Image super-resolution method based on knowledge distillation

An image and knowledge technology, which is applied in the field of image super-resolution based on knowledge distillation, can solve problems such as slow speed, consume large computing resources, and weak learning ability of small models, and achieve the effect of fast running speed and reduced running time

Pending Publication Date: 2022-04-15
HANGZHOU ARCVIDEO TECHNOLOGY CO LTD
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The present invention aims at the problem that the image super-resolution model in the prior art is complex in design, consumes a large amount of computing resources and is slow in speed, and the simple small model has weak learning ability and cannot achieve a good super-resolution effect, and provides Image super-resolution method based on knowledge distillation

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 super-resolution method based on knowledge distillation
  • Image super-resolution method based on knowledge distillation
  • Image super-resolution method based on knowledge distillation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0063] An image super-resolution method based on knowledge distillation, the method includes;

[0064] Step 1, prepare the input and label required for training the super-resolution model, the input is a low-resolution image, and the label is a high-resolution image;

[0065] Step 2, the acquisition of complex models, based on the noise level image super-resolution method to obtain complex models;

[0066] Step 3, obtaining a simple model, constructing a multi-scale simple super-resolution model;

[0067] Step 4, model training, splicing the super-resolution models of steps 2 and 3, and training simple models based on complex models.

[0068] The preparation of the super-resolution image includes;

[0069] Obtain high-resolution images, and obtain high-resolution images through an image database;

[0070] Image block acquisition, for the acquired high-resolution image, image block sampling is performed through the set image block size, and multiple image blocks are obtained...

Embodiment 2

[0097] On the basis of Embodiment 1, the image super-resolution method based on knowledge distillation implemented in this embodiment,

[0098] Step 1: Prepare the input and labels required for training the super-resolution model. The input is a low-resolution image, and the label is a high-resolution image.

[0099] Step 1.1: The dataset consists of 5 public image databases, namely Vimeo, RealSR, REDS, DIV2K and Flickr2K, where Vimeo and REDS are video datasets, and each scene sequence consists of multiple consecutive images. Each database provides high-resolution images, and some databases provide corresponding low-resolution images. The present invention only uses high-resolution images, and the low-resolution images are generated from high-resolution images. The content, size, and number of images contained in each database are different.

[0100] Step 1.2: For the Vimeo and REDS datasets, in order to avoid duplication of image content, only one frame image is taken for e...

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 relates to an image processing technology, and discloses an image super-division method based on knowledge distillation, which comprises the following steps: preparing an input and a label required for training a super-division model, the input being a low-resolution image, and the label being a high-resolution image; a complex model is obtained, and the complex model is obtained through an image super-division method based on the noise level; obtaining a simple model, and constructing a multi-scale simple super-division model; and model training: splicing the super-division models in the step 2 and the step 3, and training a simple model based on a complex model. The input of the model is a low-resolution image, the simple model learns the super-resolution knowledge learned by the complex model through the trained complex super-resolution model by using the knowledge distillation technology, a better super-resolution result can be obtained, and the operation speed is higher.

Description

technical field [0001] The invention relates to image processing technology, in particular to an image super-resolution method based on knowledge distillation. Background technique [0002] Image super-resolution is to improve the spatial resolution of images, for example, the resolution of a picture is doubled from 960*540 to 1920*1080, which is convenient for users to watch on large-size display devices. Image super-resolution is one of the basic problems in image processing, and has a wide range of practical needs and application scenarios. With the popularization of mobile high-definition playback devices, the demand for converting low-resolution photos into high-definition and being able to quickly view and share them on various devices has never been higher. [0003] In recent years, with the continuous development of deep learning technology, convolutional neural network (CNNS) has received extensive attention in the direction of computer vision and has achieved rema...

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
Patent Type & Authority Applications(China)
IPC IPC(8): G06T3/40G06N3/04G06K9/62G06V10/774G06V10/82
Inventor 徐烂烂陈梅丽谢亚光孙彦龙
Owner HANGZHOU ARCVIDEO TECHNOLOGY CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap