Google Inc.

X Development LLC

Arthur Robert Pope

Ioannis Stamos

Phillip Ellsworth Stahlfeld

Stephen Charles Hsu

Vivek Verma

Ananya Gupta

Om Prakash Ravi

Xinyue Li

Aviva Cheryl Shwaid

Yanxi Liu

Kshitij Naresh Nikhal

Methods, computer systems, and apparatus, including computer programs encoded on computer storage media, for training a classification neural network. The system generates, from a set of object-specific data, one or more meta-learning datasets for one or more respective initial training tasks. The system determines values for a set of meta parameters by performing meta-learning with a classification neural network on the one or more meta-learning datasets. The system obtains a set of labeled training examples for a characteristic-detection task. The system determines based at least on one of the values for the set of meta parameters and using the set of labeled training examples, target values for the network parameters for the classification neural network to perform the characteristic-detection task.

Meta-learning for detecting object anomaly from images

Computer-implemented methods and systems of detecting curbs include receiving a cloud of three-dimensional (3D) data points acquired along street locations. A plurality of vertical scanlines (e.g., arrays of 3D data points obtained at given horizontal locations) are identified within the cloud of 3D data points. One or more curb points indicating the potential presence of a curb in the plurality of vertical scanlines are identified. A synthesized set of non-overlapping curb curves are generated in order to close gaps between certain curb points while removing certain other outlier curb points. Successive curb curves in the synthesized set of non-overlapping curb curves are then identified as belonging to one or more curb segments.

Systems and methods for detecting curbs in three-dimensional point clouds descriptive of streets

Computer-implemented methods and systems of detecting and modeling curbs include receiving a cloud of three-dimensional (3D) data points acquired along street locations. A plurality of vertical scanlines (e.g., arrays of 3D data points obtained at given horizontal locations) are identified within the cloud of 3D data points. One or more curb points indicating the potential presence of a curb in the plurality of vertical scanlines are identified. A synthesized set of non-overlapping curb curves are generated in order to close gaps between certain curb points while removing certain other outlier curb points. Successive curb curves in the synthesized set of non-overlapping curb curves are then identified as belonging to one or more curb segments. A curb model that fits each of the one or more curb segments to one or more curb model shape representations can then be generated.

Systems and methods for detecting and modeling curb curves in complex urban scenes

Systems and methods for reducing a point cloud data set are provided. According to aspects of the present disclosure, a method includes receiving a point of a point cloud data set, the point having three-dimensional coordinates. The point's coordinates are mapped to a location to determine whether a different point's coordinates have already been mapped to the location. The point is discarded when a different point's coordinates have been mapped to the location.

Systems and methods for reducing a data set

The aspects described herein relate to replacing pixels in images in order to remove obstructions from images. In one example, an image of a scene having symmetrical features and image information identifying at least one pixel of the image corresponding to a hole to be filled may be received. This hole may correspond to an obstruction in the image. A set of symmetry axes may be identified based on the symmetrical features. A symmetry map identifying correspondences between different pixels in the image based on the set of symmetry axes may be generated. A correspondence between the at least one pixel corresponding to a hole to be filled and a second pixel of the image is identified based at least in part on the symmetry map and the image information. The at least one pixel may be altered based on the identified correspondence in order to remove the obstruction.

Growing community of inventors

Palo Alto, CA, United States of America

Arthur Robert Pope