Unlock instant, AI-driven research and patent intelligence for your innovation.

Deconvolution processing method and device, electronic equipment and medium

A processing method and deconvolution technology, applied in the field of convolutional neural network, can solve the problem of slow calculation speed of deconvolution, and achieve the effect of overcoming slow calculation speed

Pending Publication Date: 2022-06-03
PURPLE MOUNTAIN LAB
View PDF1 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] Technical purpose: In view of the above technical problems, the present invention discloses a deconvolution processing method, device, electronic equipment and medium, which solves the problem of slow deconvolution calculation speed and can be widely used in various scenarios of deep learning

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Deconvolution processing method and device, electronic equipment and medium
  • Deconvolution processing method and device, electronic equipment and medium
  • Deconvolution processing method and device, electronic equipment and medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0055] like figure 1 As shown, this embodiment provides a deconvolution processing method, including the steps:

[0056] Step 1: Convolve the input feature image with each sub-convolution kernel to obtain an intermediate feature image corresponding to the number of sub-convolution kernels; obtained after processing;

[0057] The original deconvolution kernel is the initial deconvolution kernel before the transformation is performed by the method of the present invention;

[0058] Step 2, online rearrangement of the intermediate feature picture to obtain an output feature picture.

[0059] In the present invention, the original deconvolution kernel is split into a plurality of sub-convolution kernels in advance, and the deconvolution operation of the feature picture is converted into a convolution operation, so as to speed up the deconvolution calculation speed, and a dedicated convolution accelerator can be used for accelerated calculation .

[0060] Specifically, offline ...

Embodiment 2

[0097] like Figure 4 As shown, this embodiment provides a deconvolution processing device, including:

[0098] The offline rearrangement module is used to rearrange the original deconvolution kernel offline to obtain multiple subconvolution kernels;

[0099] The convolution processing module is used to convolve the input feature image with each sub-convolution kernel to obtain the intermediate feature image corresponding to the number of sub-convolution kernels;

[0100] The online rearrangement module is used to rearrange the intermediate feature pictures online to obtain the output feature pictures.

[0101] Preferably, the offline rearrangement module includes:

[0102] A rotation module for rotating the original deconvolution kernel by 180°;

[0103] Zero-padding module, which is used according to the size K of the original deconvolution kernel d and the stride S of the deconvolution d , judge whether the zero-filling condition is satisfied, if so, perform zero-filli...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a deconvolution processing method and device, electronic equipment and a medium, and the method comprises the steps: carrying out the convolution calculation of an input feature picture and each sub-convolution kernel obtained through preprocessing, and obtaining intermediate feature pictures corresponding to the number of the sub-convolution kernels, the sub-convolution kernels are obtained by performing offline rearrangement preprocessing on an original deconvolution kernel; and carrying out online rearrangement on the intermediate feature picture to obtain an output feature picture. According to the method, the deconvolution kernel is divided into a plurality of sub-convolution kernels, the input feature picture and each sub-convolution kernel are subjected to convolution calculation, and finally, the deconvolution result is obtained through online data rearrangement, so that the calculation speed is improved, a special deconvolution accelerator does not need to be designed, and the calculation efficiency is improved. The method can be used for multiple deep learning application scenes such as super-resolution reconstruction, generative adversarial networks and target detection.

Description

technical field [0001] The present invention relates to the technical field of convolutional neural networks, and in particular, to a deconvolution processing method, device, electronic device and medium. Background technique [0002] Deep neural network (DNN) is the main component of deep learning. With the development of deep learning, DNN has shown excellent results in many fields. Deconvolution is an upsampling operator in DNN. The main purpose of upsampling is to enlarge the original image so that the image can be displayed on a higher resolution display device. It is widely used in super-resolution reconstruction, generative adversarial networks (GANs). ), target detection and other algorithms. Many dedicated neural network accelerators are designed to accelerate convolution, matrix multiplication and other operators and achieve good results. However, there is no corresponding acceleration effect for deconvolution. General deconvolution calculations need to first dete...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06N3/04
CPCG06N3/045
Inventor 王稷琛巫远黄永明夏井新
Owner PURPLE MOUNTAIN LAB