Image processing model training method and device, image recommendation method and device, equipment and medium

Abstract

The invention provides an image processing model training method and device, an image recommendation method and device, equipment and a medium, relates to the technical field of artificial intelligence, in particular to the technical fields of natural language processing, deep learning and the like, and can be used for application scenes such as generative retrieval, document intelligent editing, intelligent assistants, virtual assistants, intelligent e-commerce and the like. The training method comprises the steps of obtaining a first sample image with annotation data and a plurality of unannotated second sample images, wherein the annotation data comprises a classification result and an interpretation text for a plurality of preset evaluation dimensions; performing fine tuning training on the multi-modal visual understanding large model by using the first sample image and the corresponding annotation data; classifying the plurality of second sample images by using the multi-modal visual understanding large model subjected to fine tuning training; constructing a plurality of image groups, wherein each image group comprises at least two images with different categories in the same evaluation dimension; and training the lightweight image processing model based on the plurality of image groups.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, in particular to the technical field of natural language processing and deep learning, which can be used in generative retrieval, intelligent editing of documents, intelligent assistants, virtual assistants, intelligent e-commerce and other application scenarios, and specifically relates to a training method of an image processing model, an image recommendation method based on a large model, a training device of an image processing model, an image recommendation device based on a large model, an electronic device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] Artificial intelligence is a discipline that studies enabling computers to simulate some thinking processes and intelligent behaviors of humans (such as learning, reasoning, thinking, planning, etc.), which has both hardware and software technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing, etc.; artificial intelligence software technologies mainly include natural language processing technology, computer vision technology, speech recognition technology, and machine learning / deep learning, big data processing technology, knowledge graph technology, etc.

[0003] The methods described in this section can not necessarily be the methods that have been previously conceived or adopted. Unless otherwise indicated, nothing in this section should be assumed to be prior art merely because of its inclusion in this section. Similarly, unless otherwise indicated, nothing in this section should be assumed to have been admitted prior to the filing date of the present application. SUMMARY

[0004] The present disclosure provides a training method of an image processing model, an image recommendation method based on a large model, a training device of an image processing model, an image recommendation device based on a large model, an electronic device, a computer readable storage medium and a computer program product.

[0005] According to an aspect of the present disclosure, a training method of an image processing model is provided, including: obtaining a first sample image with labeled data and a plurality of second sample images without labels, the labeled data of the first sample image including true value classification results and true value explanation texts of a plurality of preset evaluation dimensions; generating a first prediction result of the first sample image by using a multimodal visual understanding large model, the prediction result including prediction classification results and prediction explanation texts of the plurality of evaluation dimensions; performing fine-tuning training on the multimodal visual understanding large model based on the first prediction result and the labeled data; classifying the plurality of second sample images based on the plurality of evaluation dimensions by using the multimodal visual understanding large model after the fine-tuning training, to obtain a second prediction result of the plurality of second sample images; constructing a plurality of image groups based on the second prediction result, each image group including at least two images of different categories in the same evaluation dimension; and training a lightweight image processing model based on the plurality of image groups, to obtain a trained image processing model.

[0006] According to another aspect of the present disclosure, a large model-based image recommendation method is provided, including: inputting a plurality of user historical interest images of a target user into a trained image processing model, to obtain a prediction score of each user historical interest image for a plurality of preset evaluation dimensions, wherein the trained image processing model is trained by using the above training method; determining a preference score of the target user for the same evaluation dimension based on the prediction scores of the plurality of user historical interest images for the same evaluation dimension; and determining a target image for recommendation to the target user based on the preference scores of the target user for the plurality of evaluation dimensions.

[0007] According to another aspect of the present disclosure, a training device of an image processing model is provided, including: an obtaining unit configured to obtain a first sample image with labeled data and a plurality of second sample images without labels, the labeled data of the first sample image including true value classification results and true value explanation texts of a plurality of preset evaluation dimensions; a first generating unit configured to generate a first prediction result of the first sample image by using a multimodal visual understanding large model, the prediction result including prediction classification results and prediction explanation texts of the plurality of evaluation dimensions; a first training unit configured to perform fine-tuning training on the multimodal visual understanding large model based on the first prediction result and the labeled data; a second generating unit configured to classify the plurality of second sample images based on the plurality of evaluation dimensions by using the multimodal visual understanding large model after the fine-tuning training, to obtain a second prediction result of the plurality of second sample images; a constructing unit configured to construct a plurality of image groups based on the second prediction result, each image group including at least two images of different categories in the same evaluation dimension; and a second training unit configured to train a lightweight image processing model based on the plurality of image groups, to obtain a trained image processing model.

[0008] According to another aspect of the present disclosure, there is provided a large model-based image recommendation apparatus, comprising: a scoring unit configured to input a plurality of user historical interest images of a target user into a trained image processing model to obtain a predicted score of each user historical interest image for a plurality of preset evaluation dimensions, wherein the trained image processing model is trained by the training apparatus described above; a first determination unit configured to determine a preference score of the target user for an evaluation dimension based on the predicted scores of the plurality of user historical interest images for the same evaluation dimension; and a second determination unit configured to determine a target image for recommendation to the target user based on the preference scores of the target user for the plurality of evaluation dimensions.

[0009] According to another aspect of the present disclosure, there is provided an electronic device, comprising: at least one processor; and a memory communicatively connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the above method.

[0010] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to perform the above method.

[0011] According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the above method.

[0012] According to one or more embodiments of the present disclosure, the present disclosure fine-tunes a multi-modal visual understanding large model by using the labeled classification results and labeled explanation texts of the first sample images, so that the multi-modal visual understanding large model has the judgment ability for a plurality of preset evaluation dimensions under the constraint of multi-modal information. Further, the fine-tuned multi-modal visual understanding large model is used to classify a plurality of unlabeled second sample images to obtain the classification results of each second sample image for each evaluation dimension. Then, a plurality of image groups are constructed, each image group containing images of different categories of the same evaluation dimension, and a lightweight image processing model is trained using the image groups, so that a high-performance model for generating image evaluation results can be obtained at low cost.

[0013] It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0014] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments and together with the description serve to explain exemplary implementations of the application. The illustrated embodiments are exemplary only and not limiting of the scope of the appended claims. In all

[0015] Figure 1 A schematic diagram illustrating an exemplary system in which various methods described herein can be implemented according to embodiments of the present disclosure is shown;

[0016] Figure 2 A flowchart illustrating a method of training an image processing model according to embodiments of the present disclosure is shown;

[0017] Figure 3 A flowchart illustrating fine-tuning training of a multi-modal visual understanding large model based on a first prediction result and labeled data according to embodiments of the present disclosure is shown;

[0018] Figure 4 A flowchart illustrating training of a lightweight image processing model based on a plurality of image groups to obtain a trained image processing model according to embodiments of the present disclosure is shown;

[0019] Figure 5 A schematic diagram illustrating constructing image pairs and training a lightweight image processing model according to embodiments of the present disclosure is shown;

[0020] Figure 6 A flowchart illustrating a large model-based image recommendation method according to embodiments of the present disclosure is shown;

[0021] Figure 7 A block diagram illustrating a structure of a training apparatus of an image processing model according to embodiments of the present disclosure is shown;

[0022] Figure 8 A block diagram illustrating a structure of a large model-based image recommendation apparatus according to embodiments of the present disclosure is shown; and

[0023] Figure 9 A block diagram illustrating an exemplary electronic device that can be used to implement embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0024] Exemplary embodiments of the present disclosure are described herein with reference to the accompanying drawings, in which various details are set forth to facilitate an understanding of the present disclosure. However, it will be apparent to those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present disclosure. Likewise, the description is not to be construed to limit the scope of the claims. In order to clearly illustrate the present disclosure, the following description will be given with reference to the accompanying drawings.

[0025] In the present disclosure, the terms "first", "second", and the like are used to describe various elements only for the purpose of distinguishing one element from another, and the terms are not intended to imply a relative importance of the elements. In some examples, a first element and a second element can refer to the same instance of the element, and in some cases, they can refer to different instances of the element based on the context of description.

[0026] The terminology used in the description of the various examples in the present disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless specifically defined otherwise, an element can be one or more than one, if the quantity of elements is not specifically limited by the context. Furthermore, the term "and / or" used in the present disclosure encompasses any one and all possible combinations of the listed items.

[0027] In the related art, measuring the visual content perception in the recommendation field is a long-standing problem. At present, there are many models to measure the noise and blur degree of pictures, but this quantification and human language are loosely coupled. When it comes to the feeling or more abstract perception of visual content, the existing model is often difficult to fit.

[0028] To solve the above problems, the present disclosure fine-tunes a multi-modal visual understanding large model by using the labeled classification result and the labeled explanation text of the first sample image, so that the multi-modal visual understanding large model has the judgment ability for the preset multiple evaluation dimensions under the constraint of multi-modal information. Further, the fine-tuned multi-modal visual understanding large model is used to classify the unlabeled multiple second sample images to obtain the classification results of each second sample image for each evaluation dimension. Then, a plurality of image groups are constructed, each image group containing images of different categories of the same evaluation dimension, and a lightweight image processing model is trained using the image groups, so that a high-performance model for generating image evaluation results can be obtained at low cost.

[0029] Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0030] Figure 1 A schematic diagram of an example system 100 in which the various methods and apparatus described herein can be implemented according to embodiments of the present disclosure is shown. Referring to Figure 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. The client devices 101, 102, 103, 104, 105, and 106 can be configured to execute one or more application programs.

[0031] In embodiments of the disclosure, the server 120 can run one or more services or software applications that enable the methods of the disclosure to be performed.

[0032] In certain embodiments, the server 120 can also provide other services or software applications, which can include non-virtual and virtual environments. In certain embodiments, these services can be provided as web-based services or cloud services, for example, to users of the client devices 101, 102, 103, 104, 105, and / or 106 under a software as a service (SaaS) model.

[0033] In Figure 1 In the illustrated configuration, the server 120 can include one or more components that implement the functionality performed by the server 120. These components can include software components that are executable by one or more processors, hardware components, or combinations thereof. Users operating the client devices 101, 102, 103, 104, 105, and / or 106 can in turn utilize one or more client applications to interact with the server 120 to utilize the services provided by the components. It will be appreciated that various different system configurations are possible, which can differ from system 100. Thus, Figure 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

[0034] A user can use the client devices 101, 102, 103, 104, 105, and / or 106 to engage in human-machine interactions. The client devices can provide interfaces that enable users of the client devices to interact with the client devices. The client devices can also output information to the users via the interfaces. Although Figure 1 Only six client devices are depicted, but one of skill in the art will appreciate that the disclosure can support any number of client devices.

[0035] 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, etc. 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 (such as Google Chrome OS); or include various mobile operating systems, such as Microsoft Windows Mobile OS, iOS, Windows Phone, and Android. Portable handheld devices may include cellular phones, smartphones, tablet computers, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. Gaming systems may include various handheld gaming devices, internet-enabled gaming devices, etc. Client devices are capable of executing a variety of different applications, such as various internet-related applications, communication applications (such as email applications), and short message service (SMS) applications, and may use various communication protocols.

[0036] The network 110 may be any type of network known to those skilled in the art that can 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, the 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 blockchain network, 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.

[0037] Server 120 may include one or more general-purpose computers, specialized server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running virtual operating systems, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that may be virtualized to maintain a server's virtual storage device). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0038] The computing units in the server 120 can run one or more operating systems including any of the operating systems described above, as well as any commercially available server operating systems. Server 120 can also run any of a variety of additional server applications and / or mid-tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0039] In some embodiments, the server 120 can include one or more applications to analyze and consolidate data feeds and / or event updates from users of the client devices 101, 102, 103, 104, 105, and / or 106. The server 120 can also include one or more applications to display the data feeds and / or real-time events via one or more display devices of the client devices 101, 102, 103, 104, 105, and / or 106.

[0040] In some embodiments, the server 120 can be a server of a distributed system, or a server combined with a blockchain. The server 120 can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. The cloud server is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and virtual private server (VPS, Virtual Private Server) services.

[0041] The system 100 can also include one or more databases 130. In certain embodiments, these databases can be used to store data and other information. For example, one or more of the databases 130 can be used to store information such as audio files and video files. The databases 130 can reside in a variety of locations. For example, databases used by the server 120 can reside locally to the server 120, or can be remote from the server 120 and can communicate with the server 120 via a network- or application-specific connection. The databases 130 can be of different types. In certain embodiments, databases used by the server 120 can be, for example, relational databases. One or more of these databases can store, update, and retrieve data to and from the databases in response to commands.

[0042] In certain embodiments, one or more of the databases 130 can also be used by applications to store application data. Databases used by applications can be different types of databases, such as key-value stores, object stores, or regular stores backed by file systems.

[0043] Figure 1The system 100 can be configured and operated in various ways to enable the application of various methods and apparatuses described according to the present disclosure.

[0044] According to one aspect of the present disclosure, a method for training an image processing model is provided. As shown in the figure, the method comprises the following steps: Figure 2 S201, obtaining a first sample image with labeled data and a plurality of second sample images without labels, the labeled data of the first sample image comprising true value classification results and true value explanation texts of a plurality of preset evaluation dimensions respectively; S202, generating a first prediction result of the first sample image by using a multimodal visual understanding large model, the prediction result comprising prediction classification results and prediction explanation texts of the plurality of evaluation dimensions respectively; S203, fine-tuning the multimodal visual understanding large model based on the first prediction result and the labeled data; S204, classifying the plurality of second sample images based on the plurality of evaluation dimensions by using the fine-tuned multimodal visual understanding large model to obtain a second prediction result of the plurality of second sample images; S205, constructing a plurality of image groups based on the second prediction result, each image group comprising at least two images of different categories within the same evaluation dimension; and S206, training a lightweight image processing model based on the plurality of image groups to obtain a trained image processing model.

[0045] Thus, by fine-tuning the multimodal visual understanding large model using the labeled classification results and labeled explanation texts of the first sample image, the model is capable of making judgments for the plurality of preset evaluation dimensions under the constraint of multimodal information. Furthermore, by using the fine-tuned multimodal visual understanding large model to classify the plurality of second sample images without labels, classification results of each second sample image for each evaluation dimension are obtained. Then, by constructing a plurality of image groups, each image group containing images of different categories within the same evaluation dimension, and training a lightweight image processing model using the image groups, a high-performance model for generating image evaluation results can be obtained at low cost.

[0046] The labeled data of the first sample image obtained in step S201 can be manually labeled, machine (e.g., large model) labeled, or obtained through other methods or a combination of multiple methods, which is not limited herein. The second sample images can be obtained by any means. The number of second sample images can be much larger than the number of first sample images.

[0047] In some embodiments, the multi-modal visual understanding large model can include various models capable of understanding visual content, such as ERNIE-4.5, Qwen-VL, and the like. The multi-modal visual understanding large model used in the present disclosure is a pre-trained large model, that is, a large model that has been pre-trained to receive, understand, process, and / or generate multi-modal information such as visual and textual information. By fine-tuning the multi-modal visual understanding large model using steps S201-S203, the model can be made to have the ability to classify images in a plurality of predetermined evaluation dimensions with relatively small training costs.

[0048] In some embodiments, the multi-modal visual understanding large model can include a visual encoding layer (e.g., a ViT layer), a cross-attention layer (e.g., a CrossAttn layer), and a large language model decoding layer (e.g., a Qwen, ERNIE, or the like).

[0049] In some embodiments, the plurality of evaluation dimensions can include dimensions for evaluating images from a subjective perception or an objective perspective. Examples of evaluation dimensions are shown in Table 1.

[0050] Table 1: Examples of evaluation dimensions

[0051] Evaluation dimensions Positive cue words (Chinese) Negative cue words (Chinese) Brightness Bright image Dark image Noise level Clear image Noisy image Color richness Color-rich image Dull / low-saturation image Sharpness Sharp image Blurred image Contrast High-contrast image Low-contrast image Naturalness Natural image Synthetic image Warmth Warm image Cool image Relaxation Relaxing image Stressful image

[0052] It can be understood that the above are only examples of evaluation dimensions and are not intended to limit the scope of the present disclosure. In addition to the above dimensions, other dimensions capable of evaluating images from a subjective perception or an objective perspective are within the scope of the present disclosure. The positive and negative prompt words described above can be used for true value explanation text.

[0053] In some embodiments, the classification result of each evaluation dimension can be a binary classification result or a multi-classification result. The explanation text can be a natural language-based text that explains the classification result of the image in the evaluation dimension. By fine-tuning the multi-modal visual understanding large model using the classification result and the explanation text together, the multi-modal capabilities of the model, i.e., the visual capabilities and the text processing capabilities, can be fully utilized, so that the fine-tuned model can comprehensively master the evaluation standards of the plurality of evaluation dimensions, thereby generating more accurate classification labels in subsequent steps.

[0054] In step S202, the output format of the multi-modal visual understanding large model can be: {"label":"brightness-positive", "explanation":"The overall picture is transparent and bright, with fresh and natural color tones"}. The label field is the classification result of a specific evaluation dimension, and the explanation field is the explanation text. The specific prompt text can be designed and input into the multi-modal visual understanding large model together with the first sample image, so that the large model outputs the above format of result.

[0055] According to some embodiments, as shown in FIG. 3, the step S203 of fine-tuning the multi-modal visual understanding large model based on the first prediction result and the labeled data can include: a step S301 of calculating, based on the true value classification result and the predicted classification result of the first sample image for the plurality of evaluation dimensions, a class cross-entropy loss value of the plurality of evaluation dimensions; a step S302 of calculating, based on the true value explanation text and the predicted explanation text of the first sample image for the plurality of evaluation dimensions, a sequence cross-entropy loss value of the plurality of evaluation dimensions; and a step S303 of fine-tuning the multi-modal visual understanding large model based on the class cross-entropy loss value and the sequence cross-entropy loss value of the plurality of evaluation dimensions. Figure 3 Accordingly, by the above manner, the label prediction and explanation generation capabilities of the large model can be simultaneously optimized, so that the understanding of the large model for the multi-dimensional evaluation system is more comprehensive and accurate.

[0056] In some embodiments, in the step S301, the class cross-entropy loss value of the corresponding true value classification result and the predicted classification result can be calculated for each evaluation dimension. For each evaluation dimension, the true value classification result of the first sample image is first represented as a one-hot vector, and the predicted classification result is represented as a probability distribution, and then the class cross-entropy loss of the evaluation dimension is calculated. The total class cross-entropy loss can be summed or averaged for all evaluation dimensions to measure the deviation of the large model in class discrimination.

[0057] In some embodiments, in the step S302, the sequence cross-entropy loss value of the corresponding true value explanation text and the predicted explanation text can be calculated for each evaluation dimension. The true value explanation text can be tokenized and recorded as a sequence {w t} after tokenization, and the predicted probability of the corresponding position decoded by the large model is {q t}. An autoregressive language modeling manner can be used to calculate the loss value of each token, and the total sequence cross-entropy loss is obtained by accumulating the explanation texts of all evaluation dimensions.

[0058] In some embodiments, in the step S303, the class cross-entropy loss value and the sequence cross-entropy loss value can be summarized according to a preset weight to obtain a total loss value, and the fine-tuning training is performed based on the total loss value.

[0059] According to some embodiments, the step S303 of fine-tuning the multi-modal visual understanding large model based on the class cross-entropy loss value and the sequence cross-entropy loss value of the plurality of evaluation dimensions can include: fine-tuning the decoder in the multi-modal visual understanding large model in a low-rank adapter (LoRA) and parameter efficient fine-tuning (PEFT) manner.

[0060] Accordingly, by the above manner, the label prediction and explanation generation capabilities of the large model can be simultaneously optimized, so that the understanding of the large model for the multi-dimensional evaluation system is more comprehensive and accurate.

[0061] Thus, by the above manner, only a small number of parameters need to be fine-tuned to inject image evaluation knowledge, reducing resource consumption and accelerating model convergence speed.

[0062] In step S204, when labeling the plurality of second sample images, only classification results can be generated without generating explanation texts again. In step S202 and step S204, different prompt texts can be input to the multi-modal visual understanding large model to instruct the large model to output different results for the first sample image or the second sample image. When labeling the plurality of second sample images, the output format of the multi-modal visual understanding large model can be {"label": "brightness-positive"}.

[0063] According to some embodiments, the true value classification result and the predicted classification result can be binary determination results, and each image group can include positive images and negative images for the same evaluation dimension. In such embodiments, the image group can be referred to as an image pair.

[0064] Thus, by using binary determination and selecting positive and negative samples in the same dimension, clear positive and negative signals can be provided for contrastive learning, pairing rules can be simplified, training complexity can be reduced, and convergence can be accelerated.

[0065] In one exemplary embodiment, in step S205, the positive image for the "brightness" dimension can be an image with a binary determination result of "brightness-positive", and the negative image can be an image with a binary determination result of "brightness-negative".

[0066] In some embodiments, the lightweight image processing model can include various lightweight models for processing images and capable of generating evaluation results (e.g., classification results, scores, etc.), such as a combination of a Visualized-BGE model and a number of MLP-based scoring heads. After inputting an image into the image processing model, the image processing model can finally output evaluation results in one or more evaluation dimensions based on its internal processing mechanism.

[0067] According to some embodiments, the lightweight image processing model can include an image encoding model and a plurality of scoring models corresponding to a plurality of evaluation dimensions. For example, as shown in FIG. 6, the lightweight image processing model can include an image encoding model 601 and a plurality of scoring models 602 corresponding to a plurality of evaluation dimensions. Figure 4As shown, step S206, training the lightweight image processing model based on the plurality of image groups to obtain the trained image processing model can include: step S401, inputting the positive image and the negative image for the target evaluation dimension in one image group into the image encoding model respectively to obtain the positive image encoding feature and the negative image encoding feature; step S402, inputting the positive image encoding feature and the negative image encoding feature into the scoring model corresponding to the target evaluation dimension respectively to obtain the positive image score and the negative image score for the target evaluation dimension; step S403, calculating the contrast loss value based on the positive image score and the negative image score; and step S404, adjusting the parameters of the lightweight image processing model based on the contrast loss value.

[0068] In this way, the visual features are first extracted using the image encoding model, and then independently regressed by the scoring model dedicated to each dimension, so that the gradients corresponding to each dimension do not interfere with each other, which can improve the scoring accuracy and retain the flexibility of model expansion.

[0069] In some embodiments, in step S401, two images with opposite labels under the current target evaluation dimension can be selected from the image group, and the positive image I + and the negative image I- are respectively input into the image encoding model of the lightweight image processing model. The image encoding model can adopt a backbone network based on visual Transformer (ViT) to perform multi-layer self-attention calculation and multi-scale feature fusion on the input image, and output a fixed-length global vector representation. After processing by the encoding model, the positive image encoding feature vector E+ and the negative image encoding feature vector E- are obtained, which have consistent dimensions, facilitating subsequent processing by the scoring model.

[0070] In some embodiments, in step S402, the encoding features E + and E- can be input into the scoring model corresponding to the target evaluation dimension one by one. Each scoring model is composed of a linear mapping layer or a lightweight multi-layer perceptron (MLP) and is used to compress the 1024-dimensional image feature into a single scalar score. The scoring model can first perform layer normalization on the input feature, then map it to a hidden space through a fully connected layer, and finally output a normalized score through an activation function such as Sigmoid or Tanh, thereby obtaining the positive image score S + and the negative image score S-, both of which are used to measure the tendency of the image in the current evaluation dimension.

[0071] In some embodiments, in step S403, after S + and S- are obtained, a contrast loss function can be used to evaluate the difference between them to drive the optimization learning of the image processing model.

[0072] According to some embodiments, the difference between the positive image score and the negative image score can be negatively correlated with the contrast loss value.

[0073] Thus, the contrast loss value decreases as the difference between the positive and negative scores increases, and the training process actively widens the gap between the positive and negative scores in the same dimension, thereby enhancing the model's ability to distinguish in the target dimension.

[0074] In an exemplary embodiment, the contrast loss can adopt a margin ranking loss (Margin Ranking Loss) in the form of L = max(0, m - (S + – S-)), where m represents a preset margin threshold. If S + is significantly higher than S-, the loss tends to zero; if the difference is small, the loss is positive, forcing subsequent gradient adjustments to the model parameters to increase the positive score or decrease the negative score.

[0075] In some embodiments, after calculating the contrast loss value in step S404, the weight parameters in the image encoding model and the scoring model can be updated simultaneously through a backpropagation algorithm.

[0076] In some embodiments, during the gradient backpropagation process, only the last N layers of self-attention weights or the inserted low-rank adapter can be updated, and the remaining parameters remain frozen. In this way, the memory usage can be reduced while maintaining the original semantic embedding capability. After iterative training of multiple image groups, an image processing model with higher discrimination accuracy in each evaluation dimension and faster inference speed can be obtained.

[0077] Figure 5 A schematic diagram of constructing image pairs and training a lightweight image processing model is shown. The fine-tuned multi-modal visual understanding large model receives a plurality of second sample images and a prompt text to output classification results for the plurality of second sample images. The prompt text is used to indicate the task performed by the multi-modal visual understanding large model, i.e., outputting the classification results. Then, a plurality of image pairs can be constructed according to the classification results of the second sample images. These image pairs can be input into the lightweight image processing model to obtain image feature vectors, and then the scoring model in the upper pyramid outputs the scoring results for the corresponding evaluation dimensions, and finally the contrast loss value is calculated to adjust the parameters of the image processing model.

[0078] According to some embodiments, the trained image processing model can be used for image recommendation operations. The image recommendation operations include: step SA01, inputting a plurality of user historical interest images of a target user into the trained image processing model to obtain a predicted score of each user historical interest image for a plurality of evaluation dimensions; step SA02, determining a preference score of the target user for an evaluation dimension based on the predicted scores of the plurality of user historical interest images for the same evaluation dimension; and step SA03, determining a target image for recommendation to the target user based on the preference scores of the target user for the plurality of evaluation dimensions.

[0079] Thus, by converting the multi-dimensional average scores of the historical images of interest of the user into a user preference vector, the interest of the user can be quantified at a multi-dimensional level, which helps personalized recommendation.

[0080] It can be understood that the operation of step SA01 is similar to the operation of inputting positive images and negative images into the image processing model as described above, and will not be repeated here.

[0081] In some embodiments, in step SA02, for each evaluation dimension, the average, median or other statistical value of the predicted scores of the plurality of user historical interest images in the evaluation dimension can be determined as the preference score of the target user for the evaluation dimension.

[0082] According to some embodiments, step SA03, determining a target image for recommendation to the target user based on the preference scores of the target user for the plurality of evaluation dimensions can include: inputting a candidate image into the trained image processing model to obtain predicted scores of the candidate image for the plurality of evaluation dimensions; and determining a recommendation weight of the candidate image based on the predicted scores of the candidate image for the plurality of evaluation dimensions and the preference scores of the target user for the plurality of evaluation dimensions.

[0083] Thus, by the above-mentioned method, it can be determined whether the candidate image is suitable for the user by using the preferences of the user in each dimension, so that the recommendation result is more in line with the needs of the user.

[0084] In some embodiments, each-dimensional image scores are obtained for the candidate image, and the preference scores determined based on the user historical interest images are combined to adjust the personalized up-weight or down-weight of the resource, so as to optimize the visual style of the overall list page.

[0085] According to some embodiments, determining a recommendation weight of the candidate image based on the predicted scores of the candidate image for the plurality of evaluation dimensions and the preference scores of the target user for the plurality of evaluation dimensions can include: calculating the inner product of the predicted scores of the candidate image for the plurality of evaluation dimensions and the preference scores of the target user for the plurality of evaluation dimensions to obtain a weight change amount; and adjusting the original recommendation weight of the candidate image based on the weight change amount.

[0086] Therefore, by calculating the inner product of the scores of each dimension of the candidate image and the scores of each dimension of the user preference, multi-dimensional matching can be completed with only one matrix operation, and the computational overhead is linearly expanded, which is suitable for real-time deployment of massive candidate scenarios.

[0087] According to another aspect of the present disclosure, a large model-based image recommendation method is provided. Figure 6 As shown, the method includes: step S601, inputting multiple user historical interest images of the target user into a trained image processing model to obtain a predicted score for each user historical interest image for multiple preset evaluation dimensions, wherein the trained image processing model is trained using the above-mentioned training method; step S602, determining the target user's preference score for the evaluation dimension based on the predicted scores of multiple user historical interest images for the same evaluation dimension; and step S603, determining the target image for recommendation to the target user based on the target user's preference scores for multiple evaluation dimensions.

[0088] It is understandable that Figure 6 The operations of step S601 to step S603 in the embodiment can refer to the above description of step SA01 to step SA03, which will not be repeated here.

[0089] According to some embodiments, step S603, determining a target image for recommendation to a target user based on the target user's preference scores for multiple evaluation dimensions may include: inputting a candidate image into a trained image processing model to obtain predicted scores of the candidate image for multiple evaluation dimensions; and determining a recommendation weight of the candidate image based on the predicted scores of the candidate image for multiple evaluation dimensions and the target user's preference scores for multiple evaluation dimensions.

[0090] Therefore, through the above method, the user's preferences in various dimensions can be used to determine whether the candidate image is suitable for the user, so that the recommendation results are more in line with the user's needs.

[0091] According to some embodiments, determining the recommendation weight of a candidate image based on the predicted scores of the candidate image for multiple evaluation dimensions and the target user's preference scores for multiple evaluation dimensions may include: calculating the inner product of the predicted scores of the candidate image for multiple evaluation dimensions and the target user's preference scores for multiple evaluation dimensions to obtain a weight change; and adjusting the original recommendation weight of the candidate image based on the weight change.

[0092] Therefore, by calculating the inner product of the scores of each dimension of the candidate image and the scores of each dimension of the user preference, multi-dimensional matching can be completed with only one matrix operation, and the computational overhead is linearly expanded, which is suitable for real-time deployment of massive candidate scenarios.

[0093] According to another aspect of the present disclosure, a device for training an image processing model is provided. As shown in Figure 7 The device 700 includes an acquisition unit 710 configured to acquire a first sample image with labeled data and a plurality of second sample images without labels, the labeled data of the first sample image including true value classification results and true value explanation texts of a plurality of preset evaluation dimensions respectively; a first generation unit 720 configured to generate a first prediction result of the first sample image by using a multimodal visual understanding large model, the prediction result including predicted classification results and predicted explanation texts of the plurality of evaluation dimensions respectively; a first training unit 730 configured to fine-tune the multimodal visual understanding large model based on the first prediction result and the labeled data; a second generation unit 740 configured to classify the plurality of second sample images based on the plurality of evaluation dimensions by using the fine-tuned multimodal visual understanding large model to obtain a second prediction result of the plurality of second sample images; a construction unit 750 configured to construct a plurality of image groups based on the second prediction result, each image group including at least two images of different categories within the same evaluation dimension; and a second training unit 760 configured to train a lightweight image processing model based on the plurality of image groups to obtain a trained image processing model.

[0094] It can be understood that the operations and effects of the units 710-760 in the device 700 can be referred to the descriptions of steps S201-S206 above.

[0095] According to another aspect of the present disclosure, a device for image recommendation based on a large model is provided. As shown in Figure 8 The device 800 includes a scoring unit 810 configured to input a plurality of user historical interest images of a target user into a trained image processing model to obtain predicted scores of each user historical interest image for a plurality of preset evaluation dimensions, wherein the trained image processing model is obtained by using the device 700; a first determination unit 820 configured to determine a preference score of the target user for an evaluation dimension based on predicted scores of the plurality of user historical interest images for the same evaluation dimension; and a second determination unit 830 configured to determine a target image for recommendation to the target user based on the preference scores of the target user for the plurality of evaluation dimensions.

[0096] It can be understood that the operations and effects of the units 810-830 in the device 800 can be referred to the descriptions of steps S601-S603 above.

[0097] In the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good customs.

[0098] According to embodiments of the present disclosure, an electronic device, a readable storage medium and a computer program product are also provided.

[0099] Reference Figure 9 A block diagram of an electronic device 900 that can be used as a server or a client of the present disclosure will now be described, which is an example of a hardware device that can be applied to aspects of the present disclosure. The electronic device is intended to represent a wide variety of digital electronic computer devices such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computing devices. The electronic device can also represent a wide variety of mobile devices such as personal digital assistants, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present disclosure described and / or claimed in this document.

[0100] As shown in Figure 9 , the electronic device 900 includes a computing unit 901 that 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 into a random access memory (RAM) 903 from a storage unit 908. In the RAM 903, various programs and data required for the operation of the electronic device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other through a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0101] A plurality of components in the electronic 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 can be any type of device that can input information to the electronic device 900, and can receive inputted digital or character information, and generate key signal inputs related to user settings and / or function controls of the electronic device, and can include, but is not limited to, a mouse, a keyboard, a touch screen, a track pad, a track ball, a jog wheel, a microphone, and / or a remote controller. The output unit 907 can be any type of device that can present information, and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 908 can include, but is not limited to, a magnetic disk, an optical disk. The communication unit 909 allows the electronic device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0102] The computing unit 901 can be various 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 special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 901 performs various methods, processes, and / or procedures described above. For example, in some embodiments, these methods, processes, and / or procedures can 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 can be loaded and / or installed onto the electronic device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded onto the RAM 903 and executed by the computing unit 901, one or more steps of the methods, processes, and / or procedures described above can be performed. Alternatively, in other embodiments, the computing unit 901 can be configured to perform these methods, processes, and / or procedures by any other appropriate means, such as by means of firmware.

[0103] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0104] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, or entirely on a remote machine or server.

[0105] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can 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 the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0106] 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 the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; 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 can be received in any form, including acoustic, speech, or tactile input.

[0107] 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 a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0108] The computer system can 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. A server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS (Virtual Private Server, or simply VPS) services. The server can also be a server of a distributed system, or a server combined with a blockchain.

[0109] It should be understood that various forms of flow shown above can be used with reordering, additions, or removals of steps. For example, steps recited in the disclosure can be executed in parallel, in serial, or in different orders, as long as the desired results of the technology disclosed in the disclosure are achieved, which are not limited herein.

[0110] While embodiments or examples of this disclosure have been described with reference to the figures, it will be understood that the methods, systems, and devices described above are merely exemplary embodiments or examples, and the scope of the application is not limited to these embodiments or examples. Various elements of the embodiments or examples can be omitted or substituted by equivalents thereof. Furthermore, the steps can be performed in a different order than described in the disclosure. Further, various elements of the embodiments or examples can be combined in various ways. It is important that as technology evolves, many of the elements described herein can be substituted by equivalents which serve the same function.

Claims

1. A method for training an image processing model, comprising: Acquire a first sample image with labeled data and a plurality of unlabeled second sample images, wherein the labeled data of the first sample image includes true value classification results and true value explanation texts of respective preset multiple evaluation dimensions; Generate a first prediction result for the first sample image using a multimodal visual understanding model, the prediction result including predicted classification results and predicted explanation texts for each of the multiple evaluation dimensions; Fine-tuning the multimodal visual understanding model based on the first prediction result and the labeled data; Using the fine-tuned multimodal visual understanding large model, classify the plurality of second sample images based on the plurality of evaluation dimensions to obtain second prediction results for the plurality of second sample images; constructing a plurality of image groups based on the second prediction result, each image group including at least two images of different categories within the same evaluation dimension; as well as Based on the multiple image groups, a lightweight image processing model is trained to obtain a trained image processing model.

2. The method according to claim 1, wherein The true value classification result and the predicted classification result are binary judgment results, and each image group includes positive images and negative images for the same evaluation dimension.

3. The method according to claim 2, wherein: The lightweight image processing model includes an image coding model and multiple scoring models corresponding to the multiple evaluation dimensions. The step of training the lightweight image processing model based on the multiple image groups to obtain a trained image processing model includes: Inputting positive images and negative images for a target evaluation dimension in an image group into the image coding model respectively to obtain positive image coding features and negative image coding features; Inputting the positive image coding features and the negative image coding features into a scoring model corresponding to the target evaluation dimension respectively to obtain a positive image score and a negative image score for the target evaluation dimension; calculating a contrast loss value based on the positive image score and the negative image score; and Based on the contrast loss value, parameters of the lightweight image processing model are adjusted.

4. The method according to claim 3, wherein: The difference between the positive image score and the negative image score is negatively correlated with the contrast loss value.

5. The method according to any one of claims 1 to 4, wherein Fine-tuning the multimodal visual understanding model based on the first prediction result and the labeled data includes: Calculating category cross entropy loss values ​​for the multiple evaluation dimensions based on the true classification results and the predicted classification results of the first sample image for the multiple evaluation dimensions; Calculating sequence cross entropy loss values ​​of the multiple evaluation dimensions based on the true explanation text and the predicted explanation text of the first sample image for the multiple evaluation dimensions; and Based on the category cross entropy loss values ​​and sequence cross entropy loss values ​​of the multiple evaluation dimensions, the multimodal visual understanding large model is fine-tuned and trained.

6. The method according to claim 5, wherein: The fine-tuning training of the multimodal visual understanding large model based on the category cross entropy loss values ​​and the sequence cross entropy loss values ​​of the multiple evaluation dimensions includes: The decoder in the multimodal visual understanding model is fine-tuned using a low-rank adapter and efficient parameter fine-tuning.

7. The method according to claim 3, wherein: The trained image processing model is used for the following image recommendation operations: Inputting multiple historical interest images of a target user into the trained image processing model to obtain a predicted score for each historical interest image of the user with respect to the multiple evaluation dimensions; Determining the target user's preference score for the evaluation dimension based on the predicted scores of the multiple user historical interest images for the same evaluation dimension; as well as Based on the preference scores of the target user with respect to the multiple evaluation dimensions, a target image is determined for recommendation to the target user.

8. The method according to claim 7, wherein: The determining of the target image to be recommended to the target user based on the target user's preference scores for the multiple evaluation dimensions includes: Inputting the candidate image into the trained image processing model to obtain predicted scores of the candidate image for the multiple evaluation dimensions; and The recommendation weight of the candidate image is determined based on the predicted scores of the candidate image with respect to the multiple evaluation dimensions and the preference scores of the target user with respect to the multiple evaluation dimensions.

9. The method according to claim 8, wherein The determining of the recommendation weight of the candidate image based on the predicted scores of the candidate image for the multiple evaluation dimensions and the target user's preference scores for the multiple evaluation dimensions includes: Calculating the inner product of the predicted scores of the candidate image for the multiple evaluation dimensions and the preference scores of the target user for the multiple evaluation dimensions to obtain a weight change; and Based on the weight change, the original recommendation weight of the candidate image is adjusted.

10. A large model-based image recommendation method, comprising: Inputting multiple historical user interest images of a target user into a trained image processing model to obtain a predicted score for each of the historical user interest images with respect to multiple preset evaluation dimensions, wherein the trained image processing model is trained using the method of any one of claims 1 to 9; Determining the target user's preference score for the evaluation dimension based on the predicted scores of the multiple user historical interest images for the same evaluation dimension; and Based on the preference scores of the target user with respect to the multiple evaluation dimensions, a target image is determined for recommendation to the target user.

11. The method according to claim 10, wherein: The determining of the target image to be recommended to the target user based on the target user's preference scores for the multiple evaluation dimensions includes: Inputting the candidate image into the trained image processing model to obtain predicted scores of the candidate image for the multiple evaluation dimensions; and The recommendation weight of the candidate image is determined based on the predicted scores of the candidate image with respect to the multiple evaluation dimensions and the preference scores of the target user with respect to the multiple evaluation dimensions.

12. The method according to claim 11, wherein The determining of the recommendation weight of the candidate image based on the predicted scores of the candidate image for the multiple evaluation dimensions and the target user's preference scores for the multiple evaluation dimensions includes: Calculating the inner product of the predicted scores of the candidate image for the multiple evaluation dimensions and the preference scores of the target user for the multiple evaluation dimensions to obtain a weight change; and Based on the weight change, the original recommendation weight of the candidate image is adjusted.

13. A training device for an image processing model, comprising: an acquiring unit configured to acquire a first sample image with annotated data and a plurality of unannotated second sample images, wherein the annotated data of the first sample image includes a true value classification result and a true value explanation text of each of the plurality of preset evaluation dimensions; A first generating unit is configured to generate a first prediction result of the first sample image using a multimodal visual understanding large model, wherein the prediction result includes a predicted classification result and a predicted explanation text for each of the multiple evaluation dimensions; A first training unit is configured to fine-tune the multimodal visual understanding model based on the first prediction result and the labeled data; a second generating unit configured to classify the plurality of second sample images based on the plurality of evaluation dimensions using the fine-tuned multimodal visual understanding large model to obtain second prediction results for the plurality of second sample images; a constructing unit configured to construct a plurality of image groups based on the second prediction result, each image group including at least two images of different categories within the same evaluation dimension; as well as The second training unit trains the lightweight image processing model based on the multiple image groups to obtain a trained image processing model.

14. An image recommendation device based on a large model, comprising: a scoring unit configured to input a plurality of historical user interest images of a target user into a trained image processing model to obtain a predicted score for each historical user interest image with respect to a plurality of preset evaluation dimensions, wherein the trained image processing model is trained using the apparatus according to claim 13; A first determining unit is configured to determine the target user's preference score for the evaluation dimension based on the predicted scores of the plurality of user historical interest images for the same evaluation dimension; and The second determining unit is configured to determine a target image for recommendation to the target user based on the preference scores of the target user with respect to the multiple evaluation dimensions.

15. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 12.

16. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-12.

17. A computer program product comprising a computer program, wherein When the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

Images