Patents by Inventor GUO-CHIN SUN

GUO-CHIN SUN has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).

  • Publication number: 20220253996
    Abstract: A method, apparatus, and device for labeling images of PCBs includes obtaining an image to be tested; comparing the image to be tested to a reference image to generate an image mask, the image mask includes several connected domains; detecting defects of the image to be tested; when at least one defect detected in the image to be tested, obtaining a coordinate of the at least one defect; based on a central coordinate of the connected domains and the coordinate of the at least one defect, determining the connected domains to be defect connected domains or normal connected domains; generating a first image mask and a second image mask; and processing the first image mask and the second image mask with the image to be tested to obtain a defect element image corresponding to the defect connected domains and a normal element image corresponding to the normal connected domains.
    Type: Application
    Filed: January 27, 2022
    Publication date: August 11, 2022
    Inventors: GUO-CHIN SUN, CHIN-PIN KUO
  • Publication number: 20220222084
    Abstract: A method for loading multiple neural network model includes compiling at least two neural network models and generating at least two binary model files corresponding to the at least two neural network models. One of the at least two binary model files is an original model file which is taken as the basic model, and differences between the two files are calculated and recorded using preset difference calculation method. A differences file is generated, and the basic model and the differences file are compressed using a preset compression method, to generate an input file. Such input file is input into a neural network accelerator, the input file being decompressed to obtain the basic model and the differences file. The basic model and the differences file are loaded into the neural network accelerator. An electronic device and a non-volatile storage medium performing the above-described method are also disclosed.
    Type: Application
    Filed: January 11, 2022
    Publication date: July 14, 2022
    Inventors: GUO-CHIN SUN, CHIN-PIN KUO
  • Publication number: 20220215247
    Abstract: A method for processing multiple modes of data with cross-learning and sharing to avoid duplicated learning generates a weighting when a neural network model is being trained with a plurality of multiple modes of training samples. The neural network model includes an input layer, a neural network backbone coupled to the input layer, and a plurality of different output layers coupled to the neural network backbone. Results of testing are output by inputting the obtained weighting into the neural network model and testing a multiple modes of test sample with the neural network model. The need for many neural network models is avoided. An electronic device and a non-transitory storage medium are also disclosed.
    Type: Application
    Filed: December 30, 2021
    Publication date: July 7, 2022
    Inventors: GUO-CHIN SUN, CHIN-PIN KUO, TUNG-TSO TSAI
  • Publication number: 20220198678
    Abstract: In a method of distinguishing objects in images, a first image segmentation model is applied to segment a first segmented image including a first object from a test image. A second image segmentation model is applied to segment a second segmented image including a second object from the test image. A third segmented image marking the first object and the second object is obtained according to first coordinates of the first object in the first segmented image and/or second coordinates of the second object in the second segmented image. The method can segment different objects from an image quickly and accurately.
    Type: Application
    Filed: December 21, 2021
    Publication date: June 23, 2022
    Inventors: CHIN-PIN KUO, GUO-CHIN SUN, YUEH CHANG, CHUNG-YU WU
  • Patent number: 11354801
    Abstract: A method for detecting a tumor from images which are required to be shrunken in resolution obtains one or more first images. Then, the method segments or divides the detection images into a number of detection image blocks according to an input size of training data of a convolutional neural network architecture, before segmenting, each of the plurality of detection image blocks comprising coordinate values. The detection image blocks are input into a preset tumor detection model to generate image blocks of a result of the detection images. The method merges the image blocks into a single image according to the coordinate values of each detection image block. Colors of normal areas, abnormal areas, and overlapping areas of the abnormal areas are all different. The method generates a final detection according to color depths in the image. A tumor detection device and a non-transitory storage medium are provided.
    Type: Grant
    Filed: February 12, 2020
    Date of Patent: June 7, 2022
    Assignee: HON HAI PRECISION INDUSTRY CO., LTD.
    Inventors: Tzu-Chen Lin, Chin-Pin Kuo, Tung-Tso Tsai, Guo-Chin Sun, I-Hua Chen, Wan-Jhen Lee
  • Publication number: 20220164978
    Abstract: For the benefit of pedestrians, a method for identifying and locating positions of obstacles moving on a pedestrian sidewalk acquires an image of the sidewalk and processes the image to divide it. The divided image comprises classifications of objects in the image on a pixel by pixel basis. The classifying of objects in the divided image comprises the sidewalk classification, and classification of the obstacles appears in the image. Pixels surrounding the obstacles are acquired in terms of number and classifications. Positions of the obstacles are determined based on a preset threshold, the classifications of adjacent pixels of the obstacles, and the pixel number of the adjacent pixel in each object classification. An apparatus and a system applying the method are also disclosed.
    Type: Application
    Filed: November 11, 2021
    Publication date: May 26, 2022
    Inventors: YUEH CHANG, CHIN-PIN KUO, GUO-CHIN SUN
  • Publication number: 20220058530
    Abstract: A method for optimizing the conversion of a deep learning model to process other data, applied in a device, includes converting a first deep learning model to obtain a second deep learning model, obtaining a weighting arrangement of the two models according to their deep learning frameworks and performing a quantization on the two models. A similarity in weighting between the two models is analyzed to produce a weighting analysis based on the first and second weighting arrangement and the first and second model quantization result weighting. The two models are tested to establish a model performance analysis. One or more suggestions for optimization are obtained based on the weighting analysis and the model performance analysis, and are applied to optimize the second deep learning model, an optimized second deep learning model being employed to process the other data.
    Type: Application
    Filed: August 18, 2021
    Publication date: February 24, 2022
    Inventors: TZU-CHEN LIN, GUO-CHIN SUN, CHIH-TE LU, TUNG-TSO TSAI, JUNG-HAO YANG, CHUNG-YU WU, WAN-JHEN LEE
  • Publication number: 20220044061
    Abstract: A data labeling model training method, an electronic device employing the method, and a storage medium are provided. The method acquires medical image data. An improved quality of the medical image data to be used for training the data labeling model is obtained by filtering the medical data, so as to enable training with higher-quality training material. The data labeling model is used to label medical data with improved efficiency and accuracy.
    Type: Application
    Filed: August 4, 2021
    Publication date: February 10, 2022
    Inventors: Tung-Tso TSAI, Chin-Pin KUO, Wan-Jhen LEE, Guo-Chin SUN
  • Publication number: 20220036192
    Abstract: An image processing method and an electronic device are disclosed, the method acquires training data in response to receiving an image processing instruction and trains a deep learning model with the training data by using a preset deep learning framework to obtain an initial model. A data type of the initial model is converted to increase the data processing speed of the model. A correction layer is added to the converted initial model and the initial model is optimized by training weights of the correction layer to obtain an image processing model to further increase the data processing speed. After acquiring an image from the image processing instruction, the image processing model can be used to process the image and an image processing result can be outputted. The image is processed based on the optimized image processing model, and the data processing speed and accuracy are guaranteed.
    Type: Application
    Filed: July 27, 2021
    Publication date: February 3, 2022
    Inventors: I-HUA CHEN, TUNG-TSO TSAI, GUO-CHIN SUN, TZU-CHEN LIN, WAN-JHEN LEE, CHIN-PIN KUO
  • Publication number: 20220028055
    Abstract: A product defect detection method which includes acquiring a detection image of a product to be detected is provided. The method further includes dividing the detection image into a first preset number of detection blocks. Once a detection result of each detection block is obtained by inputting each detection block into a preset defect recognition model, according to a position of each detection block in the detection image, a detection result of the product is determined according to the detection result of each detection block.
    Type: Application
    Filed: July 22, 2021
    Publication date: January 27, 2022
    Inventors: WAN-JHEN LEE, TUNG-TSO TSAI, CHIN-PIN KUO, TZU-CHEN LIN, GUO-CHIN SUN
  • Publication number: 20210390336
    Abstract: A method for automatically selecting a suitable network model for machine deep learning includes identifying a category of an input content and selecting a training set based on the identified category. Several matched network models are selected based on the identified category. A test network model is selected based on performance data corresponding to the matched network models, and the selected model is trained based on the input content and the selected training set. When an output result in a display interface is not satisfactory, an optimization operation is executed for adjusting the structure of the test network model. An apparatus, an electronic device, and a storage medium applying the method are also disclosed.
    Type: Application
    Filed: June 1, 2021
    Publication date: December 16, 2021
    Inventors: Wan-Jhen Lee, Tung-Tso Tsai, Chin-Pin Kuo, Tzu-Chen Lin, Guo-Chin Sun
  • Publication number: 20210357806
    Abstract: A machine learning model training method includes receiving a problem type to be processed, determining a machine learning model corresponding to the problem type, receiving sample data to train the determined machine learning model, analyzing training results of the determined machine learning model obtained after training is completed and displaying the training results that meet preset conditions, and providing a machine learning model corresponding to the training result that meets the preset conditions.
    Type: Application
    Filed: September 18, 2020
    Publication date: November 18, 2021
    Inventors: GUO-CHIN SUN, TUNG-TSO TSAI, TZU-CHEN LIN, WAN-JHEN LEE, CHIN-PIN KUO
  • Publication number: 20210326641
    Abstract: A method for selecting a deep learning network which is optimal for solving an image processing task obtaining a type of the image processing task, selecting a data set according to the type of problem, and dividing selected data set into training data and test data. Similarities between different training data are calculated, and a batch size of the training data is adjusted according to the similarities of the training data. A plurality of deep learning networks is selected according to the type of problem, and the plurality of deep learning networks is trained through the training data to obtain network models. Each of the network models is tested through the test data, and the optimal deep learning network with the best test result is selected from the plurality of deep learning networks appropriate for image processing.
    Type: Application
    Filed: April 9, 2021
    Publication date: October 21, 2021
    Inventors: TUNG-TSO TSAI, CHIN-PIN KUO, GUO-CHIN SUN, TZU-CHEN LIN, WAN-JHEN LEE
  • Publication number: 20210089886
    Abstract: A method for processing data based on a neural network trained by different methods includes: dividing sample data into a training set and a test set; training a predetermined neural network to obtain a first detection model based on the training set; testing the first detection model based on the test set to count a first precision rate; cleaning the training set and the test set according to selected cleaning method; adjusting the first detection model by a predetermined rule and training adjusted first detection model based on cleaned training set to obtain a second detection model; testing the second detection model based on cleaned test set to count a second precision rate; selecting the first detection model or the second detection model as a final detection model based on a comparison between the first precision rate and the second precision rate.
    Type: Application
    Filed: March 23, 2020
    Publication date: March 25, 2021
    Inventors: TZU-CHEN LIN, TUNG-TSO TSAI, GUO-CHIN SUN, CHIN-PIN KUO, WAN-JHEN LEE
  • Publication number: 20210012485
    Abstract: A method for detecting a tumor from images which are required to be shrunken in resolution obtains one or more first images. Then, the method segments or divides the detection images into a number of detection image blocks according to an input size of training data of a convolutional neural network architecture, before segmenting, each of the plurality of detection image blocks comprising coordinate values. The detection image blocks are input into a preset tumor detection model to generate image blocks of a result of the detection images. The method merges the image blocks into a single image according to the coordinate values of each detection image block. Colors of normal areas, abnormal areas, and overlapping areas of the abnormal areas are all different. The method generates a final detection according to color depths in the image. A tumor detection device and a non-transitory storage medium are provided.
    Type: Application
    Filed: February 12, 2020
    Publication date: January 14, 2021
    Inventors: TZU-CHEN LIN, CHIN-PIN KUO, TUNG-TSO TSAI, GUO-CHIN SUN, I-HUA CHEN, WAN-JHEN LEE
  • Publication number: 20200410356
    Abstract: A method for optimizing a data model is used in a device. The device acquires data information and selecting at least two data models according to the data information, and utilizes the data information to train the at least two data models. The device acquires each accuracy of the at least two data models, determines a target data model which has greatest accuracy between the at least two data models, and optimizes the target data model.
    Type: Application
    Filed: March 6, 2020
    Publication date: December 31, 2020
    Inventors: CHIN-PIN KUO, TUNG-TSO TSAI, GUO-CHIN SUN, TZU-CHEN LIN, WAN-JHEN LEE
  • Publication number: 20200285955
    Abstract: A method for accelerating deep learning includes calling up an entire deep learning architecture. Such architecture includes a data operation program of a convolutional layer and a data operation program of a fully connecting layer. A data operation program of the convolutional layer is obtained, the data operation program of the fully connecting layer is discarded, and the data operation program of the convolutional layer is loaded to a first processor. The data operation program for the fully connecting layer is then applied in similar manner to a second processor of the user terminal, the second processor continuing to perform operations on the fully connecting layer, thereby completing the entire deep learning architecture and training on the user terminal.
    Type: Application
    Filed: July 11, 2019
    Publication date: September 10, 2020
    Inventors: CHIN-PIN KUO, TUNG-TSO TSAI, GUO-CHIN SUN