-
公开(公告)号：US20230042275A1
公开(公告)日：2023-02-09
申请号：US17966396
申请日：2022-10-14
Applicant: Socionext Inc.
Inventor： Yukihiro SASAGAWA
Abstract: A network quantization method is a network quantization method of quantizing a neural network, and includes a database construction step of constructing a statistical information database on tensors that are handled by neural network, a parameter generation step of generating quantized parameter sets by quantizing values included in each tensor in accordance with the statistical information database and the neural network, and a network construction step of constructing a quantized network by quantizing the neural network with use of the quantized parameter sets. The parameter generation step includes a quantization-type determination step of determining a quantization type for each of a plurality of layers that make up the neural network.
-
2.发明申请公开(公告)号：US20210073635A1
公开(公告)日：2021-03-11
申请号：US17014699
申请日：2020-09-08
Applicant: SOCIONEXT INC.
Inventor： Yukihiro SASAGAWA
Abstract: A quantization parameter optimization method includes: determining a cost function in which a regularization term is added to an error function, the regularization term being a function of a quantization error that is an error between a weight parameter of a neural network and a quantization parameter that is a quantized weight parameter; updating the quantization parameter by use of the cost function; and determining, as an optimized quantization parameter of a quantization neural network, the quantization parameter with which a function value derived from the cost function satisfies a predetermined condition, the optimized quantization parameter being obtained as a result of repeating the updating, the quantization neural network being the neural network, the weight parameter of which has been quantized, wherein the function value derived from the regularization term and an inference accuracy of the quantization neural network are negatively correlated.
-
公开(公告)号：US20230335016A1
公开(公告)日：2023-10-19
申请号：US18341466
申请日：2023-06-26
Applicant: Socionext Inc.
Inventor： Tetsuya TANAKA , Yukihiro SASAGAWA
CPC classification number: G09B29/106 , G06T7/70
Abstract: A map information update method includes: obtaining one or more projection relationships; obtaining, for each projection relationship, reprojection error information; calculating, for each of one or more landmarks, a first sum value based on all items of reprojection error information associated with the landmark; calculating, for each of one or more keyframes, a second sum value based on all items of reprojection error information associated with the keyframe; inferring from the first sum value, for each landmark, a position information update value of an item of position information about the landmark, and updating the item of position information about the landmark using the position information update value; and inferring from the second sum value, for each keyframe, a pose information update value of an item of pose information about the keyframe, and updating the item of pose information about the keyframe using the pose information update value.
-
公开(公告)号：US20220398693A1
公开(公告)日：2022-12-15
申请号：US17891687
申请日：2022-08-19
Applicant: Socionext Inc.
Inventor： Yukihiro SASAGAWA
IPC: G06T5/00
Abstract: A color image inpainting method includes: obtaining a color image of an object to be recognized, the color image including a missing portion where at least part of image information is missing; obtaining an infrared image of the object; identifying the missing portion in the color image; and inpainting the missing portion in the color image identified in the identifying. The inpainting includes inpainting the missing portion by using information which is obtained from the infrared image and corresponds to the missing portion to obtain an inpainted color image of the object.
-
公开(公告)号：US20220036160A1
公开(公告)日：2022-02-03
申请号：US17506303
申请日：2021-10-20
Applicant: SOCIONEXT INC.
Inventor： Yukihiro SASAGAWA
IPC: G06N3/04
Abstract: An inference model generating method is a method for generating a third inference model using a trained first inference model and a trained second inference model, when a type of output data that is output from the first inference model is the same as a type of input data that is input to the second inference model, the method including: preparing a first partial inference model that includes a portion of the first inference model from an input layer through a predetermined intermediate layer; preparing a second partial inference model that includes a portion of the second inference model from a predetermined intermediate layer to an output layer; and generating the third inference model by disposing a glue layer between the first partial inference model and the second partial inference model.
-
Patent Agency Ranking
公开(公告)号：US20250029321A1
公开(公告)日：2025-01-23
申请号：US18903684
申请日：2024-10-01
Applicant: Socionext Inc.
Inventor： Yukihiro SASAGAWA
IPC: G06T15/20 , G06V10/771 , G06V10/82 , H04N13/282
Abstract: A computer-implemented image processing method of synthesizing a free viewpoint image on a display projection surface from plurality of captured images, the method including: acquiring the plurality of captured images with a plurality of respective cameras; estimating projection surface residual data by machine learning using the plurality of captured images and viewpoint data as inputs, the projection surface residual data representing a difference between a bowl-shaped predefined projection surface and the display projection surface; and acquiring the free viewpoint image by mapping the plurality of captured images onto the display projection surface using information about the predefined projection surface, the projection surface residual data, and the viewpoint data.
-
公开(公告)号：US20210209470A1
公开(公告)日：2021-07-08
申请号：US17210097
申请日：2021-03-23
Applicant: SOCIONEXT INC.
Inventor： Yukihiro SASAGAWA
Abstract: A network quantization method of quantizing a neural network includes: constructing a statistical information database of tensors handled by the neural network obtained when a plurality of test datasets are input to the neural network; generating a quantization parameter set by quantizing values of the tensors; and quantizing the neural network using the quantization parameter set. In the generating, based on the statistical information database, a quantization step interval in a high-frequency region is set to be narrower than a quantization step interval in a low-frequency region, the high-frequency region including a value, among the values of the tensors, having a frequency that is a maximum, and the low-frequency region including a value of the tensors that has a lower frequency than in the high-frequency region and a frequency that is not zero.
-
公开(公告)号：US20160364882A1
公开(公告)日：2016-12-15
申请号：US15247337
申请日：2016-08-25
Applicant: SOCIONEXT INC.
Inventor： Yukihiro SASAGAWA , Tatsuya TETSUKAWA , Michael Bi MI , Chua Tien PING , Ryuta NAKANISHI , Naoki NOJIRI
CPC classification number: G06T7/2033 , G06K9/00 , G06K9/00335 , G06K9/00362 , G06K9/00986 , G06K9/4642 , G06K9/4661 , G06K2009/3291 , G06T7/246 , G06T2207/10016 , G06T2207/20081 , G06T2207/30201 , G06T2207/30232 , G06T2207/30268
Abstract: An image recognition system for detecting and tracking at least an image portion associated with a predefined object from a moving picture is configured to be able to perform: an object detection processing step of detecting the object; a tracking point specification processing step of specifying a predetermined point as a tracking point; a tracking target recognition processing step of recognizing an actual tracking target based on the tracking point; a tracking processing step of tracking the tracking target; and a determination processing step of determining the type of the tracking target's behavior. The tracking point specification processing step and the determination processing step are implemented by software, while the object detection processing step, the tracking target recognition processing step, and the tracking processing step are implemented by hardware.
Abstract translation: 用于检测和跟踪与运动图像相关联的预定对象的图像部分的图像识别系统被配置为能够执行：对象检测处理步骤，检测对象; 指定预定点作为跟踪点的跟踪点指定处理步骤; 跟踪目标识别处理步骤，基于跟踪点识别实际跟踪目标; 跟踪跟踪目标的跟踪处理步骤; 以及确定处理步骤，确定跟踪目标的行为的类型。 跟踪点指定处理步骤和确定处理步骤由软件实现，而对象检测处理步骤，跟踪目标识别处理步骤和跟踪处理步骤由硬件实现。
-
公开(公告)号：US20250036951A1
公开(公告)日：2025-01-30
申请号：US18913473
申请日：2024-10-11
Applicant: Socionext Inc.
Inventor： Yukihiro SASAGAWA
IPC: G06N3/08
Abstract: A neural network generation method includes: decomposing a trained teacher neural network including M layers into N subnetworks to generate a trained teacher neural network including N subnetworks; and generating a trained student neural network by (i) inputting a data set into each of the trained teacher neural network and a student neural network including N layers and (ii) training the student neural network. The generating of the trained student neural network includes: associating N teacher outputs and N student outputs in order of processing from an input layer toward an output layer; and determining weight data for each of the N layers in order of association, the N teacher outputs corresponding one to one to the N subnetworks, the N student outputs corresponding one to one to the N layers of the student neural network.
-
10.发明申请公开(公告)号：US20230046001A1
公开(公告)日：2023-02-16
申请号：US17975168
申请日：2022-10-27
Applicant: Socionext Inc.
Inventor： Tetsuya TANAKA , Yukihiro SASAGAWA
Abstract: A map information update method includes: (a) obtaining map information; (b) obtaining landmark observed positions indicating positions of one or more landmarks in a captured image; (c) adding that includes (i) generating added map information by adding information pertaining to the landmark observed positions to the map information, and (ii) updating the map information obtained in (a) to the added map information; (d) predicting that includes (i) calculating predicted map information based on the map information updated in (c), by using a neural network inference engine that has been trained, and (ii) updating the map information to the predicted map information; and updating information that includes (i) calculating updated map information based on the map information updated in (d), by using a gradient method, and (ii) updating the map information to the updated map information.
-
-
-
-
-
-
-
-
-
Search Results
11
records
(took 0.015 seconds)
