Hangzhou Hikvision Digital Technology Co., Ltd.

Di Xie

Shiliang Pu

Yuan Zhang

Li Wang

Xiaoyang Wu

Lulu Zhou

Haiming Sun

Tao Song

Weijie Chen

Ping Yang

A deep neural network operation method and apparatus are provided. The method comprises: obtaining an input feature map of a network layer; displacing respectively, according to a preset displacement parameter, each of channels of the input feature map of the network layer along axes, to obtain a displaced feature map, wherein the preset displacement parameter comprises displacement amounts of the channel in the axes; and performing convolution operation on the displaced feature map with a 1×1 convolution kernel to obtain an output feature map of the network layer. The operation efficiency of the DNN can be improved through the above method.

Deep neural network operation method and apparatus

An operation method and apparatus for a network layer in a Deep Neural Network are provided. The method includes: acquiring a weighted tensor of the network layer in the Deep Neural Network, the weighted tensor comprising a plurality of filters; converting each filter into a linear combination of a plurality of fixed-point convolution kernels by splitting the filter, wherein a weight value of each of the fixed-point convolution kernels is a fixed-point quantized value having a specified bit-width; for each filter, performing a convolution operation on input data of the network layer and each of the fixed-point convolution kernels, respectively, to obtain a plurality of convolution results, and calculating a weighted sum of the obtained convolution results based on the linear combination of the plurality of fixed-point convolution kernels of the filter to obtain an operation result of the filter; determining output data of the network layer.

Operation method and apparatus for network layer in deep neural network

A deep neural network training method and apparatus and a computer device are provided. The deep neural network training method includes: obtaining task attributes of nodes in a current network layer in a tree-like network topology (S); performing cluster analysis on the nodes in the current network layer based on the task attributes of the nodes in the current network layer and extracting a common part of task attributes of multiple nodes in a same category as a task attribute for a parent node of the multiple nodes (S); training a network parameter of each parent node based on a task attribute of this parent node (S); and determining that training of a deep neural network corresponding to the tree-like network topology is completed, after completion of training of all nodes in all network layers (S). The operation efficiency of deep learning can be improved through this solution.

Deep neural network training method and apparatus, and computer device

Embodiments of methods and apparatuses for object detection and of computer devices are disclosed. The method for object detection includes: acquiring an image to be detected that is captured by an image capturing means; inputting the image to be detected into a fully convolutional neural network obtained by training to generate an object upper-vertex confidence distribution diagram, an object lower-vertex confidence distribution diagram, and an object upper-and-lower-vertex correlation diagram for the image to be detected; for the object upper-vertex confidence distribution diagram and the object lower-vertex confidence distribution diagram respectively, determining upper-vertex objects and lower-vertex objects in the image to be detected by using a preset object determination method; for each first vertex, calculating a correlation value of a connection line connecting the first vertex object and each of second vertex object respectively by mapping the upper-vertex objects and the lower-vertex object onto the object upper-and-lower-vertex correlation diagram; and based on the correlation values, determining a connection line having a maximum correlation value as a specified object by matching the upper-vertex objects and lower-vertex objects. The accuracy of object detection can be improved through the present solution.

Target detection method and apparatus, and computer device

Embodiments of the present application provide a method, apparatus, and electronic device for eliminating distortion of a distorted image. Side information components of a distorted image are generated. The distorted image is resulted from image processing on an original image. Side information components represent distortion features of the distorted image with respect to original image. Distorted image color components of the distorted image and the side information components are input into a pre-established convolutional neural network model for convolution filtering to obtain distortion-eliminated image color components. The convolutional neural network model is obtained through training based on a preset training set. The training set comprises original sample images, distorted image color components of multiple distorted images corresponding to each of the original sample images, and side information components of each of the distorted images. Thereby, a CNN-based distortion elimination process different from related art is provided.

Image distortion correction method and device and electronic device

The embodiments of the present application provide an image segmentation method, an image segmentation apparatus, and a fully convolutional network system. The method includes: acquiring a target image to be processed; acquiring image feature data of the target image; inputting the image feature data into a pre-trained target network far image segmentation to obtain an output; wherein the target network is a fully convolutional network comprising a hybrid context network structure, and the hybrid context network structure is configured to extract a plurality of reference features at a predetermined scale and fuse them into a target feature that matches a scale of a target object in a segmented image; and wherein the target network is trained with sample images containing target objects at different scales; and obtaining an image segmentation result for the target image based on the output. With this technical solution, the effectiveness of segmentation of target objects of different sizes in the image can be improved while ensuring a large receptive field.

Image segmentation method, apparatus, and fully convolutional network system

Embodiments of the present application disclose a target detection method and device, and relate to the technical field of video processing. The method comprises: obtaining an image sequence to be detected from a video to be detected according to an image sequence determining algorithm based on video timing (S), extracting a first CNN feature of the image sequence to be detected based on a pre-trained CNN model, performing feature fusion on the first CNN feature based on a second CNN feature to obtain a first fused CNN feature of the image sequence to be detected (S); inputting the first fused CNN feature into the first-level classifier, and obtaining first candidate target regions of the image sequence to be detected from an output of the first-level classifier (S); determining a first input region of the second-level classifier based on the first candidate target regions (S); obtaining a third CNN feature of the first input region based on the first fused CNN feature (S); inputting the third CNN feature into the second-level classifier, and obtaining a target detection result for the image sequence to be detected based on the output of the second-level classifier (S).

Growing community of inventors

Zhejiang, China

Di Xie