CN114429548A

CN114429548A - Image processing method, neural network and training method, device and equipment thereof

Info

Publication number: CN114429548A
Application number: CN202210107300.4A
Authority: CN
Inventors: 韩文华
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-01-28
Filing date: 2022-01-28
Publication date: 2022-05-03

Abstract

The disclosure provides an image processing method, a neural network and a training method, device and equipment thereof, relating to the technical field of artificial intelligence, in particular to computer vision, image processing and deep learning technology. The method comprises the following steps: performing cooperative processing on an image to be processed by utilizing a plurality of coding layers to obtain a plurality of first feature maps with different sizes; performing feature extraction on each first feature map to obtain at least one second feature map with a target size, wherein the target size comprises the size of the first feature map with the smallest size in the plurality of first feature maps; performing feature extraction on the first feature map with the minimum size to obtain a third feature map; determining a first feature map to be fused in the second feature map based on the size of the third feature map; fusing the first feature map to be fused with the third feature map to obtain a fourth feature map; and performing cooperative processing on the fourth feature map by using a plurality of decoding layers to obtain an image processing result.

Description

Image processing method, neural network and training method, device and equipment thereof

Technical Field

The present disclosure relates to the field of artificial intelligence technologies, and in particular, to computer vision, image processing, and deep learning technologies, and in particular, to an image processing method, a neural network training method, a neural network, an electronic device, a computer-readable storage medium, and a computer program product.

Background

Artificial intelligence is the subject of research that makes computers simulate some human mental processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), both at the hardware level and at the software level. The artificial intelligence hardware technology generally comprises technologies such as a sensor, a special artificial intelligence chip, cloud computing, distributed storage, big data processing and the like, and the artificial intelligence software technology mainly comprises a computer vision technology, a voice recognition technology, a natural language processing technology, machine learning/deep learning, a big data processing technology, a knowledge graph technology and the like.

With the development and application of artificial intelligence related technologies, a strong demand for intelligent and automatic technologies emerges from more and more fields, one of which is the image/video processing field. At present, portrait segmentation is an important branch in the field, and the application and land size is large, so that each application program has great demands on portrait segmentation, but at the same time has high requirements, especially the most important in precision and speed indexes.

The approaches described in this section are not necessarily approaches that have been previously conceived or pursued. Unless otherwise indicated, it should not be assumed that any of the approaches described in this section qualify as prior art merely by virtue of their inclusion in this section. Similarly, unless otherwise indicated, the problems mentioned in this section should not be considered as having been acknowledged in any prior art.

Disclosure of Invention

The present disclosure provides an image processing method, a training method of a neural network, an electronic device, a computer-readable storage medium, and a computer program product.

According to an aspect of the present disclosure, there is provided an image processing method. The method comprises the following steps: performing cooperative processing on an image to be processed by utilizing a plurality of coding layers to obtain a plurality of first feature maps with different sizes; performing feature extraction on each first feature map in the plurality of first feature maps to obtain at least one second feature map with a target size, wherein the target size comprises the size of the first feature map with the smallest size in the plurality of first feature maps; performing feature extraction on the first feature map with the minimum size to obtain a third feature map which is different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third feature map has the same size as the first feature map with the minimum size; determining at least one first feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps based on the size of the third feature map; fusing the at least one first feature map to be fused with the third feature map to obtain a fourth feature map; and performing cooperative processing on the fourth feature map by using a plurality of decoding layers to obtain an image processing result.

According to an aspect of the present disclosure, a method of training a neural network is provided. The neural network includes a plurality of encoding layers, a plurality of first feature extraction sub-networks following the plurality of encoding layers, a second feature extraction sub-network following the plurality of encoding layers, and a plurality of decoding layers following the second feature extraction sub-network. The method comprises the following steps: acquiring a sample image and a real label of the sample image; performing cooperative processing on the sample image by using a plurality of coding layers to obtain a plurality of first sample feature maps with different sizes; for each first sample feature map in the plurality of first sample feature maps, processing the first sample feature map by using at least one first feature extraction sub-network corresponding to the first sample feature map in the plurality of first feature extraction sub-networks to obtain at least one second sample feature map with a target size, wherein the target size comprises the size of the first feature map with the smallest size in the plurality of first feature maps; processing the first sample feature map with the minimum size by using a second feature extraction sub-network to obtain a third sample feature map which is different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third sample feature map has the same size as the first sample feature map with the minimum size; determining at least one first sample feature map to be fused in at least one second sample feature map corresponding to each of the plurality of first sample feature maps based on the size of the third sample feature map; fusing at least one first sample feature map to be fused with the third sample feature map to obtain a fourth sample feature map; performing cooperative processing on the fourth sample feature map by using a plurality of decoding layers to obtain a prediction label of the sample image; and calculating a loss value based on the predicted tag and the real tag to adjust a parameter of the neural network.

According to another aspect of the present disclosure, a neural network for image processing is provided. The neural network includes: the image processing device comprises a plurality of encoding layers, a plurality of image processing units and a plurality of image processing units, wherein the encoding layers are configured to carry out cooperative processing on images to be processed by the encoding layers so as to output a plurality of first feature maps with different sizes; a first feature extraction sub-network corresponding to each of the plurality of first feature maps following the plurality of encoding layers, configured to process the first feature maps to output at least one second feature map having a target size comprising a size of a smallest-sized first feature map of the plurality of first feature maps; a second feature extraction sub-network following the plurality of encoding layers, configured to process the first feature map with the smallest size to output a third feature map different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third feature map has the same size as the first feature map with the smallest size; a first fusion layer behind the second feature extraction sub-network, configured to fuse at least one first feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps with the third feature map to output a fourth feature map, wherein the at least one first feature map to be fused is determined based on a size of the fourth feature map; and a plurality of decoding layers behind the second feature extraction sub-network configured to perform cooperative processing on the third feature map to output an image processing result.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the image processing method or the neural network training method described above.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to execute the image processing method or the neural network training method described above.

According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program, when executed by a processor, implements the image processing method or the neural network training method described above.

According to one or more embodiments of the disclosure, the first feature maps of multiple scales are processed for multiple times in the encoding stage, so that a second feature map of a target size including the size of the deepest layer feature can be generated for each first feature map, and then the deepest layer semantic feature is fused with the second feature map of the deepest layer feature before being upsampled, so that semantic information is enriched, and the precision of an image processing result is improved.

It should be understood that the statements in this section are not intended to identify key or critical features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the embodiments and, together with the description, serve to explain the exemplary implementations of the embodiments. The illustrated embodiments are for purposes of illustration only and do not limit the scope of the claims. Throughout the drawings, identical reference numbers designate similar, but not necessarily identical, elements.

FIG. 1 illustrates a schematic diagram of an exemplary system in which various methods described herein may be implemented, according to an embodiment of the present disclosure;

FIG. 2 shows a flow chart of an image processing method according to an exemplary embodiment of the present disclosure;

fig. 3 shows a flowchart of feature extraction for each first feature map according to an exemplary embodiment of the present disclosure;

fig. 4 illustrates a flowchart of feature extraction for a first feature map of a smallest size according to an exemplary embodiment of the present disclosure;

FIG. 5 illustrates a flow diagram for collaborative processing of a third feature graph with multiple decoding layers according to an exemplary embodiment of the present disclosure;

FIG. 6 shows a flow chart of a method of training a neural network according to an exemplary embodiment of the present disclosure;

FIG. 7 shows a block diagram of a neural network, according to an example embodiment of the present disclosure;

FIG. 8 shows a block diagram of a neural network, according to an example embodiment of the present disclosure; and

FIG. 9 illustrates a block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

In the present disclosure, unless otherwise specified, the use of the terms "first", "second", etc. to describe various elements is not intended to define a positional relationship, a temporal relationship, or an importance relationship of the elements, and such terms are used only to distinguish one element from another. In some examples, a first element and a second element may refer to the same instance of the element, and in some cases, based on the context, they may also refer to different instances.

The terminology used in the description of the various described examples in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element may be one or a plurality of. Furthermore, the term "and/or" as used in this disclosure is intended to encompass any and all possible combinations of the listed items.

In the related art, the existing semantic segmentation network model has a poor effect when performing portrait segmentation, and particularly when performing semantic segmentation on a portrait image with a small size, detail information is seriously lost, and edge accuracy is very low.

In order to solve the above problem, according to the present disclosure, the first feature maps of multiple scales are processed for multiple times in the encoding stage, so that a second feature map of a target size including the size of the deepest layer feature can be generated for each first feature map, and then the deepest layer semantic feature is fused with the second feature map of the size before upsampling, thereby enriching semantic information and improving the precision of an image processing result.

Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

Fig. 1 illustrates a schematic diagram of an exemplary system 100 in which various methods and apparatus described herein may be implemented in accordance with embodiments of the present disclosure. Referring to fig. 1, the system 100 includes one or

more client devices

101, 102, 103, 104, 105, and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120.

Client devices

101, 102, 103, 104, 105, and 106 may be configured to execute one or more applications.

In an embodiment of the present disclosure, the server 120 may run one or more services or software applications that enable the image processing method or the training method of the neural network to be performed.

In some embodiments, the server 120 may also provide other services or software applications that may include non-virtual environments and virtual environments. In certain embodiments, these services may be provided as web-based services or cloud services, for example, provided to users of

client devices

101, 102, 103, 104, 105, and/or 106 under a software as a service (SaaS) model.

In the configuration shown in fig. 1, server 120 may include one or more components that implement the functions performed by server 120. These components may include software components, hardware components, or a combination thereof, which may be executed by one or more processors. A user operating a

client device

101, 102, 103, 104, 105, and/or 106 may, in turn, utilize one or more client applications to interact with the server 120 to take advantage of the services provided by these components. It should be understood that a variety of different system configurations are possible, which may differ from system 100. Accordingly, fig. 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

The user may use

client devices

101, 102, 103, 104, 105, and/or 106 to upload pre-processed images to be processed or view processed images. The client device may provide an interface that enables a user of the client device to interact with the client device, e.g., the client may obtain the image to be processed through the capture device and send the preview image to the server. The client device may also output information to the user via the interface, e.g., the client may output images processed by the server to the user. Although fig. 1 depicts only six client devices, those skilled in the art will appreciate that any number of client devices may be supported by the present disclosure.

Client devices

101, 102, 103, 104, 105, and/or 106 may include various types of computer devices, such as portable handheld devices, general purpose computers (such as personal computers and laptops), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors or other sensing devices, and so forth. These computer devices may run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, Linux, or Linux-like operating systems (e.g., GOOGLE Chrome OS); or include various Mobile operating systems such as MICROSOFT Windows Mobile OS, iOS, Windows Phone, Android. Portable handheld devices may include cellular telephones, smart phones, tablets, Personal Digital Assistants (PDAs), and the like. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. The gaming system may include a variety of handheld gaming devices, internet-enabled gaming devices, and the like. The client device is capable of executing a variety of different applications, such as various Internet-related applications, communication applications (e.g., email applications), Short Message Service (SMS) applications, and may use a variety of communication protocols.

Network 110 may be any type of network known to those skilled in the art that may support data communications using any of a variety of available protocols, including but not limited to TCP/IP, SNA, IPX, etc. By way of example only, one or more networks 110 may be a Local Area Network (LAN), an ethernet-based network, a token ring, a Wide Area Network (WAN), the internet, a virtual network, a Virtual Private Network (VPN), an intranet, an extranet, a Public Switched Telephone Network (PSTN), an infrared network, a wireless network (e.g., bluetooth, WIFI), and/or any combination of these and/or other networks.

The server 120 may include one or more general purpose computers, special purpose server computers (e.g., PC (personal computer) servers, UNIX servers, mid-end servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and/or combination. The server 120 may include one or more virtual machines running a virtual operating system, or other computing architecture involving virtualization (e.g., one or more flexible pools of logical storage that may be virtualized to maintain virtual storage for the server). In various embodiments, the server 120 may run one or more services or software applications that provide the functionality described below.

The computing units in server 120 may run one or more operating systems including any of the operating systems described above, as well as any commercially available server operating systems. The server 120 may also run any of a variety of additional server applications and/or middle tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, and the like.

In some implementations, the server 120 may include one or more applications to analyze and consolidate data feeds and/or event updates received from users of the

client devices

101, 102, 103, 104, 105, and 106. Server 120 may also include one or more applications to display data feeds and/or real-time events via one or more display devices of

client devices

101, 102, 103, 104, 105, and 106.

In some embodiments, the server 120 may be a server of a distributed system, or a server incorporating a blockchain. The server 120 may also be a cloud server, or a smart cloud computing server or a smart cloud host with artificial intelligence technology. The cloud Server is a host product in a cloud computing service system, and is used for solving the defects of high management difficulty and weak service expansibility in the traditional physical host and Virtual Private Server (VPS) service.

The system 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store information such as audio files and video files. The data store 130 may reside in various locations. For example, the data store used by the server 120 may be local to the server 120, or may be remote from the server 120 and may communicate with the server 120 via a network-based or dedicated connection. The data store 130 may be of different types. In certain embodiments, the data store used by the server 120 may be a database, such as a relational database. One or more of these databases may store, update, and retrieve data to and from the database in response to the command.

In some embodiments, one or more of the databases 130 may also be used by applications to store application data. The databases used by the application may be different types of databases, such as key-value stores, object stores, or regular stores supported by a file system.

The system 100 of fig. 1 may be configured and operated in various ways to enable application of the various methods and apparatus described in accordance with the present disclosure.

According to an aspect of the present disclosure, there is provided an image processing method. As shown in fig. 2, the image processing method includes: step S201, performing cooperative processing on an image to be processed by utilizing a plurality of coding layers to obtain a plurality of first feature maps with different sizes; step S202, performing feature extraction on each first feature map in the plurality of first feature maps to obtain at least one second feature map with a target size, wherein the target size comprises the size of the first feature map with the minimum size in the plurality of first feature maps; step S203, performing feature extraction on the first feature map with the minimum size to obtain a third feature map which is different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third feature map has the same size as the first feature map with the minimum size; step S204, determining at least one first feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps based on the size of the third feature map; step S205, fusing at least one first feature graph to be fused with the third feature graph to obtain a fourth feature graph; and step S206, carrying out cooperative processing on the fourth feature map by utilizing a plurality of decoding layers to obtain an image processing result.

Therefore, the first feature maps with multiple scales are processed for multiple times in the encoding stage, the second feature map with the size of the deepest feature under the target size can be generated for each first feature map, and then the deepest semantic feature is fused with the second feature map under the size before being subjected to up-sampling, so that semantic information is enriched, and the precision of an image processing result is improved.

According to some embodiments, the image to be processed may be an image including a portrait, and the image processing result may be a portrait segmentation result. The image to be processed may also be a frame in a video. In other words, the above-described image processing method may be a method of performing portrait segmentation on a video including a portrait. The steps and effects of the image processing method will be described in the present disclosure by using the human image segmentation task as an example, but the method is not intended to be limited, and a person skilled in the art can apply the method of the present disclosure to other image processing methods, and all the methods are within the scope of the present disclosure.

The image to be processed including the portrait may have a smaller size, such as 256 × 256 or 192 × 192, or a size smaller than a preset size, compared to other types of images. Therefore, when it is determined that the image to be processed is an image satisfying the preset condition, it may be processed using the image processing method of the present disclosure.

The multiple coding layers can perform operations such as convolution, activation, pooling, full connection, down sampling and the like on the image to be processed so as to extract image information of different semantic levels in the image to be processed. In some embodiments, an existing neural network for encoding an image may be used as a plurality of encoding layers, and parameters and structures of the encoding layers may be designed by themselves, which is not limited herein.

According to some embodiments, the plurality of encoding layers may include a plurality of down-sampling layers that down-sample the image to be processed step by step. The down-sampling layer may be, for example, a pooling layer or a convolution layer. It will be appreciated that other layers may also be included between the two downsampled layers, for example, a convolutional layer that may not be downsampled may be included to further extract semantic information at the current size. The first feature map may be, for example, a feature map obtained after the last downsampling, that is, a feature map including the deepest semantic information.

According to some embodiments, when convolutional layers are included in the plurality of coding layers, the convolutional layers may use separable convolutions to reduce the number of parameters of the neural network, increase the training speed of the neural network, and improve the performance of the neural network. In some embodiments, separable convolutions may be used in convolutional layers without downsampling, thereby enabling a reduction in the number of parameters of the neural network while preserving as much detail information as possible.

In some embodiments, the plurality of down-sampling layers included in the plurality of coding layers may include four down-sampling layers of magnification 2. That is, in the encoding process, feature maps of 1/2, 1/4, 1/8, 1/16 and other four sizes of the size of the image to be processed can be obtained, and the first feature map may be a feature map of 1/16 size, for example. Through experiments, the loss of detail information may be caused by using the down-sampling layer with the multiplying power of less than four and 2, so that the accuracy of the portrait segmentation result is influenced, the improvement brought by using the down-sampling layer with the multiplying power of more than four and 2 is very limited, the parameter quantity is increased, and the performance of the neural network is reduced.

According to some embodiments, as shown in fig. 3, the step S202 of performing feature extraction on each of the plurality of first feature maps may include: step S301, the first feature map is processed by using the first void convolution layer to obtain a corresponding second feature map, wherein the size of the corresponding second feature map is smaller than that of the first feature map. Therefore, by adopting the hole convolution when generating the second feature map with a smaller size based on the first feature map, the receptive field can be increased, and the edge accuracy of the portrait segmentation result can be improved.

The hole convolution is a special convolution that increases the receptive field by interval sampling or the like without changing the parameter number and the amount of computation. The hole convolution can utilize the hyper-parametric expansion rate to determine the distance between adjacent sampling points, thereby adjusting the range of the receptive field.

In accordance with some embodiments, the expansion rate of the first hole convolution layer is determined based on a stride step size of the first hole convolution layer. Thus, by determining the expansion rate based on the ratio of the feature map sizes before and after downsampling (i.e., the stride step size), the edge accuracy of subsequent portrait segmentation can be improved to the greatest extent. In one exemplary embodiment, the expansion rate of the first hole convolution layer may be equal to the stride step size. For example, the expansion ratio of the first hole convolution layer for generating 1/4 the second profile based on the 1/2 first profile may be 2, the expansion ratio of the first hole convolution layer for generating 1/8 the second profile based on 1/2 the first profile may be 4, and the expansion ratio of the first hole convolution layer for generating 1/16 the second profile based on 1/2 the first profile may be 8.

According to some embodiments, the step S202 of performing feature extraction on each of the plurality of first feature maps may further include: step S302, utilizing a second upsampling layer to upsample the first characteristic diagram; and step S303, processing the first feature map after upsampling by using at least one first convolution layer to obtain a corresponding second feature map, wherein the size of the corresponding second feature map is larger than that of the first feature map.

According to some embodiments, the step S202 of performing feature extraction on each of the plurality of first feature maps may further include: step S304, processing the first feature map corresponding to at least one second convolution layer to obtain a corresponding second feature map, wherein the size of the corresponding second feature map is equal to the size of the first feature map.

According to some embodiments, as shown in fig. 4, the step S203 of extracting the features of the first feature map with the smallest size may include: step S401, processing the first characteristic diagram with the minimum size by utilizing a plurality of second void convolution layers connected in parallel to obtain a plurality of fourth characteristic diagrams; s402, fusing the fourth feature maps to obtain a fifth feature map; and step S403, processing the fifth characteristic diagram by using the plurality of serially connected third void convolution layers to obtain a third characteristic diagram. Therefore, the deepest semantic features are processed by using the parallel and serial hole convolutions, information of different receptive field sizes can be further fused under deep semantics, semantic information included by the feature map is enriched, and therefore the accuracy of the portrait segmentation result can be improved.

In some embodiments, the first feature map may be processed using convolution of four holes with respective expansion rates of 1, 2, 4, and 6 to obtain four fourth feature maps. These fourth feature maps may have the same dimensions. It is understood that, those skilled in the art can set the number of the second void convolution layers and the expansion rate thereof according to the requirement, and the setting is not limited herein.

In some embodiments, the plurality of second hole convolution layers may be further connected in parallel with other neural network layers, such as a pooling layer, to further enrich semantic information included in the output plurality of fourth feature maps.

According to some embodiments, the step S402 of fusing the plurality of fourth feature maps may be, for example, directly splicing the plurality of fourth feature maps, or performing point-to-point addition on the plurality of fourth feature maps, or performing further processing using a 1 × 1 convolution kernel on the basis of the foregoing manner, or performing fusion using other manners, which is not limited herein. In an exemplary embodiment, the plurality of fourth feature maps are fused by using a direct stitching manner, so as to fully retain image information acquired through different receptive fields at the deepest semantic level.

After obtaining the fifth feature map, the fifth feature map may be processed by a plurality of third hole convolution layers connected in series, respectively, to obtain a third feature map.

In some embodiments, the fifth feature map may be processed using convolution of three holes with respective expansion rates of 2, 4, and 6 to obtain a third feature map. The third feature map may have the same size as the fifth feature map. It is understood that, those skilled in the art can set the number of the third void convolution layers and the expansion rate thereof according to the requirement, and the setting is not limited herein.

After the third feature map is obtained, at least one first feature map to be fused may be determined in at least one second feature map corresponding to each of the plurality of first feature maps based on the size of the third feature map. In some embodiments, the second feature map having the same size as the third feature map may be determined as the first feature map to be fused, so that these first feature maps to be fused are fused with the third feature map to obtain the fourth feature map.

After the fourth feature map is obtained, it may be processed using a plurality of decoders to obtain an image processing result.

The multiple decoding layers may perform operations such as upsampling, deconvolution, activation, pooling, full concatenation, etc. on the feature map to decode image information in the feature map into an image. In some embodiments, an existing neural network for decoding an image feature may be used as a plurality of decoding layers, and parameters and structures of the decoding layers may also be designed by themselves, which is not limited herein.

According to some embodiments, the plurality of encoding layers may include a plurality of upsampling layers that upsample the third feature map step by step. The upsampling layer may be, for example, an deconvolution layer. In one exemplary embodiment, the plurality of upsampling layers may include, for example, three upsampling layers of magnification 2, thereby enabling a step-wise upsampling of the third feature map of size 1/16 into feature maps of sizes 1/8, 1/4, and 1/2. Through gradual up-sampling, semantic information under each scale can be reserved as much as possible, and particularly, under the condition that an image to be processed is small, detailed information can be fully reserved so as to improve the edge precision of a portrait segmentation result.

According to some embodiments, the plurality of decoding layers comprises a plurality of upsampling layers which upsample the fourth feature map step by step. As shown in fig. 5, the step S206 of performing cooperative processing on the third feature map by using a plurality of decoding layers to obtain an image processing result may include: step S501, each up-sampling layer in a plurality of up-sampling layers is used for up-sampling the characteristic graph received by the up-sampling layer to obtain a second characteristic graph to be fused; step S502, determining at least one third feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps based on the size of the second feature map to be fused; and step S503, fusing the second feature graph to be fused and at least one third feature graph to be fused. The operations of steps S502 and S503 in fig. 5 are similar to the operations of steps S204 and S205 in fig. 2, and are not repeated herein.

According to some embodiments, the number of the at least one second feature map corresponding to each first feature map is the same as the number of the plurality of first feature maps, and the size of the corresponding at least one second feature map is associated with the size of the plurality of first feature maps. Therefore, when the up-sampling is carried out step by step, the feature map obtained after each up-sampling is fused with the second feature map generated in the encoding stage, so that the semantic information is enriched to the maximum extent.

According to some embodiments, the plurality of decoding layers may further include restoring the feature map to a network structure having a size of an image to be processed to obtain an image processing result.

According to another aspect of the present disclosure, a method of training a neural network is provided. The neural network includes a plurality of encoding layers, a plurality of first feature extraction sub-networks following the plurality of encoding layers, a second feature extraction sub-network following the plurality of encoding layers, and a plurality of decoding layers following the second feature extraction sub-network. As shown in fig. 6, the method includes: s601, obtaining a sample image and a real label of the sample image; in step S602, a plurality of coding layers are used to perform cooperative processing on a sample image to obtain a plurality of first sample feature maps with different sizes; step S603, for each of the plurality of first sample feature maps, processing the first sample feature map by using at least one of the plurality of first feature extraction subnetworks corresponding to the first sample feature map to obtain at least one second sample feature map having a target size, where the target size includes a size of a first feature map having a smallest size among the plurality of first feature maps; step S604, processing the first sample feature map with the minimum size by using a second feature extraction sub-network to obtain a third sample feature map which is different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third sample feature map has the same size as the first sample feature map with the minimum size; step S605, determining at least one first to-be-fused sample feature map in at least one second sample feature map corresponding to each of the plurality of first sample feature maps based on the size of the third sample feature map; step S606, fusing at least one first sample feature map to be fused with the third sample feature map to obtain a fourth sample feature map; step S607, the fourth sample feature map is processed cooperatively by a plurality of decoding layers to obtain a prediction label of the sample image; and step S608, calculating a loss value based on the predicted label and the real label to adjust parameters of the neural network.

It is understood that the operations of steps S602-S607 in fig. 6 are similar to the operations of steps S201-S206 in fig. 2, and are not repeated herein.

Therefore, the first feature maps with multiple scales are processed for multiple times in the encoding stage, the second feature map with the size of the deepest feature under the target size can be generated for each first feature map, and then the deepest semantic features are fused with the second feature map under the size before being sampled, so that the semantic information is enriched, and the accuracy of image processing of the neural network obtained by training through the training method is improved.

According to another aspect of the present disclosure, a neural network is provided. As shown in fig. 7, the neural network 700 includes: a plurality of encoding layers 710 configured to perform a cooperative processing on the image to be processed 702 by the plurality of encoding layers to output a plurality of first feature maps having different sizes; a first feature extraction subnetwork 720, subsequent to the plurality of encoding layers, corresponding to each of the plurality of first feature maps, configured to process the first feature map to output at least one second feature map 760 having a target size, the target size comprising a size of a smallest-sized first feature map of the plurality of first feature maps; a second feature extraction sub-network 730 following the plurality of encoding layers, configured to process the first feature map with the smallest size to output a third feature map different from each of at least one second feature map corresponding to each of the plurality of first feature maps, where the third feature map is the same size as the first feature map with the smallest size; a first fusion layer 740 behind the second feature extraction sub-network, configured to fuse at least one first feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps with the third feature map to output a fourth feature map, wherein the at least one first feature map to be fused is determined based on a size of the fourth feature map; and a plurality of decoding layers 750 after the second feature extraction sub-network configured to perform cooperative processing on the third feature map to output an image processing result. The operations of the

layers

710, 730, and 750 in the neural network 700 are similar to the operations of the steps S201-S203, and S206 in fig. 2, respectively, and the operation of the layer 740 is similar to the steps S204 and S205 in fig. 2, and are not repeated herein.

According to some embodiments, as shown in FIG. 8, the plurality of decode layers 850 may include: a plurality of upsampling layers 852 that upsample the fourth feature map step by step, each of the plurality of upsampling layers being configured to upsample the received feature map to obtain a second feature map to be fused; and a plurality of second fusion layers 854 corresponding to the plurality of upsampling layers, each of the plurality of second fusion layers being configured to fuse the received second feature map to be fused and at least one third feature map to be fused to output, the at least one third feature map to be fused being determined in the at least one second feature map corresponding to each of the plurality of first feature maps based on a size of the second feature map to be fused. The to-be-processed image 802, the image processing result 804, the plurality of encoding layers 810, the first feature extraction sub-network 820, the second feature extraction sub-network 830, the first fusion layer 840, and the plurality of decoding layers 850 in fig. 8 are similar to the corresponding network structure in fig. 7, and are not repeated here.

According to some embodiments, the first feature extraction subnetwork may comprise: at least one first hole convolution layer configured to perform hole convolution on a corresponding first feature map to output a corresponding second feature map, wherein the size of the corresponding second feature map is smaller than that of the first feature map.

According to some embodiments, the first feature extraction subnetwork may comprise: a second upsampling layer configured to upsample the corresponding first feature map; and at least one first convolution layer configured to convolve the corresponding first feature maps after upsampling to output corresponding second feature maps, wherein the size of the corresponding second feature maps is larger than that of the first feature maps.

According to some embodiments, the first feature extraction subnetwork may comprise: at least one second convolution layer configured to convolve the corresponding first feature map to output a corresponding second feature map, wherein the size of the corresponding second feature map is equal to the size of the first feature map.

According to some embodiments, the second feature extraction sub-network may comprise: a plurality of second hole convolution layers connected in parallel, each of the plurality of second hole convolution layers being configured to perform hole convolution on the first feature map having the smallest size to output a fourth feature map; a fusion layer subsequent to the plurality of second hole convolution layers, configured to fuse the plurality of fourth feature maps output by the plurality of second hole convolution layers to output a fifth feature map; the plurality of third hole convolution layers connected in series after the fusion layer are configured to successively perform hole convolution for a plurality of times on the fifth feature map to output a third feature map.

According to some embodiments, the number of the at least one second feature map output by the first feature extraction sub-network may be the same as the number of the plurality of first feature maps, and the size of the at least one second feature map output may be associated with the size of the plurality of first feature maps.

In the technical scheme of the disclosure, the processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the personal information of the related user all accord with the regulations of related laws and regulations, and do not violate the common customs of public order.

According to an embodiment of the present disclosure, there is also provided an electronic device, a readable storage medium, and a computer program product.

Referring to fig. 9, a block diagram of a structure of an electronic device 800, which may be a server or a client of the present disclosure, which is an example of a hardware device that may be applied to aspects of the present disclosure, will now be described. Electronic device is intended to represent various forms of digital electronic computer devices, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 9, the apparatus 900 includes a computing unit 901, which can perform various appropriate actions and processes in accordance with a computer program stored in a Read Only Memory (ROM)902 or a computer program loaded from a storage unit 908 into a Random Access Memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The calculation unit 901, ROM 902, and RAM 903 are connected to each other via a bus 904. An input/output (I/O) interface 905 is also connected to bus 904.

A number of components in the device 900 are connected to the I/O interface 905, including: an input unit 906, an output unit 907, a storage unit 908, and a communication unit 909. The input unit 906 may be any type of device capable of inputting information to the device 900, and the input unit 906 may receive input numeric or character information and generate key signal inputs related to user settings and/or function controls of the electronic device, and may include, but is not limited to, a mouse, a keyboard, a touch screen, a track pad, a track ball, a joystick, a microphone, and/or a remote control. Output unit 907 may be any type of device capable of presenting information and may include, but is not limited to, a display, speakers, a video/audio output terminal, a vibrator, and/or a printer. Storage unit 908 may include, but is not limited to, a magnetic disk, an optical disk. The communication unit 909 allows the device 900 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers and/or chipsets, such as bluetooth (TM) devices, 802.11 devices, WiFi devices, WiMax devices, cellular communication devices, and/or the like.

The computing unit 901 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 901 performs the respective methods and processes described above, such as an image processing method or a training method of a neural network. For example, in some embodiments, the image processing method or the training method of the neural network may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 908. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 900 via ROM 902 and/or communications unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the image processing method or the training method of the neural network described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured by any other suitable means (e.g. by means of firmware) to perform an image processing method or a training method of a neural network.

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The Server can be a cloud Server, also called a cloud computing Server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service ("Virtual Private Server", or simply "VPS"). The server may also be a server of a distributed system, or a server incorporating a blockchain.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be performed in parallel, sequentially or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.

Although embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it is to be understood that the above-described methods, systems and apparatus are merely exemplary embodiments or examples and that the scope of the present invention is not limited by these embodiments or examples, but only by the claims as issued and their equivalents. Various elements in the embodiments or examples may be omitted or may be replaced with equivalents thereof. Further, the steps may be performed in an order different from that described in the present disclosure. Further, various elements in the embodiments or examples may be combined in various ways. It is important that as technology evolves, many of the elements described herein may be replaced with equivalent elements that appear after the present disclosure.

Claims

1. An image processing method comprising:

performing cooperative processing on an image to be processed by utilizing a plurality of coding layers to obtain a plurality of first feature maps with different sizes;

performing feature extraction on each of the plurality of first feature maps to obtain at least one second feature map having a target size, wherein the target size comprises the size of the first feature map with the smallest size in the plurality of first feature maps;

performing feature extraction on the first feature map with the minimum size to obtain a third feature map which is different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third feature map has the same size as the first feature map with the minimum size;

determining at least one first feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps based on the size of the third feature map;

fusing the at least one first feature map to be fused with the third feature map to obtain a fourth feature map; and

and performing cooperative processing on the fourth feature map by using a plurality of decoding layers to obtain an image processing result.

2. The method of claim 1, wherein the plurality of decoding layers includes a plurality of upsampling layers that upsample the fourth feature map step by step,

wherein the performing the cooperative processing on the third feature map by using the plurality of decoding layers to obtain an image processing result comprises:

utilizing each up-sampling layer in the plurality of up-sampling layers to up-sample the characteristic diagram received by the up-sampling layer so as to obtain a second characteristic diagram to be fused;

determining at least one third feature map to be fused in at least one second feature map corresponding to each of the plurality of first feature maps based on the size of the second feature map to be fused; and

and fusing the second feature map to be fused and the at least one third feature map to be fused.

3. The method of claim 1 or 2, wherein the feature extracting each of the plurality of first feature maps to obtain at least one second feature map having different sizes comprises:

and processing the first characteristic diagram by utilizing the first void convolution layer to obtain a corresponding second characteristic diagram, wherein the size of the corresponding second characteristic diagram is smaller than that of the first characteristic diagram.

4. The method of claim 3, wherein a rate of expansion of the first hole convolution layer is determined based on a stride step size of the first hole convolution layer.

5. The method of any one of claims 1-4, wherein said feature extracting each of the plurality of first feature maps to obtain at least one second feature map having a different size comprises:

upsampling the first feature map using a second upsampling layer; and

and processing the first feature map after upsampling by utilizing at least one first convolution layer to obtain a corresponding second feature map, wherein the size of the corresponding second feature map is larger than that of the first feature map.

6. The method of any one of claims 1-5, wherein the feature extracting each of the plurality of first feature maps to obtain at least one second feature map having a different size comprises:

and processing the corresponding first feature map by using at least one second convolution layer to obtain a corresponding second feature map, wherein the size of the corresponding second feature map is equal to that of the first feature map.

7. The method of any one of claims 1-6, wherein the feature extracting a first feature map of the plurality of first feature maps of a smallest size comprises:

processing the first characteristic diagram with the minimum size by utilizing a plurality of second void convolution layers connected in parallel to obtain a plurality of fourth characteristic diagrams;

fusing the plurality of fourth feature maps to obtain a fifth feature map; and

and processing the fifth feature map by utilizing a plurality of third void convolution layers connected in series to obtain the third feature map.

8. The method according to any one of claims 1-7, wherein the number of the at least one second feature map corresponding to each of the first feature maps is the same as the number of the plurality of first feature maps, and the size of the corresponding at least one second feature map is associated with the size of the plurality of first feature maps.

9. The method of any one of claims 1-8, wherein the plurality of coding layers comprises four 2-magnification downsampling layers, and the plurality of first upsampling layers comprises three 2-magnification upsampling layers.

10. The method of any of claims 1-9, wherein the plurality of encoded layers and the plurality of decoded layers comprise at least one convolutional layer using separable convolution.

11. The method according to any one of claims 1-10, wherein the image to be processed is an image comprising a portrait and the image processing result is a portrait segmentation result.

12. A method of training a neural network, the neural network comprising a plurality of encoding layers, a plurality of first feature extraction sub-networks following the plurality of encoding layers, a second feature extraction sub-network following the plurality of encoding layers, and a plurality of decoding layers following the second feature extraction sub-network, the method comprising:

acquiring a sample image and a real label of the sample image;

carrying out cooperative processing on the sample image by utilizing a plurality of coding layers to obtain a plurality of first sample characteristic graphs with different sizes;

for each first sample feature map in the plurality of first sample feature maps, processing the first sample feature map by using at least one first feature extraction sub-network corresponding to the first sample feature map in the plurality of first feature extraction sub-networks to obtain at least one second sample feature map with a target size, wherein the target size comprises the size of the first feature map with the smallest size in the plurality of first feature maps;

processing the first sample feature map with the minimum size by using a second feature extraction sub-network to obtain a third sample feature map which is different from at least one second feature map corresponding to each of the plurality of first feature maps, wherein the third sample feature map has the same size as the first sample feature map with the minimum size;

determining at least one first sample feature map to be fused in at least one second sample feature map corresponding to each of the plurality of first sample feature maps based on the size of the third sample feature map;

fusing the at least one first sample feature map to be fused with the third sample feature map to obtain a fourth sample feature map;

performing cooperative processing on the fourth sample feature map by using a plurality of decoding layers to obtain a prediction label of the sample image; and

calculating a loss value based on the predicted tag and the real tag to adjust a parameter of the neural network.

13. A neural network for image processing, comprising:

the image processing device comprises a plurality of encoding layers, a plurality of image processing units and a plurality of image processing units, wherein the encoding layers are configured to carry out cooperative processing on images to be processed by the encoding layers so as to output a plurality of first feature maps with different sizes;

a first feature extraction sub-network, subsequent to the plurality of encoding layers, corresponding to each of the plurality of first feature maps, configured to process the first feature map to output at least one second feature map having a target size, the target size including a size of a smallest-sized first feature map of the plurality of first feature maps;

a second feature extraction sub-network after the plurality of coding layers, configured to process the first feature map with the smallest size to output a third feature map different from at least one second feature map corresponding to each of the plurality of first feature maps, where the third feature map has the same size as the first feature map with the smallest size;

a first fusion layer following the second feature extraction sub-network, configured to fuse at least one first feature map to be fused from among at least one second feature map corresponding to each of the plurality of first feature maps with the third feature map to output a fourth feature map, wherein the at least one first feature map to be fused is determined based on a size of the third feature map; and

a plurality of decoding layers behind the second feature extraction sub-network configured to perform cooperative processing on the fourth feature map to output an image processing result.

14. The neural network of claim 13, wherein the plurality of decoding layers comprises:

a plurality of upsampling layers that upsample the fourth feature map step by step, each of the plurality of upsampling layers being configured to upsample the received feature map to obtain a second feature map to be fused; and

a plurality of second fusion layers corresponding to the plurality of upsampling layers, each of the plurality of second fusion layers being configured to fuse the received second feature map to be fused and at least one third feature map to be fused to output, the at least one third feature map to be fused being determined in at least one second feature map corresponding to each of the plurality of first feature maps based on a size of the second feature map to be fused.

15. A neural network as claimed in claim 13 or 14, wherein the first feature extraction sub-network comprises:

at least one first hole convolution layer configured to perform hole convolution on a corresponding first feature map to output a corresponding second feature map, wherein the size of the corresponding second feature map is smaller than that of the first feature map.

16. A neural network as claimed in any one of claims 13 to 15, wherein the first feature extraction sub-network comprises:

a second upsampling layer configured to upsample the corresponding first feature map; and

at least one first convolution layer configured to convolve the corresponding first feature map after upsampling to output a corresponding second feature map, wherein the size of the corresponding second feature map is larger than that of the first feature map.

17. A neural network as claimed in any one of claims 13 to 16, wherein the first feature extraction sub-network comprises:

at least one second convolution layer configured to convolve the corresponding first feature map to output a corresponding second feature map, wherein the size of the corresponding second feature map is equal to the size of the first feature map.

18. A neural network as claimed in any one of claims 13 to 17, wherein the second feature extraction sub-network comprises:

a plurality of second hole convolution layers connected in parallel, each of the plurality of second hole convolution layers being configured to perform hole convolution on the first feature map of smallest size to output a fourth feature map;

a fusion layer subsequent to the plurality of second hole convolution layers, configured to fuse a plurality of fourth feature maps output by the plurality of second hole convolution layers to output a fifth feature map;

a plurality of third hole convolution layers connected in series after the fusion layer is configured to successively perform a plurality of hole convolutions on the fifth feature map to output the third feature map.

19. The neural network of any one of claims 13-18, wherein the first feature extraction sub-network outputs the same number of at least one second feature map as the plurality of first feature maps, the size of the at least one second feature map of the output being associated with the size of the plurality of first feature maps.

20. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein

The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-12.

21. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-12.

22. A computer program product comprising a computer program, wherein the computer program realizes the method of any one of claims 1-12 when executed by a processor.