知搜-全球专利检索

  1. Login
  2. Sign Up

Search Results

319 records (took 0.032 seconds)

    System and method for deep machine learning for computer vision applications
    22.
    发明授权
    System and method for deep machine learning for computer vision applications 有权

    公开(公告)号：US11429805B2

    公开(公告)日：2022-08-30

    申请号：US16872199

    申请日：2020-05-11

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Haoyu Ren ,  Mostafa El-Khamy ,  Jungwon Lee ,  Aman Raj

    IPC: G06K9/62 ,  G06N20/00 ,  G06N7/00 ,  G06T7/593 ,  G06T7/11 ,  G06T15/00

    Abstract: A computer vision (CV) training system, includes: a supervised learning system to estimate a supervision output from one or more input images according to a target CV application, and to determine a supervised loss according to the supervision output and a ground-truth of the supervision output; an unsupervised learning system to determine an unsupervised loss according to the supervision output and the one or more input images; a weakly supervised learning system to determine a weakly supervised loss according to the supervision output and a weak label corresponding to the one or more input images; and a joint optimizer to concurrently optimize the supervised loss, the unsupervised loss, and the weakly supervised loss.

    Method and apparatus for neural network quantization
    23.
    发明授权
    Method and apparatus for neural network quantization 有权

    公开(公告)号：US11321609B2

    公开(公告)日：2022-05-03

    申请号：US15433531

    申请日：2017-02-15

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Yoo Jin Choi ,  Mostafa El-Khamy ,  Jungwon Lee

    IPC: G06N3/08 ,  G06N3/063 ,  G06F17/16

    Abstract: Apparatuses and methods of manufacturing same, systems, and methods for performing network parameter quantization in deep neural networks are described. In one aspect, diagonals of a second-order partial derivative matrix (a Hessian matrix) of a loss function of network parameters of a neural network are determined and then used to weight (Hessian-weighting) the network parameters as part of quantizing the network parameters. In another aspect, the neural network is trained using first and second moment estimates of gradients of the network parameters and then the second moment estimates are used to weight the network parameters as part of quantizing the network parameters. In yet another aspect, network parameter quantization is performed by using an entropy-constrained scalar quantization (ECSQ) iterative algorithm. In yet another aspect, network parameter quantization is performed by quantizing the network parameters of all layers of a deep neural network together at once.

    System and method for IQ mismatch calibration and compensation
    26.
    发明授权
    System and method for IQ mismatch calibration and compensation 有权

    公开(公告)号：US11129122B2

    公开(公告)日：2021-09-21

    申请号：US16549996

    申请日：2019-08-23

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Tiangao Gou ,  Pranav Dayal ,  Niranjan Ratnakar ,  Gennady Feygin ,  Jungwon Lee

    IPC: H04W56/00 ,  H03D3/00 ,  H04L27/36

    Abstract: A method for providing IQ mismatch (IQMM) compensation includes: sending a single tone signal at an original frequency; determining a first response of an impaired signal at the original frequency and a second response of the impaired signal at a corresponding image frequency; determining an estimate of a frequency response of the compensation filter at the original frequency based on the first response and the second response; repeating the steps of sending the single tone signal, determining the first response and the second response, and determining the estimate of the frequency response of the compensation filter by sweeping the single tone signal at a plurality of steps to determine a snapshot of the frequency response of the compensation filter; converting the frequency response of the compensation filter to a plurality of time-domain filter taps of the compensation filter by performing a pseudo-inverse of a time-to-frequency conversion matrix; and determining a time delay that provides a minimal LSE for the corresponding time domain filter taps.

    VIDEO DEPTH ESTIMATION BASED ON TEMPORAL ATTENTION
    29.
    发明申请
    VIDEO DEPTH ESTIMATION BASED ON TEMPORAL ATTENTION 有权

    公开(公告)号：US20210027480A1

    公开(公告)日：2021-01-28

    申请号：US16841618

    申请日：2020-04-06

    Applicant: Samsung Electronics Co., Ltd.

    Inventor： Haoyu Ren ,  Mostafa El-Khamy ,  Jungwon Lee

    IPC: G06T7/579 ,  G06T7/194 ,  G06N3/08

    Abstract: A method of depth detection based on a plurality of video frames includes receiving a plurality of input frames including a first input frame, a second input frame, and a third input frame respectively corresponding to different capture times, convolving the first to third input frames to generate a first feature map, a second feature map, and a third feature map corresponding to the different capture times, calculating a temporal attention map based on the first to third feature maps, the temporal attention map including a plurality of weights corresponding to different pairs of feature maps from among the first to third feature maps, each weight of the plurality of weights indicating a similarity level of a corresponding pair of feature maps, and applying the temporal attention map to the first to third feature maps to generate a feature map with temporal attention.

Patent Agency Ranking