Disclosed is a system and associated methods for generating and rendering a polyhedral point cloud that represents a scene with multi-faceted primitives. Each multi-faceted primitive stores multiple sets of values that represent different non-positional characteristics that are associated with a particular point in the scene from different angles. For instance, the system generates a multi-faceted primitive for a particular point of the scene that is captured in first capture from a first position and a second capture from a different second position. Generating the multi-faceted primitive includes defining a first facet with a first surface normal oriented towards the first position and first non-positional values based on descriptive characteristics of the particular point in the first capture, and defining a second facet with a second surface normal orientated towards the second position and second non-positional values based on different descriptive characteristics of the particular point in the second capture.

A sensor data processing system and method is described. Contemplated systems and methods derive a first recognition trait of an object from a first data set that represents the object in a first environmental state. A second recognition trait of the object is then derived from a second data set that represents the object in a second environmental state. The sensor data processing systems and methods then identifies a mapping of elements of the first and second recognition traits in a new representation space. The mapping of elements satisfies a variance criterion for corresponding elements, which allows the mapping to be used for object recognition. The sensor data processing systems and methods described herein provide new object recognition techniques that are computationally efficient and can be performed in real-time by the mobile phone technology that is currently available.

An object is to provide an information processing apparatus capable of reducing redundant computation in CNN. An information processing apparatus (1) according to the present disclosure includes a masking operator (2) configured to use mask channel in input feature maps to mask pixels of feature channels in the input feature maps and to generate masked feature channels, and a convolution operator (3) configured to perform a convolution operation between the masked feature channels and convolution kernel to generate output feature maps.

An object is to provide an information processing apparatus capable of reducing redundant computation in CNN. An information processing apparatus (1) according to the present disclosure includes a masking operator (2) configured to use mask channel in input feature maps to mask pixels of feature channels in the input feature maps and to generate masked feature channels, and a convolution operator (3) configured to perform a convolution operation between the masked feature channels and convolution kernel to generate output feature maps.

A system and method for providing object-level driver attention reasoning with a graph convolution network that include receiving image data associated with a plurality of image clips of a surrounding environment of a vehicle and determining anchor objectness scores and anchor importance scores associated with relevant objects included within the plurality of image clips. The system and method also include analyzing the anchor objectness scores and anchor importance scores associated with relevant objects and determining top relevant objects with respect to an operation of the vehicle. The system and method further include passing object node features and edges of an interaction graph through the graph convolution network to update features of each object node through interaction with other object nodes and determining importance scores for the top relevant objects.

When it is assumed that a large-scale Deep Neural Network for autonomous driving applied compression, there are problems of a decrease in recognition accuracy of a post-compression Neural Network (NN) model and an increase in a compression design period, due to a large number of harmful or unnecessary training images (invalid training images). A training image selection unit B100 calculates an influence value on an inference, and generates an indexed training image set 1004-1 necessary for an NN compression design, by using the influence value. A neural network compression unit P200 notified of the result via a memory P300 compresses the NN.

One or more embodiments of the present disclosure include systems and methods that use neural architecture fusion to learn how to combine multiple separate pre-trained networks by fusing their architectures into a single network for better computational efficiency and higher accuracy. For example, a computer implemented method of the disclosure includes obtaining multiple trained networks. Each of the trained networks may be associated with a respective task and has a respective architecture. The method further includes generating a directed acyclic graph that represents at least a partial union of the architectures of the trained networks. The method additionally includes defining a joint objective for the directed acyclic graph that combines a performance term and a distillation term. The method also includes optimizing the joint objective over the directed acyclic graph.

A neural network is initially trained to remove errors and is later fine tuned to remove less-effective portions (e.g., kernels) from the initially trained network and replace them with further trained portions (e.g., kernels) trained with data after the initial training.

A neural network is initially trained to remove errors and is later fine tuned to remove less-effective portions (e.g., kernels) from the initially trained network and replace them with further trained portions (e.g., kernels) trained with data after the initial training.

An image processing method is provided. An image including a target object is obtained. Image segmentation is performed on the image. A mask image of the target object is determined based on the image segmentation performed on the image. A first feature extraction is performed on the image. A first predicted value associated with the target object is determined based on a first feature extraction result of the first feature extraction performed on the image. A second feature extraction is performed on the mask image. A second predicted value associated with the target object is determined based on a second feature extraction result of the second feature extraction performed on the mask image. A target predicted value associated with the target object is determined according to the first predicted value and the second predicted value.