging sensor and a thermal emissivity map can be generated for the geographic area using the thermal imaging data and the weighted emissivity model. The emissivity values from the weighted emissivity model can be used to estimate thermal emissivity values.

Abstract: Image processing methods, devices, and storage media are provided. One of the image processing methods includes: obtaining an image comprising a target object; inputting the image into a mask prediction model to obtain a first mask image corresponding to the target object in the image; inputting the first mask image and the image comprising the target object into a mask quality unification model to obtain a second mask image, wherein the mask quality unification model is configured to adjust a fineness associated with the first mask image to a target fineness to obtain the second mask image, and wherein the second mask image has image semantic information consistent with image semantic information of the image comprising the target object; and inputting the image comprising the target object and the second mask image into a target image obtaining model to obtain a target image corresponding to the target object.

Abstract: Techniques are described for estimating pose of a camera located on a vehicle. An exemplary method of estimating camera pose includes obtaining, from a camera located on a vehicle, an image including a lane marker on a road on which the vehicle is driven, and estimating a pose of the camera such that the pose of the camera provides a best match according to a criterion between a first position of the lane marker determined from the image and a second position of the lane marker determined from a stored map of the road.

Abstract: A method and apparatus for determining a location of a signal light, a storage medium, a program and a roadside device provided by the present application, which relate to intelligent traffic technology and automatic driving technology, including: for a plurality of frames of images captured including the signal light, determining horizontal edges and vertical edges in each of the images; combining, according to the obtained horizontal edges and vertical edges, to obtain a plurality of rectangular frames; and determining, according to the plurality of rectangular frames and preset size information of a frame of the signal light, the location of the signal light in the image.

Abstract: The present disclosure is directed to systems and methods for estimating machine rotation state. The rotation state can include the rotation speed and/or angle as the machine rotates. In some implementations, an estimating machine rotation state system can receive image data, timestamped consecutively, from camera devices of a machine. The estimating machine rotation state system can generate optical flow vectors for the pixels of the image data and determine the hue and intensity of the pixels based on the optical flow vectors. Using the hues, intensities, and timestamps of the image data, the estimating machine rotation state system can generate an intensity plot over time. The estimating rotation states system can then determine a rotation speed of the machine based on the intensity plot. The estimating rotation states system can also determine a rotation angle of the machine by integrating the intensity plot over time.

Abstract: Systems and methods for calibration of a surgical tool having a tracker. At least one camera captures images of the tracker at a first exposure and captures images of the surgical tool at a second exposure with a different exposure time and/or illumination than the first exposure. Controller(s) recognize a pose of the tracker based on the first exposure images and recognize a geometry of the surgical tool based on the second exposure images. The controller(s) correlate the recognized pose of the tracker and the recognized geometry of the surgical tool to define a relationship between the tracker and the surgical tool and calibrate the surgical tool based on the defined relationship.

Abstract: Systems and methods to perform noise-adaptive non-blind deblurring on an input image that includes blur and noise involve implementing a first neural network on the input image to obtain one or more parameters and performing regularized deconvolution to obtain a deblurred image from the input image. The regularized deconvolution uses the one or more parameters to control noise in the deblurred image. A method includes implementing a second neural network to remove artifacts from the deblurred image and provide an output image.

Abstract: This disclosure describes one or more implementations of a depth prediction system that generates accurate depth images from single input digital images. In one or more implementations, the depth prediction system enforces different sets of loss functions across mix-data sources to generate a multi-branch architecture depth prediction model. For instance, in one or more implementations, the depth prediction model utilizes different data sources having different granularities of ground truth depth data to robustly train a depth prediction model. Further, given the different ground truth depth data granularities from the different data sources, the depth prediction model enforces different combinations of loss functions including an image-level normalized regression loss function and/or a pair-wise normal loss among other loss functions.

Abstract: In an embodiment, a volumetry method is provided. The method can include receiving one or more 2D images of a site including one or more containers configured to hold a liquid, the 2D images being acquired at a first time. The method can also include identifying a container within the 2D images. The method can further include selecting an identified container within the 2D images for volumetric analysis based upon a degree of occlusion of its front facing surface. The method can additionally include classifying a portion of the selected container containing the liquid. The method can further include determining the volume of liquid held within the container based upon the classified portion.

Abstract: Systems and methods are disclosed for universal body movement translation and character rendering. Motion data from a source character can be translated and used to direct movement of a target character model in a way that respects the anatomical differences between the two characters. The movement of biomechanical parts in the source character can be converted into normalized values based on defined constraints associated with the source character, and those normalized values can be used to inform the animation of movement of biomechanical parts in a target character based on defined constraints associated with the target character.

Abstract: A method of processing an image for improving image quality is provided. The method includes inputting an image into a first artificial neural network as an input variable to create a feature map and inputting an image into a second artificial neural network as an input variable to create an estimated class probability distribution map for each of frequency channels. Further, the method may comprise determining a class for each of the frequency channels on the basis of a probability distribution included in the estimated class probability distribution map, creating an estimated class map that includes a determined class, converting the estimated class map into an estimated frequency map, combining the estimated frequency map with the feature map to create a combined feature map, and inputting the combined feature map into a third artificial neural network as an input variable to create an output image.

Abstract: A system includes sensors onboard an autonomous vehicle, a processor, and a memory. The memory stores instructions for the processor to receive a first image pair from the sensors at a first time and a second image pair from the sensors at a second time. Each image pair includes at least one static feature in an environment of the autonomous vehicle. The memory also stores instructions to determine a distance travelled by the autonomous vehicle between the first and second times, and to determine a correction to a disparity map based on (1) a first disparity associated with the static feature(s) and the first image pair, (2) a second disparity associated with the static feature(s) and the second image pair, and (3) the distance travelled. The memory also stores instructions to cause the correction to be applied to the disparity map.