An example device is described for facilitating polygon localization. In various aspects, the device can comprise a processor. In various instances, the device can comprise a non-transitory machine-readable memory that can store machine-readable instructions. In various cases, the processor can execute the machine-readable instructions, which can cause the processor to localize a polygon depicted in an image, based on execution of a deep learning pipeline. In various aspects, the deep learning pipeline can comprise a circular-softmax block.

A cross reality system enables portable devices to access stored maps and efficiently and accurately render virtual content specified in relation to those maps. The system may process images acquired with a portable device to quickly and accurately localize the portable device to the persisted maps by constraining the result of localization based on the estimated direction of gravity of a persisted map and the coordinate frame in which data in a localization request is posed. The system may actively align the data in the localization request with an estimated direction of gravity during the localization processing, and/or a portable device may establish a coordinate frame in which the data in the localization request is posed aligned with an estimated direction of gravity such that the subsequently acquired data for inclusion in a localization request, when posed in that coordinate frame, is passively aligned with the estimated direction of gravity.

Methods, systems, and non-transitory computer readable storage media are disclosed that utilizes machine learning models for patch retrieval and deformation in completing three-dimensional digital shapes. In particular, in one or more implementations the disclosed systems utilize a machine learning model to predict a coarse completion shape from an incomplete 3D digital shape. The disclosed systems sample coarse 3D patches from the coarse 3D digital shape and learn a shape distance function to retrieve detailed 3D shape patches in the input shape. Moreover, the disclosed systems learn a deformation for each retrieved patch and blending weights to integrate the retrieved patches into a continuous surface.

A cross reality system enables portable devices to access stored maps and efficiently and accurately render virtual content specified in relation to those maps. The system may process images acquired with a portable device to quickly and accurately localize the portable device to the persisted maps by constraining the result of localization based on the estimated direction of gravity of a persisted map and the coordinate frame in which data in a localization request is posed. The system may actively align the data in the localization request with an estimated direction of gravity during the localization processing, and/or a portable device may establish a coordinate frame in which the data in the localization request is posed aligned with an estimated direction of gravity such that the subsequently acquired data for inclusion in a localization request, when posed in that coordinate frame, is passively aligned with the estimated direction of gravity.

Implementations relate to dynamic adaptation of images for projection by a projector, based on one or more properties of user(s) that are in an environment with the projector. The projector can be associated with an automated assistant client of a client device. In some versions of those implementations, a pose of a user in the environment is determined and, based on the pose, a base image for projecting onto a surface is warped to generate a transformed image. The transformed image, when projected onto a surface and viewed from the pose of the user, mitigates perceived differences relative to the base image. The base image (on which the transformed image is based) can optionally be generated in dependence on a distance of the user. Some implementations additionally or alternatively relate to dynamic adaptation of projection parameters (e.g., a location for projection, a size of projection) based on one or more properties of user(s) that are in an environment with the projector.

An image matching apparatus matching a first image against a second image includes an acquiring unit, a generating unit, and a determining unit. The acquiring unit acquires a frequency feature of the first image and a frequency feature of the second image. The generating unit synthesizes the frequency feature of the first image and the frequency feature of the second image, and generates a quantized synthesized frequency feature in which a value of an element is represented by a binary value or a ternary value. The determining unit calculates a score indicating a degree to which the quantized synthesized frequency feature is a square wave having a single period, and matches the first image against the second image based on the score.

An image comparison apparatus includes an image comparison portion configured to compare a reference image with a target image. The image comparison portion calculates a degree of blur of each of the reference image and the target image. The image comparison apparatus reduces a difference in the degree of blur between the reference image and the target image to within a specific range by adding blur to one of the reference image and the target image that has a smaller degree of blur. The image comparison apparatus compares the reference image and the target image whose difference in the degree of blur has been reduced to within the specific range.

An authentication processing system includes a memory storing a portrait fraud detection application, and a processing unit coupled with the memory and configured to execute the portrait fraud detection application. The portrait fraud detection application, when executed, configures the processing unit to receive a capture of a document including a portrait photo and at least one overlay, detect a face in the portrait photo among the at least one overlay in the capture, and determine the portrait photo is fraudulent; and initiate an indication the document is fraudulent.

Embodiments provide a method of using image-based tube top circle detection that includes extracting, from one of a series of images of a tube tray, a region of interest (ROI) patch having a target tube top circle and boundaries constrained by two dimensional (2D) projections of different types of tube top circle centers. The method also includes calculating an edge gradient magnitude map of the ROI patch and constructing a three dimensional (3D) map of a circle parameter space, each location in the 3D map corresponding to a circle parameter having a center location and a diameter. The method further includes accumulating weighted votes in the 3D map from edge points in the edge gradient magnitude map along edge point gradient directions, determining locations in the 3D map as circle candidates based on the accumulated votes and selecting a target tube top circle based on the greatest accumulated votes.

When there are a plurality of detection candidate objects, to detect a target object that is appropriately visible as a whole, an information processing apparatus calculates detection likelihoods in a plurality of local areas of each of the plurality of detection target candidates, and a detection reliability of each of the detection target candidates based on a distribution of the detection likelihoods.