Abstract: A computer implemented method of training an encoder to extract features from sensor data comprises generating a plurality of training examples, each training example comprising at least two data representations of a set of sensor data, the at least two data representations related by a transformation parameterized by at least one numerical transformation value; and training the encoder based on a self-supervised regression loss function applied to the training examples. The encoder extracts respective features from the at least two data representations of each training example, and at least one numerical output value is computed from the extracted features. The self-supervised regression loss function encourages the at least one numerical output value to match the at least one numerical transformation value parameterizing the transformation.

Abstract: A computer implemented method of training an encoder to extract features from sensor data comprises training a machine learning (ML) system based on a self-supervised loss function applied to a training set, the ML system comprising the encoder. The training set comprises sets of real sensor data and corresponding sets of synthetic sensor data. The encoder extracts features from each set of real and synthetic sensor data, and the self-supervised loss function encourages the ML system to associate each set of real sensor data with its corresponding set of synthetic sensor data based on their respective features.

Abstract: A computer-implemented method of computer-implemented method of perceiving structure in a point cloud comprises: applying clustering to the point cloud, and thereby identifying at least one moving object cluster within the point cloud, the point cloud comprising time-stamped points captured over a non-zero accumulation window; determining a motion model for the moving object cluster, by fitting one or more parameters of the motion model to the time-stamped points of that cluster; using the motion model to transform the time-stamped points of the moving object cluster to a common reference time; and applying a perception component to the transformed points of the moving object cluster to extract information about structure exhibited in the transformed points.

Abstract: A computer-implemented method of processing images for extracting information about known objects comprises the steps of receiving an image containing a view of a known object at a scale dependent on an object distance of the known object from an image capture location of the image; determining, from a world model representing one or more known objects in the vicinity of the image capture location, an object location of the known object, the object location and the image capture location defined in a world frame of reference; and based on the image capture location and the object location in the world frame of reference, applying image scaling to the image, to extract a rescaled image containing a rescaled view of the known object at a scale that is substantially independent of the object distance from the image capture location.

Abstract: In one aspect, hierarchical image segmentation is applied to an image formed of a plurality of pixels, by classifying the pixels according to a hierarchical classification scheme, in which at least some of those pixels are classified by a parent level classifier in relation to a set of parent classes, each of which is associated with a subset of child classes, and each of those pixels is also classified by at least one child level classifier in relation to one of the subsets of child classes, wherein each of the parent classes corresponds to a category of visible structure, and each of the subset of child classes associated with it corresponds to a different type of visible structure within that category.

Abstract: A method of annotating known objects in road images captured from a sensor-equipped vehicle, the method implemented in an annotation system and comprising: receiving at the annotation system a road image containing a view of a known object; receiving ego localization data, as computed in a map frame of reference, via localization applied to sensor data captured by the sensor-equipped vehicle, the ego localization data indicating an image capture pose of the road image in the map frame of reference; determining, from a predetermined road map, an object location of the known object in the map frame of reference, the predetermined road map representing a road layout the map frame of reference, wherein the known object is one of: a piece of road structure, and an object on or adjacent a road; computing, in an image plane defined by the image capture pose, an object projection, by projecting an object model of the known object from the object location into the image plane; and storing, in an image database, image

Abstract: A computer-implemented method of modelling a perception system, the perception system configured to receive sensor data and interpret the sensor data to generate actual perception outputs, comprises: receiving a plurality of input samples, wherein each input sample comprises sensor data and is associated with one or more training perception ground truths pertaining to one or more ground truth objects; providing the sensor data of each input sample to the perception system to be modelled, wherein the perception system interprets the sensor data, in order to generate one or more actual perception outputs for the input sample; and training a function approximator to model the perception system by: for each input sample, inputting the training perception ground truths to the function approximator, wherein the function approximator computes one or more predicted perception values by processing the training perception ground truths but not the sensor data from which the actual perception outputs are generated, and

Abstract: In one aspect, hierarchical image segmentation is applied to an image formed of a plurality of pixels, by classifying the pixels according to a hierarchical classification scheme, in which at least some of those pixels are classified by a parent level classifier in relation to a set of parent classes, each of which is associated with a subset of child classes, and each of those pixels is also classified by at least one child level classifier in relation to one of the subsets of child classes, wherein each of the parent classes corresponds to a category of visible structure, and each of the subset of child classes associated with it corresponds to a different type of visible structure within that category.

Abstract: Herein, a “perception statistical performance model” (PSPM) for modelling a perception slice of a runtime stack for an autonomous vehicle or other robotic system may be used e.g. for safety/performance testing. A first PSPM is configured to: receive a computed perception ground truth; determine from the perception ground truth, based on a set of learned parameters, a probabilistic perception uncertainty distribution, the parameters learned from a set of actual perception outputs generated using the perception slice to be modelled, in order to compute a first time series of perception outputs. A second time series of perception outputs is computed using a second PSPM for modelling a second perception slice of the runtime stack, the first PSPM learned from data of a first sensor modality of the perception slice and the time series, and the second PSPM learned independently thereof from data of a second sensor modality of the second perception slice and the second time series.

Abstract: A computer-implemented method of creating 2D annotation data for annotating one or more perception inputs comprises: receiving at the annotation computer system at least one captured frame comprising a set of 3D structure points, in which at least a portion of a structure component is captured; computing a reference position for the structure component within the frame; generating a 3D model for the structure component by selectively extracting 3D structure points of the frame based on the reference position; computing a projection of the 3D model into an image plane; and storing 2D annotation data of the computed projection in persistent computer storage for annotating the structure component within the image plane.

Abstract: Techniques are described for determining the value of individual facts in a knowledge base, and various applications of such fact values. In one example, the knowledge base is part of a question answering system. A ranking of knowledge base facts based on the number of times each of the knowledge base facts is used in answering user questions (e.g., as determined from question answering logs) is used to derive a fact value function that may then be used to determine the value of other facts included in or subsequently added to the knowledge base.

Abstract: In one aspect, hierarchical image segmentation is applied to an image formed of a plurality of pixels, by classifying the pixels according to a hierarchical classification scheme, in which at least some of those pixels are classified by a parent level classifier in relation to a set of parent classes, each of which is associated with a subset of child classes, and each of those pixels is also classified by at least one child level classifier in relation to one of the subsets of child classes, wherein each of the parent classes corresponds to a category of visible structure, and each of the subset of child classes associated with it corresponds to a different type of visible structure within that category.

Abstract: A perception model is trained to classify inputs in relation to a discrete set of leaf node classes. A hierarchical classification tree encodes hierarchical relationships between the leaf node classes. A training loss function is dependent on a classification score for a given training input a its ground truth leaf node class of the training input, but also classification scores for at least some others of the leaf node classes, with the classification scores of the other leaf node classes weighted in dependence on their hierarchical relationship to the ground truth leaf node class within the hierarchical classification tree.

Abstract: A query parsing system uses a multi-stage process to parse the text of incoming queries before attempting to answer the queries. The multi-stage configuration involves a first trained classifier to determine the query type, or intent, of the text and a plurality of second trained classifiers, where each of the second trained classifiers is configured particularly for one specific respective query type. During query processing, the first trained classifier is used on the text to identify the query type. A second trained classifier for that specific identified query type is then found and used on the text to identify what strings in the text correspond to specific entities needed to resolve the query. The identified text strings and query type are then placed into a form understandable by a knowledge base and sent to the knowledge base for resolution. The classifiers may be trained using queries and answers previously processed by the knowledge base using a rules/template resolution process.

Abstract: Techniques are described for adding knowledge to a knowledge base.