Qualcomm Incorporated

Aptina Imaging Corporation

Innopeak Technology, Inc.

Shuxue Quan

Xiaoyun Jiang

Chinchuan Andrew Chiu

Ning Bi

Lei Xu

Junchen Du

Bo Zhou

Yunke Pan

Haohong Wang

Kalin Mitkov Atanassov

Khaled Helmi El-Maleh

Ananthapadmanabhan Arasanipalai Kandhadai

Yingyong Qi

Jingqiang Li

Ying Noyes

Jianfeng Ren

Feng Guo

Ruiduo Yang

Qinwen Xu

Binglong Xie

Szepo Robert Hung

Lei Ma

Ruben Manuel Velarde

Dashan Gao

Yingen Xiong

Xin Zhong

Yi Xu

Bolan Jiang

Magdi Mohamed

Xin Li

Xinping Zhang

Michael Warren Castelloe

Andrew Chinchuan Chiu

Ashwath Harthattu

Shili Xu

Yuliang Guo

Xiangyu Du

Zhong Li

Ching-Hui Chen

A method for identifying human poses in an image, a computer system, and a non-transitory computer-readable medium are provided. The method includes obtaining the image, encoding the image into a feature map, converting the feature map into a plurality of TPDF maps corresponding to a plurality of predefined part types, and generating a global pose map based on the feature map and the plurality of TPDF maps corresponding to the predefined part types. Each TPDF map corresponds to a respective predefined part type and represents a plurality of vectors connecting each position on the image to a closest body part among the one or more body parts of the respective predefined part type. A global pose map is generated based on the feature map and the plurality of TPDF maps, identifying one or more human bodies having a plurality of body parts in a scene of the image.

Method for identifying human poses in an image, computer system, and non-transitory computer-readable medium

A method performed by an electronic device is described. The method includes receiving a set of images. The method also includes determining a motion region and a static region based on the set of images. The method further includes extracting, at a first rate, first features from the motion region. The method additionally includes extracting, at a second rate that is different from the first rate, second features from the static region. The method also includes caching the second features. The method further includes detecting at least one object based on at least a portion of the first features.

Systems and methods for object detection

Provided are systems, methods, and computer-readable medium for operating a neural network. In various implementations, the neural network can receive an input image that includes an object to be identified. The neural network can generate a plurality of initial feature maps using a convolution layers, wherein a first initial feature maps is generated using the input image. The neural network can generate an up-sampled feature map using a de-convolution layer that takes an initial feature map as input, where the up-sampled feature map has a same resolution as the previous initial feature map. The neural network can combine the up-sampled feature map and the previous initial feature map, and use the combined feature map to more accurate identify the object.

Combining convolution and deconvolution for object detection

A system and method of object detection are disclosed. In a particular implementation, a method of processing an image includes receiving, at a processor, image data associated with an image of a scene. The scene includes a road region. The method further includes detecting the road region based on the image data and determining a subset of the image data. The subset excludes at least a portion of the image data corresponding to the road region. The method further includes performing an object detection operation on the subset of the image data to detect an object. The object detection operation performed on the subset of the image data is exclusive of the at least a portion of the image data corresponding to the road region.

System and method of object detection

A method performed by an electronic device is described. The method includes interleaving multiple input image channels to produce an interleaved multi-channel input. The method also includes loading the interleaved multi-channel input to a single-instruction multiple data (SIMD) processor. The method further includes convolving the interleaved multi-channel input with a multi-channel filter.

Systems and methods for image processing in a deep convolution network

A method performed by an electronic device is described. The method includes determining overlapping areas from neighboring images. The method also includes determining a difference measure between the overlapping areas. The method further includes determining a constraint measure corresponding to at least one of the overlapping areas. The method additionally includes determining a seam based on a combination of the difference measure and the constraint measure.

Systems and methods for determining a seam

A method for calculating a feature descriptor on a single instruction, multiple data (SIMD) processor is described. The method includes generating histogram bin indexes in a first register. The method also includes generating weights in a second register. The method further includes updating an entire histogram table in a register file based on the histogram bin indexes and the weights without storing any histogram bin to memory. Histogram bins are updated in parallel with a single instruction.

Systems and methods for calculating a feature descriptor

The disclosed technology relates to image-capturing methods. In one aspect, a method includes receiving an image frame comprising a plurality of pixels and subtracting foreground pixels from the image frame to obtain background pixels. The method additionally includes determining an exposure condition for a next image frame based on at least a subset of the background pixels. The method further includes adjusting the foreground pixels such that a difference between a background luma value and a foreground luma value of the next image frame is within a predetermined range. Aspects are also directed to apparatuses configured for the methods.

Exposure metering based on background pixels

A method includes accessing, at a computing device, data descriptive of a graph representing a program. The graph includes multiple nodes representing execution steps of the program and includes multiple edges representing data transfer steps. The method also includes determining at least two heterogeneous hardware resources of the computing device that are available to execute code represented by one or more of the nodes, and determining one or more paths from a source node to a sink node based on a topology of the graph. The method further includes scheduling execution of code at the at least two heterogeneous hardware resources. The code is represented by at least one of the multiple nodes, and the execution of the code is scheduled based on the one or more paths.

Efficient execution of graph-based programs

A particular method includes generating a first result of a first integration operation performed on a first subset of elements of the plurality of elements. The first integration operation is associated with a first level of integration. The method includes generating a second result of a second integration operation performed on the first subset of elements. The second integration operation is associated with a second level of integration. The method further includes performing a third integration operation on a second subset of elements of the plurality of elements. The third integration operation is associated with the second level of integration. The third integration operation is performed based on the first result and the second result.

System and method of performing multi-level integration

A method for memory utilization by an electronic device is described. The method includes transferring a first portion of a first decision tree and a second portion of a second decision tree from a first memory to a cache memory. The first portion and second portion of each decision tree are stored contiguously in the first memory. The first decision tree and second decision tree are each associated with a different feature of an object detection algorithm. The method also includes reducing cache misses by traversing the first portion of the first decision tree and the second portion of the second decision tree in the cache memory based on an order of execution of the object detection algorithm.

Systems and methods for memory utilization for object detection

A method includes capturing an image of a scene that includes a diagram. The method further includes applying functional block recognition rules to image data of the image to recognize functional blocks of the diagram. The functional blocks include at least a first functional block associated with a first computer operation. The method further includes determining whether the functional blocks comply with functional block syntax rules. A functional graph is computer-generated based on the functional blocks complying with the functional block syntax rules. The functional graph corresponds to the diagram, and the functional graph includes the functional blocks.

Automated graph-based programming

Embodiments include a method of image processing including decomposing a reflectance spectrum for a test surface into a linear combination of reflectance spectra of a set of test targets. The coefficient vector calculated in this decomposition is used to predict a response of an imaging sensor to the test surface. A plurality of such predicted responses may be used for various applications involving color detection and/or classification, including human skin tone detection.

Systems, methods, and apparatus for image processing, for color classification, and for skin color detection

Systems, methods and apparatus for image processing by classifying pixels are described. In some systems, an exposure control operation is performed according to the pixel classifications. In some cases, the pixels are classified according to a predetermined segmentation of a color space, based on predicted sensor responses.

Systems, methods, and apparatus for exposure control

Techniques are described for automatic print image matching (PIM) parameter extraction. An original image is captured and PIM parameter data is extracted automatically based on specifics of the original image. At least one automated PIM parameter is calculated automatically from the PIM parameter data. At least one automated PIM parameter is inserted in PIM header information for communication to a rendering device to modify the original image when rendered.

Growing community of inventors

San Diego, CA, United States of America

Shuxue Quan