公开(公告)号：EP4365775A1

公开(公告)日：2024-05-08

申请号：EP23187701.0

申请日：2023-07-25

申请人： INTEL Corporation

发明人： PALLA, Alessandro ,  BERNARD, David Thomas ,  HANRAHAN, Niall

IPC分类号： G06N3/0464 ,  G06N3/048 ,  G06N3/09

CPC分类号： G06F17/153 ,  G06N3/0464 ,  G06N3/09 ,  G06N3/063 ,  G06N3/048 ,  G06N3/084 ,  G06N3/094 ,  G06N3/0475 ,  G06N3/0442

摘要： A deconvolution can be decomposed into multiple convolutions. Results of the convolutions constitute an output of the deconvolution. Zeros may be added to an input tensor of the deconvolution to generate an upsampled input tensor. Subtensors having the same size as the kernel of the deconvolution may be identified from the upsampled input tensor. A subtensor may include one or more input activations and one or more zeros. Subtensors having same distribution patterns of input activations may be used to generate a reduced kernel. The reduced kernel includes a subset of the kernel. The position of a weight in the reduced kernel may be the same as the positions of an input activation in the subtensor. Multiple reduced kernels may be generated based on multiple subtensors having different distribution patterns of activations. Each of the convolutions may use the input tensor and a different one of the reduced kernels.

公开(公告)号：EP3388981A1

公开(公告)日：2018-10-17

申请号：EP17166619.1

申请日：2017-04-13

申请人： NXP B.V.

发明人： Pineda de Gyvez, Jose de Jesus ,  Fatemi, Hamed ,  Moro Pérez, Gonzalo ,  Corporaal, Hendrik

IPC分类号： G06N3/063 ,  G06F17/15 ,  G06F17/16 ,  G06N3/04

CPC分类号： G06F9/30112 ,  G06F17/153 ,  G06K9/00335 ,  G06K9/6256 ,  G06N3/0454 ,  G06N3/063 ,  G06N3/0635

摘要： A human-machine-interface system comprising: register-file-memory, configured to store input-data; a first-processing-element-slice, a second-processing-element-slice, and a controller. Each of the processing-slices comprise: a register configured to store register-data; and a processing-element configured to apply an arithmetic and logic operation on the register-data in order to provide convolution-output-data. The controller is configured to: load input-data from the register-file-memory into the first-register as the first-register-data; and load: (i) input-data from the register-file-memory, or (ii) the first-register-data from the first-register, into the second-register as the second-register-data.

公开(公告)号：EP3346391A1

公开(公告)日：2018-07-11

申请号：EP17208104.4

申请日：2017-12-18

申请人： INTEL Corporation

发明人： KORTHIKANTI, Vijay Anand R. ,  KALAIAH, Aravind ,  WERNER, Tony L. ,  KLOSS, Carey K. ,  KHOSROWSHAHI, Amir

IPC分类号： G06F17/15

CPC分类号： G06N3/084 ,  G06F17/153 ,  G06F17/16 ,  G06N3/04

摘要： In one embodiment, a matrix operation may be performed using a plurality of input matrices, wherein the matrix operation is associated with one or more convolution operations. The plurality of input matrices may be partitioned into a plurality of input partitions, wherein the plurality of input matrices is partitioned based on a number of available processing elements. The plurality of input partitions may be distributed among a plurality of processing elements, wherein each input partition is distributed to a particular processing element of the plurality of processing elements. A plurality of partial matrix operations may be performed using the plurality of processing elements, and partial matrix data may be transmitted between the plurality of processing elements while performing the plurality of partial matrix operations. A result of the matrix operation may be determined based on the plurality of partial matrix operations.

公开(公告)号：EP3326115A1

公开(公告)日：2018-05-30

申请号：EP16753749.7

申请日：2016-07-23

申请人： Mireplica Technology, LLC

发明人： JOHNSON, William H. ,  NAGATA, Toshio

IPC分类号： G06N3/04 ,  G06K9/00

CPC分类号： G06K9/6229 ,  G06F3/0604 ,  G06F3/0644 ,  G06F3/0683 ,  G06F9/3867 ,  G06F15/80 ,  G06F17/153 ,  G06K9/00973 ,  G06K9/00986 ,  G06K9/00993 ,  G06K9/4628 ,  G06K9/6232 ,  G06K9/627 ,  G06N3/0454

摘要： Methods and systems are disclosed for increased speed of processing operations on data in two-dimensional arrays, and for detecting a feature in an image. A method for detecting a feature in an image comprises storing, in a set of data memories within a parallel processing system, first image data representing pixels of a first image. The method further comprises storing, in a memory of a host processor coupled to the parallel processing system, feature kernel data representing a set of weight matrices. A method for increased speed of processing operations on data in two-dimensional arrays comprises storing, in a set of data memories within a parallel processing system, first array data representing elements of a first array. The method further comprises, for each of multiple selected elements within the first array, performing a processing operation on the selected element to produce an output element corresponding to the selected element.

公开(公告)号：EP1229135B1

公开(公告)日：2014-01-22

申请号：EP02250686.9

申请日：2002-01-31

申请人： Perlin, Mark W.

发明人： Perlin, Mark W.

IPC分类号： C12Q1/68 ,  G06F19/00 ,  G06F17/00

CPC分类号： C12Q1/68 ,  C12Q1/6851 ,  G06F17/11 ,  G06F17/153 ,  G06F17/16 ,  G06F17/30312 ,  G06F19/18 ,  G06F19/22 ,  G06F19/24 ,  H04L9/0816 ,  H04L2209/24

摘要： The present invention pertains to a process for automatically analyzing mixed DNA samples. Specifically, the process comprises the steps of obtaining a mixed DNA sample; amplifying the DNA sample to produce a product; detecting the product to produce a signal; and analyzing the signal to determine information about the composition of the mixed DNA sample. This DNA mixture analysis is useful for finding criminals and convicting them. This mixture analysis provides high quality estimates, and can determine genotypes, mixture weights, and likelihood ratios. This analysis provides confidence measures in the results it computes, and generates reports and intuitive visualizations. The process automates a tedious manual procedure, thereby reducing the cost, time, and effort involved in DNA forensic analysis. The system can greatly accelerate the rate of DNA crime analysis, and be used to exonerate innocent people.

公开(公告)号：EP1783664A3

公开(公告)日：2011-10-05

申请号：EP06022645.3

申请日：2006-10-30

申请人： OMRON CORPORATION

发明人： Yoshiura, Takeshi Omron Corporation ,  Kato, Yutaka Omron Corporation

IPC分类号： G06K9/64 ,  G06T7/00

CPC分类号： G06K9/4652 ,  G06F17/153 ,  G06K9/6202 ,  G06T7/001 ,  G06T2207/10024

摘要： A determination target region includes a combination of a red gray image, a green gray image, and a blue gray image. A CPU part transforms a three gray value matrixes to a single first data array uniquely associated with a pixel position in the determination target region. The CPU part calculates a normalized correlation value with a single second data array transformed from a model image according to the same predetermined rule, and determines whether or not a match is found with the model image according to whether or not the normalized correlation value exceeds a predetermined threshold value.

公开(公告)号：EP1589766A3

公开(公告)日：2008-07-09

申请号：EP05013280.2

申请日：1996-10-04

申请人： MICROSOFT CORPORATION

发明人： Lee, Ming-Chieh ,  Powell, III, William Chambers

IPC分类号： H04N7/30 ,  H04N7/26

CPC分类号： H04N19/00 ,  G06F17/153 ,  G06K9/6203 ,  G06T7/223 ,  G06T9/20 ,  G06T2207/10016 ,  H04N19/186 ,  H04N19/20 ,  H04N19/23 ,  H04N19/51 ,  H04N19/517 ,  H04N19/537 ,  H04N19/54 ,  H04N19/543 ,  H04N19/563 ,  H04N19/61 ,  H04N19/63 ,  H04N19/649

摘要： A computer readable medium stores computer executable instructions for causing a computer programmed thereby to perform a method of extrapolating values of pixels of a video object (402), so as to define values for at least one pixel (412) outside of the perimeter (408) of the video object (402), the at least one pixel (412) outside of the perimeter (408) being within a block boundary (406) around the video object (402). The method comprises: scanning a line of pixels within the block boundary (406), wherein the scanning identifies at least one pixel (412) outside of the perimeter (408), and wherein each identified pixel (412) is part of a segment with two end pixels, at least one of the two end pixels having a perimeter (408) pixel value; for each identified pixel (412), if both end pixels of the segment including the identified pixel (412) have perimeter (408) pixel values; assigning to the identified pixel (412) an average of the perimeter (408) pixel values, and, if only one end pixel of the segment including the identified pixel (412) has a perimeter (408) pixel value, assigning to the identified pixel (412) the perimeter (408) pixel value.

摘要翻译： 计算机可读介质存储计算机可执行指令，用于使由其编程的计算机执行外推视频对象（402）的像素值的方法，以便定义在周边（408）之外的至少一个像素（412）的值 ），所述周边（408）外部的所述至少一个像素（412）在所述视频对象（402）周围的块边界（406）内。 该方法包括：扫描块边界（406）内的一行像素，其中扫描识别外围（408）外的至少一个像素（412），并且其中每个识别的像素（412）是具有 两个端部像素，所述两个端部像素中的至少一个具有周边（408）像素值; 对于每个识别的像素（412），如果包括所识别的像素（412）的段的两个端部像素具有周边（408）像素值，则向所识别的像素（412）分配平均的周边（408）像素值，以及 如果包括识别的像素（412）的片段的仅一个端部像素具有周边（408）像素值，则向识别的像素（412）分配周边（408）像素值。

公开(公告)号：EP0762295A3

公开(公告)日：1998-06-17

申请号：EP96305723

申请日：1996-08-02

申请人： GEN INSTRUMENT CORP

发明人： BUTLER DONALD S ,  AMANO RICHARD A

IPC分类号： G06F17/15 ,  G06T5/20 ,  G09G5/00 ,  H04N5/14 ,  H04N5/208 ,  H04N5/21 ,  G06F17/10

CPC分类号： G06T5/20 ,  G06F17/153 ,  H04N5/14 ,  H04N5/21

摘要： A two dimensional video convolver (10) generates visually acceptable images on a standard television receiver and includes a first adder stage (12, 14), a second adder stage (16, 18, 20), and two pixel time period delays (22, 24). The first adder stage simultaneously receives binary data from three adjacent pixels in a predetermined line in a first direction of a picture to be reproduced on a display during each pixel scan period. The binary data of each of three pixels is processed with a predetermined weight to generate a binary first stage output signal representative of a weighted sum of the plurality of the three pixels in the predetermined line. The second adder stage is responsive to the binary first adder stage output signal during a time period for each of three predetermined adjacent lines in a second orthogonal direction of a picture to be reproduced on a display. The second adder stage processes the binary weighted sum of each of the three lines with a predetermined weight to generate a video convolver output signal representative of a weighted sum of the three pixels in a first direction over three adjacent lines in a second orthogonal direction. Weighting is accomplished in the first and second adder stages by using binary data bits for a pixel as received when the weight for that pixel is a numerical one, and by shifting the received binary data bits by a predetermined number of y bits in a first direction to provide a multiply function when a respective weight is greater than one and equals 2
y .

公开(公告)号：EP0648359A1

公开(公告)日：1995-04-19

申请号：EP93914882.0

申请日：1993-07-05

申请人： SNELL & WILCOX LIMITED

发明人： BORER, Timothy John

IPC分类号： H04N7 ,  G06F17 ,  G06T7

CPC分类号： H04N19/527 ,  G06F17/153 ,  G06T7/262

摘要： A method of correlation signal processing particularly for the processing of correlation surfaces which are utilised in estimating the motion within sequences of video pictures, involves matching an assumed quadratic surface in the region of a stationary value.

摘要翻译： 相关信号处理的一种方法，特别是对用于估计视频图像序列内的运动的相关表面的处理，涉及匹配静止值区域内假定的二次表面。

公开(公告)号：EP0503057A1

公开(公告)日：1992-09-16

申请号：EP91919364.0

申请日：1991-09-25

申请人： EASTMAN KODAK COMPANY

发明人： GERSTENBERGER, Jeffrey, Scott

IPC分类号： G01B11 ,  G01C11 ,  G06F17 ,  G06T1 ,  G06T7 ,  H04N13

CPC分类号： G06F17/153 ,  G06T7/593 ,  G06T2207/10012 ,  H04N2013/0081

摘要： Mécanisme servant à déterminer la parallaxe entre une première et une deuxième image numérique et employant un procédé itératif d'exploration et de corrélation de configuration d'image. Ledit procédé permet de déterminer rapidement la parallaxe pour chaque pixel qui est commun aux images captées par paire d'images stéréo. En utilisant tout d'abord une sous-configuration dispersée de points pixels de référence distribués sur l'ensemble d'une première configuration d'image, on peut obtenir rapidement une sous-configuration similairement dispersée de points de raccordement, qui seront utilisés comme points de départ pour des étapes d'exploration et de corrélation plus fournies, en vue de la deuxième configuration d'image. Les valeurs de parallaxe résultantes peuvent ensuite être soumises à une fonction d'interpolation de précision des sous-pixels. Une fois que la parallaxe entre les configurations de la première et de la deuxième image a été établie, les sites peuvent être calculés en utilisant un mécanisme dépendant de la parallaxe classique.

摘要翻译： 用于确定第一和第二数字图像之间的视差的机构采用迭代图像模式搜索和相关处理，其能够快速地确定由立体图像对捕获的图像共同的每个像素的视差。 通过初始使用分布在整个第一图像阵列中的参考点像素的分散子阵列，将被用作随后更密集搜索和相关步骤的起始点的类似分散的连接点子阵列可以是 快速获得第二个图像阵列。 然后，导出的视差值可以受到子像素精度的内插函数。 一旦导出了第一和第二图像阵列之间的视差，就可以使用常规的视差相关机制来计算高程。