公开(公告)号：US20170278220A1

公开(公告)日：2017-09-28

申请号：US15126271

申请日：2016-03-17

Applicant: HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY

Inventor： Lerenhan LI ,  Nong SANG ,  Changxin GAO ,  Luxin YAN ,  Jin WANG ,  Shiwei ZHANG ,  Yuanjie SHAO ,  Juncai PENG

IPC: G06T5/00 ,  G06T7/00 ,  G06T3/40

CPC classification number: G06T5/002 ,  G06T3/4007 ,  G06T3/4076 ,  G06T7/0002 ,  G06T2207/10048 ,  G06T2207/20076

Abstract: The invention discloses a method for correcting aero-optical thermal radiation noise, comprising steps of: pretreating a degraded image to obtain a multi-scale degraded image group, conducting iteration process of obtaining an optimal solution by using a last scale estimation result as an original value of next scale estimation according to the multi-scale degraded image group, thereby facilitating original-scale bias field estimation, and restoring the degraded image according to the original-scale bias field estimated value thereby obtaining an image after aero-optical thermal radiation noise correction. The invention also discloses a system for correcting aero-optical thermal radiation noise. The invention is capable of solving problems with conventional methods, comprising poor correction effect, high complexity, and incapability in correcting the thermal radiation noise at an image level, and applicable to restoration of an image with aero-optical thermal radiation noise.

公开(公告)号：US20210365743A1

公开(公告)日：2021-11-25

申请号：US17253124

申请日：2020-06-23

Applicant: HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY

Inventor： Nong SANG ,  Chuchu HAN ,  Yuanjie SHAO ,  Ruochen ZHENG ,  Changxin GAO

IPC: G06K9/62 ,  G06K9/00 ,  G06K9/46 ,  G06T3/40 ,  G06T7/11 ,  G06T7/73 ,  G06N3/04

Abstract: The invention discloses a person search method based on person re-identification driven localization refinement. On one hand, the region of interest (ROI) conversion module converts an original input image into a small image corresponding to a ROI, and contradiction existing in part of features shared by a person re-identification network and a detection network is avoided; and on the other hand, loss of the person re-identification network can be transmitted back to the detection network in a gradient manner through the ROI conversion module, the supervision of loss of the person re-identification network for the detection bounding box output by the detection network is realized, and the adjusted detection bounding box can effectively remove background interference, contains more useful attribute information and is more suitable for person search, so that the person search accuracy is greatly improved.

公开(公告)号：US20210248378A1

公开(公告)日：2021-08-12

申请号：US16965015

申请日：2020-01-07

Applicant: HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY

Inventor： Nong SANG ,  Shiwei ZHANG ,  Zhiyuan LI ,  Changxin GAO ,  Yuanjie SHAO

IPC: G06K9/00 ,  G06T3/00 ,  G06T7/207 ,  G06T7/215 ,  G06T3/40

Abstract: A spatiotemporal action detection method includes performing object detection on all frames of a sample video to obtain a candidate object set; calculating all interframe optical flow information on the sample video to obtain a motion set; constructing a spatiotemporal convolution-deconvolution network of an attention mechanism and a motion attention mechanism of an additional object; adding both a corresponding sparse variable and a sparse constraint to obtain a network structure S after performing spatiotemporal convolution processing on each time segment of the sample video; training the network structure S with an objective function based on classification loss and loss of the sparse constraint of cross entropy; and calculating an action category and a sparse coefficient corresponding to each time segment of a test sampled video to obtain an object action spatiotemporal location.

公开(公告)号：US20180068430A1

公开(公告)日：2018-03-08

申请号：US15709454

申请日：2017-09-19

Applicant: HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY

Inventor： Nong SANG ,  Lerenhan LI ,  Luxin YAN ,  Changxin GAO ,  Yuanjie SHAO ,  Juncai PENG ,  Shiwei ZHANG ,  Jin WANG

IPC: G06T7/00 ,  G06N3/02 ,  G06F17/11 ,  G06F17/18

CPC classification number: G06T7/0002 ,  G06F17/11 ,  G06F17/18 ,  G06K9/40 ,  G06K9/4628 ,  G06N3/02 ,  G06T5/003 ,  G06T2207/20076 ,  G06T2207/20081 ,  G06T2207/20084

Abstract: A method for estimating a blur kernel size, the method including: (1) pre-processing a blurred image, to obtain an image, so that a size of the image satisfies an image input size of a multi-class convolutional neural network (CNN); (2) inputting the image into a multi-class CNN with completed training, to obtain a blur-kernel-size probability distribution vector; and (3) comparing each element in the blur-kernel-size probability distribution vector, so that an estimated blur kernel size of the blurred image is the blur kernel size corresponding to a largest element. The invention also provides a system for estimating a blur kernel size. The system includes an image pre-processing module, a training-set synthesizing module, a multi-class CNN module, and a blur kernel size estimation module.