CN115861343A - Image representation method and system of arbitrary scale based on dynamic implicit image function - Google Patents

Abstract

The invention discloses an arbitrary scale image representation method and system based on a dynamic implicit image function, wherein the method comprises the steps of obtaining an image to be processed; carrying out implicit coding processing on the image to be processed through a pre-trained coder to obtain a two-dimensional characteristic diagram; and inputting the two-dimensional characteristic diagram into a dynamic implicit image network, carrying out dynamic coordinate slicing processing on the two-dimensional characteristic diagram, and carrying out pixel value prediction processing through a double-stage multilayer perceptron to obtain an image pixel value. The embodiment of the invention can reduce the calculation cost of continuous representation of the image, improve the processing performance and can be widely applied to the technical field of artificial intelligence.

Description

Image representation method and system of any scale based on dynamic implicit image function

technical field

The invention relates to the technical field of artificial intelligence, in particular to a method and system for representing images of arbitrary scales based on dynamic implicit image functions.

Background technique

A digital image is a two-dimensional representation of the real world in the digital world, but the continuous physical world is often quantized in sensors and stored in a computer as a matrix of discrete pixels. If the image can be expressed in a continuous form, an image of any resolution can be obtained in the continuous space, thereby ensuring the accuracy of the scene described by the image. Although the continuous image representation method in the related art has excellent performance in continuous image representation, the computational cost will increase quadratically with the increase of image magnification, making super-resolution reconstruction of any scale time-consuming. Based on the above, the technical problems existing in the related technologies need to be solved urgently.

Contents of the invention

In view of this, embodiments of the present invention provide a method and system for representing images of arbitrary scales based on dynamic implicit image functions, so as to reduce computing costs and improve processing performance.

On the one hand, the present invention provides a method for representing images of arbitrary scales based on dynamic implicit image functions, including:

Get the image to be processed;

performing implicit encoding processing on the image to be processed by a pre-trained encoder to obtain a two-dimensional feature map;

Inputting the two-dimensional feature map into a dynamic implicit image network, performing dynamic coordinate slice processing on the two-dimensional feature map, and performing pixel value prediction processing through a two-stage multi-layer perceptron to obtain image pixel values.

Optionally, the performing dynamic coordinate slicing processing on the two-dimensional feature map includes:

Enter image magnification;

Obtaining a feature vector from the two-dimensional feature map is determined as a hidden code, and grouping coordinates in the two-dimensional feature map according to the hidden code to obtain a feature coordinate group;

Slicing is performed on the feature coordinate group according to the image magnification to obtain a coordinate slice.

Optionally, performing slice processing on the feature coordinate group according to the image magnification to obtain a coordinate slice includes:

determining slice intervals according to the image magnification;

The feature coordinate group is divided according to the slice interval to obtain a coordinate slice, and the coordinate slice is used to share the same hidden code for all coordinates in the slice.

Optionally, said performing pixel value prediction processing through a two-stage multi-layer perceptron includes:

Input coordinate slice and slice hidden code;

performing the first-stage processing on the coordinate slice and the slice hidden code to obtain a slice hidden vector;

Acquiring coordinates to be predicted, where the coordinates to be predicted are any coordinates in the coordinate slice;

The second-stage processing is performed on the hidden vector of the slice according to the coordinates to be predicted to obtain the pixel values of the coordinates to be predicted.

Optionally, the two-stage multilayer perceptron includes a hidden layer, and the hidden layer is composed of a linear layer and an activation function.

Optionally, before the pre-trained encoder performs implicit encoding processing on the image to be processed to obtain a two-dimensional feature map, the method further includes pre-training the encoder and a dynamic implicit image network, Specifically include:

Get training images;

performing pixel prediction processing on the training image through the encoder and the dynamic implicit image network to obtain predicted pixel values;

determining a pixel loss value based on the pixel values of the training image and the predicted pixel values;

The weight parameters of the encoder and the dynamic implicit image network are updated according to the pixel loss value to obtain a trained encoder and dynamic implicit image network.

On the other hand, an embodiment of the present invention also provides a system, including:

The first module is used to obtain the image to be processed;

The second module is used to perform implicit encoding processing on the image to be processed through a pre-trained encoder to obtain a two-dimensional feature map;

The third module is used to input the two-dimensional feature map into the dynamic implicit image network, perform dynamic coordinate slice processing on the two-dimensional feature map, and perform pixel value prediction processing through a two-stage multi-layer perceptron to obtain image pixels value.

Optionally, the third module includes:

The first submodule is used to perform dynamic coordinate slice processing on the two-dimensional feature map;

The second sub-module is used to perform pixel value prediction processing through a two-stage multi-layer perceptron.

Optionally, the first submodule includes:

The first unit is used to input image magnification;

The second unit is configured to obtain a feature vector from the two-dimensional feature map and determine it as a hidden code, and group coordinates in the two-dimensional feature map according to the hidden code to obtain a feature coordinate group;

The third unit is configured to perform slice processing on the feature coordinate group according to the image magnification to obtain coordinate slices.

Optionally, the second submodule includes:

The fourth unit is used to input coordinate slices and slice hidden codes;

The fifth unit is used to perform the first-stage processing on the coordinate slice and the slice hidden code to obtain the slice hidden vector;

The sixth unit is used to acquire coordinates to be predicted, where the coordinates to be predicted are any coordinates in the coordinate slice;

The seventh unit is configured to perform a second-stage process on the hidden vector of the slice according to the coordinates to be predicted to obtain pixel values of the coordinates to be predicted.

Compared with the prior art, the present invention adopts the above technical scheme and has the following technical effects: the embodiment of the present invention inputs the two-dimensional feature map into the dynamic implicit image network, and performs dynamic coordinate slice processing on the two-dimensional feature map, which can Make the neural network perform a many-to-many mapping from coordinate slices to pixel value slices, so that the decoder can use only one hidden code to predict all pixel values corresponding to the coordinate slice, reducing computational costs; and through a two-stage multi-layer perceptron. The pixel value prediction process obtains the image pixel value, which enables the decoder to use a non-fixed number of coordinates as input to reduce the number of hidden layers and improve processing performance.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

Fig. 1 is a flow chart of an image representation method of any scale based on a dynamic implicit image function provided by an embodiment of the present application;

FIG. 2 is an overall framework diagram of a dynamic implicit image function provided by an embodiment of the present application;

FIG. 3 is an example diagram of a coordinate slice provided by an embodiment of the present application;

FIG. 4 is a structural diagram of a dual-stage multilayer perceptron provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not intended to limit the present application.

The arbitrary-scale image representation method and system based on dynamic implicit image functions provided in the embodiments of the present application mainly involve artificial intelligence technology. Artificial Intelligence (AI) technology is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence technology is a comprehensive subject that involves a wide range of fields, including both hardware-level technology and software-level technology. Among them, artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, and mechatronics; artificial intelligence software technologies mainly include computer vision technology, voice Processing technology, natural language processing technology and machine learning/deep learning and other major directions.

Specifically, the arbitrary-scale image representation method and system based on dynamic implicit image functions provided in the embodiments of the present application can use computer vision technology in the field of artificial intelligence and machine learning/deep learning technology to analyze and process images to obtain image continuous image representation. It can be understood that, for different tasks, the methods provided in the embodiments of the present application can be executed in the corresponding application scenarios of the artificial intelligence system; and, the specific timing for executing these methods can be in the operating process of the artificial intelligence system any link.

Implicit Neural Representation Techniques: Compared with explicit representations, implicit neural representations are able to capture object details with a small number of parameters, and their differentiable properties allow backpropagation through neural rendering models. However, when implicit neural representations are applied to 2D vision tasks, they usually need to predict each pixel independently, which requires a lot of computational cost and long running time.

Local Implicit Image Function (LIIF), a novel implicit representation of images, uses a multi-layer perceptron to infer pixel values at each coordinate.

In related art, although LIIF can provide stable performance in arbitrary scale super-resolution tasks up to 30×, its computational cost increases rapidly with the increase of magnification.

In view of this, referring to FIG. 1 , an embodiment of the present invention provides an image representation method of any scale based on a dynamic implicit image function, including:

S101. Obtain an image to be processed;

S102. Perform implicit encoding processing on the image to be processed by using a pre-trained encoder to obtain a two-dimensional feature map;

S103. Input the two-dimensional feature map into a dynamic implicit image network, perform dynamic coordinate slice processing on the two-dimensional feature map, and perform pixel value prediction processing through a two-stage multilayer perceptron to obtain image pixel values.

其中，X_1st表示坐标切片的首坐标，X_last表示坐标切片的尾坐标。然后解码函数使用上述的信息来预测该坐标切片的所有像素值，即通过双阶段多层感知器(或称为粗到细的多层感知器)进行坐标的像素值预测，通过第一阶段(粗略阶段)预测切片隐向量H^*，并和待预测坐标X_i一起作为第二阶段(精细阶段)的输入，输出得到待预测坐标的像素值I_out-i。本发明实施例在训练阶段，使用预测得到的像素值I_out-i和真实图像的像素值I_gt-i计算损失函数，编码器和解码函数在自监督超分辨率任务中联合训练，而学习到的网络参数由所有图像共享。本发明实施例通过使用图像坐标分组和切片策略，使神经网络能够执行从坐标切片到像素值切片的多对多映射，而不是每次单独预测给定坐标的像素值。本发明实施例进一步提出了双阶段多层感知器(Coarse-to-FineMultilayer Perceptron,C2F-MLP)，用于执行基于动态坐标切片策略的图像解码，使得每个切片中的坐标数量能随着放大倍数的变化而变化，使用动态坐标切片策略的DIIF可以显著降低大尺度超分辨率需要的计算成本。实验结果表明，与现有的任意尺度超分辨率方法相比，DIIF实现了最佳的计算效率和超分辨率性能。In the embodiment of the present invention, a dynamic implicit image function (Dynamic Implicit Image Function, DIIF) is proposed, which is a fast and effective method for representing images of any scale. Referring to Figure 2, I _in represents the input image, and the encoder maps the input image to a 2D feature map as its DIIF representation. Given the resolution of the real image, the hidden code z ^* can be obtained from the 2D feature map, as well as the coordinate slice around the latent code

Among them, X _1st represents the first coordinate of the coordinate slice, and X _last represents the tail coordinate of the coordinate slice. Then the decoding function uses the above information to predict all pixel values of the coordinate slice, that is, the pixel value prediction of the coordinates is performed through a two-stage multi-layer perceptron (or called a coarse-to-fine multi-layer perceptron), and the first stage ( Coarse stage) Predict slice hidden vector H ^* , and take the coordinates _Xi to be predicted as the input of the second stage (fine stage), and output the pixel value I _out-i of the coordinates to be predicted. In the embodiment of the present invention, in the training phase, the predicted pixel value I _out-i and the pixel value I _gt-i of the real image are used to calculate the loss function, the encoder and the decoding function are jointly trained in the self-supervised super-resolution task, and the learning The network parameters obtained are shared by all images. The embodiment of the present invention enables the neural network to perform a many-to-many mapping from coordinate slices to pixel value slices by using image coordinate grouping and slice strategies, instead of separately predicting the pixel value of a given coordinate each time. The embodiment of the present invention further proposes a two-stage multilayer perceptron (Coarse-to-FineMultilayer Perceptron, C2F-MLP), which is used to perform image decoding based on a dynamic coordinate slicing strategy, so that the number of coordinates in each slice can be enlarged with The DIIF using the dynamic coordinate slicing strategy can significantly reduce the computational cost required for large-scale super-resolution. Experimental results show that DIIF achieves the best computational efficiency and super-resolution performance compared with existing arbitrary-scale super-resolution methods.

Further as a preferred implementation manner, the performing dynamic coordinate slicing processing on the two-dimensional feature map includes:

Enter image magnification;

Obtaining a feature vector from the two-dimensional feature map is determined as a hidden code, and grouping coordinates in the two-dimensional feature map according to the hidden code to obtain a feature coordinate group;

Slicing is performed on the feature coordinate group according to the image magnification to obtain a coordinate slice.

In the embodiment of the present invention, a vector is selected from the two-dimensional feature map as a hidden code, and the coordinates in the two-dimensional feature map that are closer to the hidden code than other hidden codes are grouped according to the hidden code to obtain a feature coordinate group . The hidden code can be shared within a coordinate group through the feature coordinate group, so that the decoder can use the hidden code only once to predict all pixel values corresponding to the coordinate group. The number of coordinates in a coordinate group is proportional to the magnification, so the greater the magnification, the greater the computational cost savings. Coordinate grouping requires the decoder to predict all pixel values of the coordinate group simultaneously, which imposes a heavy burden on the decoder when performing large-scale super-resolution. The embodiment of the present invention provides a reasonable solution to slice the feature coordinate group according to the magnification of the image to obtain a coordinate slice. By dividing a coordinate group into multiple coordinate slices, and only in the coordinate slice instead of the entire coordinate Share hidden code input within the group.

As a further preferred implementation, the said feature coordinate group is sliced according to the image magnification to obtain a coordinate slice, including:

determining slice intervals according to the image magnification;

The feature coordinate group is divided according to the slice interval to obtain a coordinate slice, and the coordinate slice is used to share the same hidden code for all coordinates in the slice.

Among them, setting an appropriate slice interval to achieve the best performance and efficiency balance, the simplest method is fixed coordinate slice, which uses a fixed slice interval in any case. However, this strategy preserves the quadratic increase in computational cost as the magnification increases. In addition, there will be two major problems of spatial discontinuity and redundant coordinates inside the coordinate slice. In order to solve these problems, the embodiment of the present invention proposes dynamic coordinate slicing, so as to adjust the slice interval when the magnification factor changes. In the embodiment of the present invention, the first strategy may be linear-order coordinate slicing, which sets the slicing interval as the magnification factor. When using linear-order coordinate slices, the computational cost of DIIF increases linearly with magnification. Another strategy is to set the slice interval as the square of the magnification factor, which is called constant-order coordinate slicing. When using constant-order coordinate slices, the computational cost of DIIF is determined only by the resolution of the input image, which remains constant as the magnification increases. In the embodiment of the present invention, the feature coordinate group is divided according to the slice interval to obtain a coordinate slice, and the coordinate slice is used to share the same hidden code for all coordinates in the slice. Referring to Fig. 3, Fig. 3 is a coordinate grouping with a magnification of 4 and a coordinate slice with a slice interval of 4. Z ^* represents the hidden code, X _1st represents the first coordinate of the coordinate slice, and X _last represents the tail coordinate of the coordinate slice.

Further as a preferred embodiment, the pixel value prediction process performed by a two-stage multi-layer perceptron includes:

Input coordinate slice and slice hidden code;

performing the first-stage processing on the coordinate slice and the slice hidden code to obtain a slice hidden vector;

Acquiring coordinates to be predicted, where the coordinates to be predicted are any coordinates in the coordinate slice;

The second-stage processing is performed on the hidden vector of the slice according to the coordinates to be predicted to obtain the pixel values of the coordinates to be predicted.

Among them, in order to perform the dynamic coordinate slicing strategy, the decoder needs to have the scalability to use a non-fixed number of coordinates as input and output corresponding pixel values. However, ordinary MLPs only allow fixed-length vectors as input. In order to solve this problem, the embodiment of the present invention proposes a two-stage multi-layer perceptron (C2F-MLP) as a decoder, which is divided into the first stage (coarse stage) for predicting the hidden vector of the slice and the first stage (coarse stage) for predicting the pixel value The second stage (fine stage). In the embodiment of the present invention, the hidden layer of the rough stage takes the boundary coordinates of the coordinate slice and its corresponding hidden code as input to generate a hidden vector of the slice. Slice latent vectors contain information about all pixel values in the slice and are used as input to the refinement stage. The computational cost of the coarse stage is determined by the number of coordinate slices, which is much smaller than the number of output coordinates due to the dynamic coordinate slice strategy. The coarse stage also allows the decoding function to exploit the spatial relationships within slices, which makes its predictions of pixel values more accurate. The hidden layer of the fine stage takes as input the slice latent vector output by the coarse stage and any coordinate in the given coordinate slice to predict the pixel value at that coordinate. The refinement stage is designed to independently predict pixel values at the coordinates to be predicted. The decoding function used in the fine stage can be expressed as:

I(X ^* ) = f _θ (z ^* ,[x _tl -v ^* ,...,x _rb -v ^* ]);

In the formula, I is the pixel value, X ^* ＝[x _tl ,…,x _rb ] is the given coordinate slice, f _θ is the decoder, z ^* is the hidden code corresponding to the coordinate slice, v ^* is the coordinate of the hidden code , x _tl and x _rb are the first and last coordinates of the coordinate slice, respectively.

Since the length of the sliced latent vector is shorter than that of the latent code, and the number of hidden layers in the fine stage is smaller, the computational cost required for the fine stage of DIIF is significantly lower compared with that of LIIF's decoder.

As a further preferred embodiment, the two-stage multilayer perceptron includes a hidden layer, and the hidden layer is composed of a linear layer and an activation function.

Referring to Figure 4, C2F-MLP divides the decoder into a coarse stage for predicting slice latent vectors and a fine stage for predicting pixel values. The hidden layer of C2F-MLP consists of a linear layer of dimension 256 followed by a ReLU activation function. In the rough stage, the hidden code z ^* , the first coordinate X _1st of the coordinate slice, the last coordinate X _last of the coordinate slice, and the area a of the pixel point under the current magnification are taken as input, and the hidden coordinate vector H _lt~rb is output as output. In the refinement stage, input coordinate latent vector and coordinate X _I to be predicted, and output I _i . To predict RGB values, the refinement stage ends with an output linear layer of dimension 3.

As a further preferred embodiment, before the pre-trained encoder performs implicit encoding on the image to be processed to obtain a two-dimensional feature map, the method further includes pre-training the encoder and dynamic implicit Image network, specifically including:

Get training images;

performing pixel prediction processing on the training image through the encoder and the dynamic implicit image network to obtain predicted pixel values;

determining a pixel loss value based on the pixel values of the training image and the predicted pixel values;

The weight parameters of the encoder and the dynamic implicit image network are updated according to the pixel loss value to obtain a trained encoder and dynamic implicit image network.

In the embodiment of the present invention, the training phase uses the predicted pixel values and the pixel values of the real image to calculate the pixel-level loss. The encoder and decoding functions are jointly trained in the self-supervised super-resolution task, while the learned network parameters are shared by all images.

On the other hand, an embodiment of the present invention also provides a system, including:

The first module is used to obtain the image to be processed;

The second module is used to perform implicit encoding processing on the image to be processed through a pre-trained encoder to obtain a two-dimensional feature map;

The third module is used to input the two-dimensional feature map into the dynamic implicit image network, perform dynamic coordinate slice processing on the two-dimensional feature map, and perform pixel value prediction processing through a two-stage multi-layer perceptron to obtain image pixels value.

Optionally, the third module includes:

The first submodule is used to perform dynamic coordinate slice processing on the two-dimensional feature map;

The second sub-module is used to perform pixel value prediction processing through a two-stage multi-layer perceptron.

Optionally, the first submodule includes:

The first unit is used to input image magnification;

The second unit is configured to obtain a feature vector from the two-dimensional feature map and determine it as a hidden code, and group coordinates in the two-dimensional feature map according to the hidden code to obtain a feature coordinate group;

The third unit is configured to perform slice processing on the feature coordinate group according to the image magnification to obtain coordinate slices.

Optionally, the second submodule includes:

The fourth unit is used to input coordinate slices and slice hidden codes;

The fifth unit is used to perform the first-stage processing on the coordinate slice and the slice hidden code to obtain the slice hidden vector;

The sixth unit is used to acquire coordinates to be predicted, where the coordinates to be predicted are any coordinates in the coordinate slice;

The seventh unit is configured to perform a second-stage process on the hidden vector of the slice according to the coordinates to be predicted to obtain pixel values of the coordinates to be predicted.

The invention proposes an image representation method and system of any scale based on a dynamic implicit image function, which is used for fast and effective representation of images of any scale. In DIIF, a pixel-based image is represented as a 2D feature map, and the decoding function takes a coordinate slice and a local feature vector as input and predicts the corresponding set of pixel values. By sharing local feature vectors inside coordinate slices, DIIF enables large-scale super-resolution reconstruction at extremely low computational cost. Experimental results demonstrate that DIIF outperforms existing arbitrary-scale super-resolution methods in terms of super-resolution performance and computational efficiency at all zoom factors. Compared with LIIF, DIIF can save up to 87% computational cost and consistently have better PSNR performance. DIIF can be efficiently applied to scenes that need to display images in real time at any resolution. By applying the embodiment of the present invention, any scaling function in image viewing/editing software, magnification and restoration of low-resolution images, and compression and storage of high-resolution images can be realized.

In some alternative implementations, the functions/operations noted in the block diagrams may occur out of the order noted in the operational diagrams. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality/operations involved. Furthermore, the embodiments presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logical flow presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are performed independently.

Furthermore, although the invention has been described in the context of functional modules, it should be understood that one or more of the described functions and/or features may be integrated into a single physical device and/or unless stated to the contrary. or software modules, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary to understand the present invention. Rather, given the attributes, functions and internal relationships of the various functional blocks in the devices disclosed herein, the actual implementation of the blocks will be within the ordinary skill of the engineer. Accordingly, those skilled in the art can implement the present invention set forth in the claims without undue experimentation using ordinary techniques. It is also to be understood that the particular concepts disclosed are illustrative only and are not intended to limit the scope of the invention which is to be determined by the appended claims and their full scope of equivalents.

The logic and/or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium, For use with instruction execution systems, devices, or devices (such as computer-based systems, systems including processors, or other systems that can fetch instructions from instruction execution systems, devices, or devices and execute instructions), or in conjunction with these instruction execution systems, devices or equipment for use. For the purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use in or in conjunction with an instruction execution system, device or device.

It should be understood that various parts of the present invention can be realized by hardware, software, firmware or their combination. In the embodiments described above, various steps or methods may be implemented by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or combination of the following techniques known in the art: Discrete logic circuits, ASICs with suitable combinational logic gates, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although the embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions and modifications can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the invention is defined by the claims and their equivalents.

The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements without violating the spirit of the present invention. These equivalent modifications or replacements are all within the scope defined by the claims of the present application.

Claims (10)

1. A method for representing images of any scale based on dynamic implicit image functions, characterized in that the method comprises: Get the image to be processed; performing implicit encoding processing on the image to be processed by a pre-trained encoder to obtain a two-dimensional feature map; Inputting the two-dimensional feature map into a dynamic implicit image network, performing dynamic coordinate slice processing on the two-dimensional feature map, and performing pixel value prediction processing through a two-stage multi-layer perceptron to obtain image pixel values. 2. The method according to claim 1, wherein said performing dynamic coordinate slice processing on said two-dimensional feature map comprises: Enter image magnification; Obtaining a feature vector from the two-dimensional feature map is determined as a hidden code, and grouping coordinates in the two-dimensional feature map according to the hidden code to obtain a feature coordinate group; Slicing is performed on the feature coordinate group according to the image magnification to obtain a coordinate slice. 3. The method according to claim 2, wherein the step of performing slice processing on the feature coordinate group according to the image magnification to obtain a coordinate slice includes: determining slice intervals according to the image magnification; The feature coordinate group is divided according to the slice interval to obtain a coordinate slice, and the coordinate slice is used to share the same hidden code for all coordinates in the slice. 4. The method according to claim 1, wherein said performing pixel value prediction processing by a two-stage multi-layer perceptron comprises: Input coordinate slice and slice hidden code; performing the first-stage processing on the coordinate slice and the slice hidden code to obtain a slice hidden vector; Acquiring coordinates to be predicted, where the coordinates to be predicted are any coordinates in the coordinate slice; The second-stage processing is performed on the hidden vector of the slice according to the coordinates to be predicted to obtain the pixel values of the coordinates to be predicted. 5. The method of claim 1, wherein the two-stage multilayer perceptron includes a hidden layer consisting of a linear layer and an activation function. 6. The method according to any one of claims 1 to 5, characterized in that, before the pre-trained encoder performs implicit encoding on the image to be processed to obtain a two-dimensional feature map, the The method also includes pre-training the encoder and the dynamic implicit image network, comprising: Get training images; performing pixel prediction processing on the training image through the encoder and the dynamic implicit image network to obtain predicted pixel values; determining a pixel loss value based on the pixel values of the training image and the predicted pixel values; The weight parameters of the encoder and the dynamic implicit image network are updated according to the pixel loss value to obtain a trained encoder and dynamic implicit image network. 7. An arbitrary-scale image representation system based on dynamic implicit image functions, characterized in that the system includes: The first module is used to obtain the image to be processed; The second module is used to perform implicit encoding processing on the image to be processed through a pre-trained encoder to obtain a two-dimensional feature map; The third module is used to input the two-dimensional feature map into the dynamic implicit image network, perform dynamic coordinate slice processing on the two-dimensional feature map, and perform pixel value prediction processing through a two-stage multi-layer perceptron to obtain image pixels value. 8. The system according to claim 7, wherein the third module comprises: The first submodule is used to perform dynamic coordinate slice processing on the two-dimensional feature map; The second sub-module is used to perform pixel value prediction processing through a two-stage multi-layer perceptron. 9. The system according to claim 8, wherein the first submodule comprises: The first unit is used to input image magnification; The second unit is configured to obtain a feature vector from the two-dimensional feature map and determine it as a hidden code, and group coordinates in the two-dimensional feature map according to the hidden code to obtain a feature coordinate group; The third unit is configured to perform slice processing on the feature coordinate group according to the image magnification to obtain coordinate slices. 10. The system according to claim 8, wherein the second submodule comprises: The fourth unit is used to input coordinate slices and slice hidden codes; The fifth unit is used to perform the first-stage processing on the coordinate slice and the slice hidden code to obtain the slice hidden vector; The sixth unit is used to acquire coordinates to be predicted, where the coordinates to be predicted are any coordinates in the coordinate slice; The seventh unit is configured to perform a second-stage process on the hidden vector of the slice according to the coordinates to be predicted to obtain pixel values of the coordinates to be predicted.

Images