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

Image filling method and device in deep learning hardware

A technology of deep learning and filling method, which is applied in the direction of neural learning method, image memory management, graphics and image conversion, etc. It can solve the problems of not paying attention to filling operation, increasing algorithm delay and power consumption, etc.

Pending Publication Date: 2021-05-14
ASR SMART TECH CO LTD
View PDF1 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The implementation of software algorithms in hardware has certain limitations on some operations. For example, directly filling the input image requires multiple data transfers, which will increase unnecessary algorithm delay and power consumption.
[0004] Existing technologies usually only focus on the increase in storage space caused by the increase in data after the filling operation, but do not pay attention to how to perform the filling operation itself

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
  • Image filling method and device in deep learning hardware
  • Image filling method and device in deep learning hardware
  • Image filling method and device in deep learning hardware

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0022] see figure 1 , the image filling method proposed in this application in deep learning hardware includes the following steps.

[0023] Step S10: Perform padding estimation on the input image A, and calculate the size of the estimated padding image A1 (called the estimated padding image). The input image A refers to the input image of the neural network, or the input feature map of any layer in the neural network. The method of calculating the estimated padding is as follows: Let the height and width of the estimated padding image A1 be H_ and W_ respectively, H_=pad_t+pad_b+H, W_=pad_l+pad_r+W. Among them, pad_t is the number of rows of pixels filled above the input image A, pad_b is the number of rows of pixels filled below the input image A, pad_l is the number of columns of pixels filled on the left side of the input image A , pad_r is the number of columns of pixels to be padded on the right side of the input image A. pad_t, pad_b, pad_l, pad_r can take any value....

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 an image filling method in deep learning hardware. The image filling method comprises the following steps, S10, performing filling estimation on the input image A, and calculating the size of a filling estimation image A1; comparing the filling estimation image A1 with the input image A, calling pixel points newly added at the edge of the input image A a first expansion set. and step S20, generating a filling template B1. The output image B1 is obtained by performing deep convolution operation on the image B, and the output image B1 simultaneously meets the condition 1 and the condition 2 by controlling the selection of the image B and adjusting one or more of the size of the convolution kernel, the value of the convolution weight and the offset value. The output image B1 is a filling template and is written into a memory address. and S30, combining the input image A with the filling template B1 to obtain an image after the input image A is subjected to actual filling processing. According to the method, image filling of any size can be realized in deep learning hardware.

Description

technical field [0001] This application relates to an image filling method in a deep learning (deep learning) algorithm. Background technique [0002] Deep learning has become a hot spot in current AI (artificial intelligence, artificial intelligence) research, and deep learning algorithms need to be implemented on a computing platform with high performance and high computing power. The performance of general-purpose processors is poor. With the crazy growth of deep learning computing requirements, general-purpose processors can no longer meet the needs of deep learning computing. Therefore, special hardware is needed for acceleration to meet the high computing power requirements of deep learning. Deep learning hardware includes CPU (central processing unit, central processing unit), GPU (graphics processing unit, graphics processing unit), NPU (Network Processing Unit, network processor), FPGA (Field-Programmable Gate Array, field programmable logic gate array ), etc. Thes...

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): G06T3/00G06T1/60G06N3/063G06N3/08
CPCG06T1/60G06N3/063G06N3/08G06T3/04Y02D10/00
Inventor 吕亚婷范名超
Owner ASR SMART TECH CO LTD