Abstract:
Certain aspects of the present disclosure are directed to methods and apparatus for deep learning in an artificial neural network. One example method generally includes receiving input data at an input to a layer of the neural network; replicating a group of neural processing units in the layer to form a superset of neural processing units, the superset comprising n instances of the group of neural processing units; processing the input data using the superset to generate output data for the layer; and determining an uncertainty of the output data. Processing the input data includes performing a dropout function by zeroing out one or more weights of a set of weights for each of the n instances of the superset of neural processing units and convolving, for each of the n instances in parallel, the input data with one or more non-zeroed out weights of the set of weights.
Abstract:
A method, which is performed by an electronic device, of automatically activating a flash for an image sensor of the electronic device is disclosed. The method may include receiving a first image including at least one text region and determining feature data characterizing the at least one text region in the first image. The method may also identify at least one candidate specular reflection region in the first image. Based on the feature data and the at least one candidate specular reflection region, the flash may be activated. Upon activating the flash, a second image including the at least one text region may be captured.
Abstract:
A method for controlling an application in a mobile device is disclosed. The method includes receiving environmental information, inferring an environmental context from the environmental information, and controlling activation of the application based on a set of reference models associated with the inferred environmental context. In addition, the method may include receiving a sound input, extracting a sound feature from the sound input, transmitting the sound feature to a server configured to group a plurality of mobile devices into at least one similar context group, and receiving, from the server, information on a leader device or a non-leader device and the at least one similar context group.
Abstract:
A method performed by an apparatus is described. The method includes receiving remote data. The method also includes adapting model processing based on the remote data. The method further includes determining a driving decision based on the adapted model processing. In some examples, adapting model processing may include selecting a model, adjusting model scheduling, and/or adjusting a frame computation frequency.
Abstract:
A method, which is performed by an electronic device, of automatically activating a flash for an image sensor of the electronic device is disclosed. The method may include receiving a first image including at least one text region and determining feature data characterizing the at least one text region in the first image. The method may also identify at least one candidate specular reflection region in the first image. Based on the feature data and the at least one candidate specular reflection region, the flash may be activated. Upon activating the flash, a second image including the at least one text region may be captured.
Abstract:
A method for grouping data items in a mobile device is disclosed. In this method, a plurality of data items and a sound tag associated with each of the plurality of data items are stored, and the sound tag includes a sound feature extracted from an input sound indicative of an environmental context for the data item. Further, the method may include generating a new data item, receiving an environmental sound, generating a sound tag associated with the new data item by extracting a sound feature from the environmental sound, and grouping the new data item with at least one of the plurality of data items based on the sound tags associated with the new data item and the plurality of data items.
Abstract:
Disclosed are techniques for estimating a 3D bounding box (3DBB) from a 2D bounding box (2DBB). Conventional techniques to estimate 3DBB from 2DBB rely upon classifying target vehicles within the 2DBB. When the target vehicle is misclassified, the projected bounding box from the estimated 3DBB is inaccurate. To address such issues, it is proposed to estimate the 3DBB without relying upon classifying the target vehicle.
Abstract:
A method performed by an apparatus is described. The method includes receiving remote data. The method also includes adapting model processing based on the remote data. The method further includes determining a driving decision based on the adapted model processing. In some examples, adapting model processing may include selecting a model, adjusting model scheduling, and/or adjusting a frame computation frequency.
Abstract:
A method includes receiving an alarm sound including information related to an emergency event. The method also includes transmitting, to a server, identification information of the mobile device and the information. The method further includes receiving, from the server, an instruction for responding to the emergency event. The method further includes outputting the instruction.
Abstract:
Disclosed are techniques for determining a motion state of a target object. In an aspect, an on-board computer of an ego vehicle detects the target object in one or more images, determines one or more first attributes of the target object based on measurements of the one or more images, determines one or more second attributes of the target object based on measurements of a map of a roadway on which the target object is travelling, and determines the motion state of the target object based on the one or more first attributes and the one or more second attributes of the target object.