Method for rapidly setting personalized photographing parameters of mobile phone based on low-parameter large model

Abstract

The invention discloses a mobile phone personalized photographing parameter rapid setting method based on a low-parameter large model, and the method comprises the steps: collecting and sorting publicly published commercial-level professional photos and historical photos shot by a user mobile phone, and creating a unified photo data set; constructing a low-parameter visual large model, and inputting the unified photo data set into the low-parameter visual large model for training; and according to the current photographing scene or theme of the user, generating personalized photographing parameters by using the trained low-parameter visual large model. The method has the remarkable effects that a user can conveniently obtain professional guidance and self-adjustment of shooting parameter setting through a mobile phone, so that the urgent requirement of a current mobile phone shooting group for quickly obtaining high-quality, personalized and specialized shooting is met.

Description

Technical Field

[0001] The present invention relates to the field of personalized mobile phone photography and machine vision technology, and more specifically, to a method for quickly setting personalized mobile phone photography parameters based on a low-parameter large model. Background Art

[0002] At present, AI-driven and embedded cameras can adjust exposure, focus, white balance, ISO and other parameters in real time to achieve accurate automatic shooting mode. AI face recognition and tracking technology allows the camera to automatically lock the face, focus and optimize exposure when shooting portraits, greatly simplifying the shooting process and lowering the technical threshold. At the same time, AI's image recognition technology can quickly identify different scene categories, such as landscapes, portraits, sports, night scenes, etc., and can automatically select the best shooting mode based on this, so that ordinary users can also take photos close to professional levels.

[0003] Currently, most mobile phones have built-in AI machine algorithms, and the camera can develop the knack of automatically identifying scenes, available light and even scene angles. In this way, when the user uses the camera to shoot, the mobile phone camera will recognize that it is shooting a person, food, scenery, animals, etc., and choose the most appropriate exposure and color adjustment for the subject, thereby improving the image quality or the quality of the film.

[0004] Usually, after the AI ​​algorithm model is debugged by the manufacturer, it is integrated into the mobile phone firmware, and the various parameters of the camera are intelligently adjusted by automatically analyzing the shooting environment and object characteristics. Since the model is pre-made and solidified by the manufacturer, it is generally impossible to predict the light, texture and other parameter information of the user's on-site shooting to update the algorithm. This makes the so-called AI photography, which may be just "one-click beautification", just roughly raising the saturation or applying filters, thus facing the inability to personalize parameter settings for users, and the photos cannot reflect user preferences. It cannot meet the user's personalized photography needs, resulting in the dilemma that users are now unwilling to use this function.

[0005] In order to effectively solve the above problems and effectively improve the quality of personalized photography with mobile phones, this application proposes a method for quickly generating personalized photography parameters for mobile phones to help users quickly take high-quality photos with personal aesthetic characteristics without professional photography knowledge. Summary of the invention

[0006] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method for quickly setting personalized photography parameters for a mobile phone based on a low-parameter large model, which can face different photography scenes and quickly match the photography parameter requirements of different users, thereby realizing the generation of personalized photography parameters.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] In a first aspect, the present invention provides a method for quickly setting personalized camera parameters of a mobile phone based on a low-parameter large model, the key of which is that it comprises the following steps:

[0009] Step 1: Data collection and analysis: Collect and organize publicly published commercial professional photos and historical photos taken by users' mobile phones to create a unified photo dataset;

[0010] Step 2: Model construction and training: construct a low-parameter visual large model, and input the unified photo dataset into the low-parameter visual large model for training;

[0011] Step 3: Personalized parameter generation: Generate personalized photography parameters based on the user's current photography scene or theme using the trained low-parameter visual large model.

[0012] Furthermore, the specific steps for creating a unified photo dataset in step 1 are as follows:

[0013] Step 1.1: Collect and obtain commercial-grade professional photo data based on publicly released commercial photo datasets and create a professional-grade photo dataset;

[0014] Step 1.2: Collect and organize historical photos of users based on their mobile phone photos to create a personalized photo dataset;

[0015] Step 1.3: The unified photo dataset is obtained based on the professional-level photo dataset and the personalized photo dataset.

[0016] Furthermore, the process of creating the professional-level photo dataset is as follows:

[0017] Get the shooting parameters of commercial-grade professional photo data;

[0018] The professional-level photo dataset is created based on the commercial-level professional photo data and its corresponding photographing parameters.

[0019] Furthermore, the creation process of the personalized photo dataset is as follows:

[0020] Users can classify their historical photos into three levels based on their own preferences.

[0021] The user classifies the user historical photos into three levels according to the personalized style, and the personalized style of each category contains at least one user historical photo of each level, so as to obtain the personalized photo data set.

[0022] Furthermore, the steps of obtaining the unified photo dataset according to the professional-level photo dataset and the personalized photo dataset are as follows:

[0023] Taking the personalized photo dataset as input, a plurality of professional-level photos having similar styles to the user's historical photos in the personalized photo dataset are screened out from the professional-level photo dataset;

[0024] Add the selected professional-level photos to the personalized photo dataset, and adjust the imbalance of the data scale of the three levels of photos in the new personalized photo dataset;

[0025] Extracting the photographing parameters of the photos in the new personalized photo dataset;

[0026] The unified photo dataset is generated based on the new personalized photo dataset and its corresponding photographing parameters.

[0027] Furthermore, the specific steps of model construction and training in step 2 are as follows:

[0028] Step 2.1, model construction: build a low-parameter visual large model;

[0029] Step 2.2, data preprocessing: clean and standardize the data in the unified photo dataset to construct a training dataset;

[0030] Step 2.3, model training: Input the training data set into the low-parameter visual large model to train and learn it.

[0031] Furthermore, the process of processing the training data set by the low-parameter large visual model is as follows:

[0032] Using grouped convolution technology, the input training data set and convolution kernels are divided into several groups, and the convolution kernels in each group only perform convolution operations with the images of the corresponding group;

[0033] Each photo is divided into a series of tiles. Based on the deep convolution technology, non-overlapping convolutions are performed on the tiles using convolution kernels of the same size as the tiles to obtain the feature values ​​of each tile.

[0034] Based on the feature values ​​of each block, the feature map of each photo output by the output layer is obtained;

[0035] Use point-by-point convolution technology to perform point-by-point convolution on the feature map to generate full-scale output;

[0036] Calculate the full-scale similarity and compare it with the given threshold. If it is not less than the given threshold, the training ends.

[0037] Furthermore, the process of generating personalized parameters in step 3 is as follows:

[0038] Step 3.1, the user opens the camera application and selects a shooting scene or subject;

[0039] Step 3.2: Extract the features of the current scene or subject through image recognition technology;

[0040] Step 3.3: Use the trained low-parameter visual large model to generate personalized photography parameters based on the user's preferences and the extracted features of the current scene or subject.

[0041] Furthermore, the personalized photography parameters include the aperture, ISO sensitivity, and shutter data of the mobile phone camera.

[0042] In a second aspect, the present application provides a mobile phone personalized photography parameter generation system based on a low-parameter large model, comprising:

[0043] The dataset creation module is used to collect and organize publicly published commercial professional photos and historical photos taken by users' mobile phones to create a unified photo dataset;

[0044] A model creation and training module, used for constructing a low-parameter visual large model, and inputting the unified photo dataset into the low-parameter visual large model for training;

[0045] The personalized parameter generation module is used to generate personalized photography parameters based on the user's current photography scene or theme using a trained low-parameter visual large model.

[0046] The present invention first obtains three types of shooting parameters (aperture, ISO sensitivity, shutter) of professional photos through publicly published commercial-grade professional photos; then, through user historical photos, a personalized style is created, and three types of shooting parameters (aperture, ISO sensitivity, shutter) of historical photos are obtained; then, the user historical photos are divided into three levels (excellent, good and ordinary) according to the user's self-preference; then, by selecting a low-parameter large model, on a commercial-grade professional photo dataset, with the overall target of three different levels of user historical photo datasets, a plurality of professional-grade photos similar to the historical photos are output, thereby expanding the historical photo dataset, constructing a unified dataset, and training and learning are performed on this basis; three types of shooting parameter datasets corresponding to the unified dataset are created; finally, the user's photo results are input into the large model, and similar photos can be quickly obtained; and according to the needs of on-site photography, the model outputs three types of shooting parameters of similar photos, namely personalized photography parameters, thereby meeting the urgent needs of the current mobile phone photography group to quickly obtain high-quality and personalized professional photography.

[0047] The remarkable effects of the present invention are:

[0048] 1. Intelligently recommend parameters to achieve personalized photography and improve user experience. By learning the user's photography style (such as photography habits and preferences, etc.), personalized photography parameters can be quickly generated, personalized photography modes can be provided, and user operations can be simplified, while achieving better shooting quality and experience.

[0049] 2. Optimize algorithms to improve photo quality. Using low-parameter large models to quickly analyze and process photos, users can adjust parameters in real time to significantly improve shooting effects. For example, in dynamic scenes, you can quickly adjust the focus and depth of field to ensure that the subject is clear and the background is soft.

[0050] 3. Simplify the learning curve, reduce shooting time, and save time and energy. Users no longer need to go through complex manual settings. Just shoot and the system will give reference parameter values, saving time for trial and error and debugging. For example, when traveling, users can quickly switch between different scenes to ensure that they do not miss any shooting opportunities. Especially for ordinary photography users, this method can serve as an intelligent guide, and they can still take more professional photos without having to master professional knowledge.

[0051] 4. It has a certain level of scene intelligence perception and improves shooting adaptability. This application obtains scene information of the object being photographed through pre-shooting with a mobile phone, and inputs a large model to generate reference parameter values, so that users do not have to worry about changes in different shooting conditions and being unable to grasp the shooting quality. In addition, this application can not only be applied to general photography, but can also be extended to multiple scenes such as video recording and slow motion shooting, increasing the scope of application.

[0052] 5. Stimulate creativity, promote creativity and experimentation, and enhance the fun of shooting. Users can conduct shooting experiments based on the system's recommendations, try different settings and effects, thereby improving their creative photography capabilities and enhancing the fun of shooting.

[0053] In general, the method for quickly setting personalized photography parameters for mobile phones based on low-parameter large models not only reduces the difficulty of taking pictures, improves the user's photography experience, and improves efficiency, but also produces beneficial effects in many aspects such as photo quality and creative expression. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of the method of the present invention;

[0055] Figure 2 It is a system principle block diagram of the present invention. DETAILED DESCRIPTION

[0056] The specific implementation manner and working principle of the present invention are further described in detail below with reference to the accompanying drawings.

[0057] Embodiment 1:

[0058] like Figure 1 As shown, this embodiment provides a method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model, and the specific steps are as follows:

[0059] Step 1: Data collection and analysis: Collect and organize publicly published commercial professional photos and historical photos taken by users' mobile phones to create a unified photo dataset;

[0060] In this example, the specific steps to create a unified photo dataset are as follows:

[0061] Step 1.1: Collect and obtain commercial-grade professional photo data based on publicly released commercial photo datasets and create a professional-grade photo dataset;

[0062] In the embodiment of the present invention, the process of creating the professional-level photo dataset is as follows:

[0063] A1. Obtain the three shooting parameters of commercial-grade professional photos in the above commercial photo dataset, including aperture, ISO sensitivity, and shutter, and define the image quality level as professional level;

[0064] A2. Create the professional-level photo dataset based on the commercial-level professional photo data and its corresponding shooting parameters.

[0065] During the specific implementation process, the professional-level photo dataset can be gradually updated and expanded in the subsequent practical stage.

[0066] Step 1.2: Collect and organize historical photos of users based on their mobile phone photos to create a personalized photo dataset;

[0067] In this example, the creation process of the personalized photo dataset is as follows:

[0068] B1. Users classify their historical photos into three levels: excellent, good and ordinary according to their own preferences;

[0069] B2. The user classifies the user's historical photos of three levels according to the personalized style, and each category of personalized style contains at least one user's historical photo of each level, thereby obtaining the personalized photo data set. That is, the user's personalized style is created. Considering the computing resources of the mobile terminal, this embodiment recommends that the total number of photos collected in the personalized photo data set is about 2,000.

[0070] Step 1.3: On a desktop platform, using a relatively new large model such as Revit (Radiance Vision Transformer), the unified photo dataset is obtained based on the professional photo dataset and the personalized photo dataset.

[0071] In this example, the steps for making the unified photo dataset are as follows:

[0072] C1. Taking the personalized photo dataset as input, multiple professional-level photos with similar styles to the user's historical photos in the personalized photo dataset are screened out from the professional-level photo dataset;

[0073] C2. Add the selected professional-level photos to the personalized photo dataset and adjust the imbalance of the data scale of the three levels of photos in the new personalized photo dataset;

[0074] C3, extracting three types of photographic parameters such as aperture, ISO, and shutter speed from the photos in the new personalized photo dataset;

[0075] C4. Generate the unified photo dataset based on the new personalized photo dataset and its corresponding three types of photo parameters, such as aperture, ISO, and shutter. The scale is about 20,000 pictures, that is, 20,000 categories are generated, and a maximum storage capacity of 150GB to 200GB is required. The main models of existing mobile phone brands basically meet the storage requirements.

[0076] Step 2: Model construction and training: construct a low-parameter visual large model, and input the unified photo dataset into the low-parameter visual large model for training;

[0077] In this example, the specific steps of model construction and training are as follows:

[0078] Step 2.1, model construction: build a low-parameter visual large model;

[0079] This embodiment selects a machine learning or deep learning model (such as a lightweight convolutional neural network, etc.) under resource-constrained conditions to analyze and predict shooting parameters. At present, lightweight visual transformers (Vision Transformer, ViTs) have shown better performance and lower latency than lightweight convolutional neural networks (CNNs, the typical representative large model is ShuffleNetV2, etc.) on resource-constrained mobile devices. Based on the need to reduce the amount of mobile computing, and with similarity as the core indicator of the accuracy of the large model, this method considers integrating the advanced technologies of the above two types of lightweight network models in reducing the amount of computing, and designs a low-parameter visual large model that meets the task requirements of this method.

[0080] Step 2.2, data preprocessing: clean and standardize the data in the unified photo dataset to construct a training dataset;

[0081] Step 2.3, model training: Input the training data set into the low-parameter visual large model to train and learn it.

[0082] The similarity calculation between images described in this method follows the principle of giving priority to local similarity and preserving global similarity, that is, the similarity between two images should calculate multiple local similarities (multiple, here refers to the number of image parts involved in the calculation being artificially set to exceed 50% of the total number of parts). If multiple parts are similar, it is directly predicted that the two given images are similar; if multiple parts are not similar, it is necessary to calculate the full similarity again. If the similarity is within the threshold, the two given images can also be judged to be similar, otherwise the two given images are not similar. The similarity calculation principle proposed by this method allows this method to reduce the amount of calculation by more than 30% compared with the existing low-parameter large model on the mobile terminal, making it more cost-effective to deploy on the mobile terminal.

[0083] Therefore, the process of processing the training data set by the low-parameter large visual model described in this embodiment is as follows:

[0084] First, to reduce the amount of computation, consider using the group convolution technique on the data set. The input image and convolution kernel are divided into several groups. The convolution kernel in each group only performs convolution operations with the image in the corresponding group;

[0085] Secondly, an image is divided into a series of patches, and then the depthwise-separable convolution technique is used to perform non-overlapping convolution on the patches with a convolution kernel as large as the patch, and a value is obtained after each convolution. Thus, by setting multiple convolution kernels, multiple values ​​will be obtained, and the values ​​will be accumulated to obtain the feature value of a patch;

[0086] After that, the feature values ​​of all sub-blocks of an image can be generated, and then the feature map of each photo output by the output layer can be obtained based on the feature values ​​of each block;

[0087] Then, the feature map is convolved point by point using pointwise convolution to generate full-scale output;

[0088] Finally, the full-scale similarity is calculated and compared with the given threshold. If it is not less than the given threshold, the training ends.

[0089] In this example, the activation function of the low-parameter visual large model selects the ReLU function.

[0090] In addition, taking into account the reality that taking photos is a creation rather than a reproduction of the scenery, and the low performance of mobile phone photography hardware, when facing the training and learning of large models, on the premise of ensuring that the full-frame similarity (that is, the image similarity) is greater than or equal to 75% (the delay is at the second level, not the millisecond level), conventional parameter adjustment modes and empirical parameters are adopted to shorten the training cycle and reduce costs, reduce the response time of the model in the actual combat stage, and improve user satisfaction, without blindly pursuing the optimal performance of the large model.

[0091] It can be seen that the low-parameter large visual model described in this embodiment has the following advantages:

[0092] 1) Low resource requirements: Large models with low parameters are streamlined and optimized to reduce memory usage and computational requirements, enabling them to run efficiently on mobile devices such as smartphones and tablets.

[0093] 2) Fast real-time response: Since mobile phone photography usually requires fast response, low-parameter large models can provide faster inference speed and are suitable for real-time application scenarios such as real-time image processing and video analysis.

[0094] 3) Support multimedia applications: The existing low-parameter large model, combined with technologies such as Camera2 API, can implement complex multimedia functions on mobile phones, such as real-time preview, photo taking, video recording, focus and face recognition, to enhance user experience.

[0095] 4) Easy to integrate. Due to its smaller model size and lower computing requirements, low-parameter large models are easier to integrate with existing mobile applications, making it easier to generate apps.

[0096] 5) High energy efficiency. Large models with low parameters can reduce energy consumption while maintaining performance, thus extending the battery life of mobile devices.

[0097] It can be seen that the above-mentioned characteristics of the low-parameter large model adopted by the present invention on mobile devices can effectively balance performance and resource consumption, support rich multimedia applications, and meet the requirements of this application for real-time and high efficiency.

[0098] Step 3: Personalized parameter generation: Generate personalized photography parameters based on the user's current photography scene or theme using the trained low-parameter visual large model.

[0099] In this example, the process of generating the personalized parameters is as follows:

[0100] Step 3.1, the user opens the camera application and selects a shooting scene or subject;

[0101] Step 3.2: Extract the features of the current scene or subject through image recognition technology;

[0102] Step 3.3: Use the trained low-parameter visual large model to generate personalized photography parameters based on the user's preferences and the extracted features of the current scene or subject.

[0103] In the specific implementation process, the learned low-parameter visual large model can be deployed in the cloud or on a mobile phone for practical use. If deployed on a mobile phone, consider encapsulating it as a mobile phone APP. The APP needs to call the professional mode of mobile phone shooting and can read and write the aperture, ISO sensitivity, and shutter data of the mobile phone camera. Consider the camera-related APIs (such as Camera2 API) included in the mobile phone system, combined with the specific development tools required by the API, such as Java, to programmatically read and write the aperture, ISO sensitivity, and shutter data of the mobile phone camera; or consider the kivy framework for Python, which can also realize the reading and writing tasks of the above parameters.

[0104] In addition, since this method uses a large model, its development tool is based on Python, so consider using tools such as Chaquopy to facilitate the needs of Python+Java mixed coding. If only Python is used, there is no need to consider the mixed coding task. Based on the kivy framework, the Android packaging of the program is completed and the APK under the Android system is generated, which is the aforementioned mobile phone APP.

[0105] When deployed in the cloud, the user's mobile phone photo taking process is as follows:

[0106] First, the user takes a test shot using the phone’s regular camera mode;

[0107] Secondly, the phone will use image recognition technology to extract the features of the current scene or subject from the test photo;

[0108] The phone will then upload the user's preferences and the extracted features of the current scene or subject to the cloud, which will use the trained low-parameter large visual model to generate personalized photography parameters and send them down;

[0109] After that, the mobile phone receives personalized shooting parameters, and the user selects the three required shooting parameters according to the on-site shooting needs, namely aperture, ISO sensitivity, shutter speed and other three shooting parameters;

[0110] Finally, the user completes the photo taking according to the selected photo parameters.

[0111] When deployed on a mobile phone and packaged as a mobile APP, the user's mobile phone photo taking service process is as follows:

[0112] First, the user takes a test shot using the regular mode;

[0113] Secondly, the test photos are used to extract the features of the current scene or subject through image recognition technology;

[0114] Then, based on the user’s preferences and the extracted features of the current scene or subject, the trained low-parameter visual large model is used to generate personalized photography parameters;

[0115] After that, the user selects the three required shooting parameters according to the on-site shooting requirements, namely aperture, ISO sensitivity, shutter speed and other three shooting parameters;

[0116] Finally, the user completes the photo taking according to the selected photo parameters.

[0117] During the shooting process, users can continue to fine-tune the selected parameters one by one, and quickly set the shooting parameters required by the user. In the UI, the fine-tuning function can use a gear-style icon to provide a toggle method to help users fine-tune.

[0118] Embodiment 2:

[0119] like Figure 2 As shown, this embodiment provides a mobile phone personalized photography parameter generation system based on a low-parameter large model for implementing the method described in Embodiment 1, including:

[0120] The dataset creation module is used to collect and organize publicly published commercial professional photos and historical photos taken by users' mobile phones to create a unified photo dataset;

[0121] A model creation and training module, used for constructing a low-parameter visual large model, and inputting the unified photo dataset into the low-parameter visual large model for training;

[0122] The personalized parameter generation module is used to generate personalized photography parameters based on the user's current photography scene or theme using a trained low-parameter visual large model.

[0123] In specific implementation, if the low-parameter large visual model is deployed on a mobile phone and packaged as a mobile phone APP, the system described in this embodiment also includes:

[0124] The encapsulation module is used to encapsulate the learned low-parameter visual large model into a mobile phone APP. The APP needs to call the professional mode of mobile phone shooting and can read and write the aperture, ISO sensitivity, and shutter data of the mobile phone camera;

[0125] A processing module, used for users to complete taking photos using the APP;

[0126] The execution module is used for the user to continue fine-tuning and quickly set the required shooting parameters based on the matched shooting parameters.

[0127] In summary, the present invention first obtains three types of shooting parameters (aperture, ISO sensitivity, shutter) of professional photos through publicly published commercial-grade professional photos; then, through user historical photos, a personalized style is created, and three types of shooting parameters (aperture, ISO sensitivity, shutter) of historical photos are obtained; then, the user's historical photos are divided into three levels (excellent, good and ordinary) according to the user's self-preference; then, by selecting a low-parameter large model, on the commercial-grade professional photo dataset, with the overall user historical photo datasets of three different levels as the target, multiple professional-grade photos similar to the historical photos are output, thereby expanding the historical photo dataset, constructing a unified dataset, and training and learning are carried out on this basis; three types of shooting parameter datasets corresponding to the unified dataset are created; finally, the user's photo results are input into the large model, and similar photos can be quickly obtained; and according to the needs of on-site photography, the model outputs three types of shooting parameters of similar photos, that is, personalized photography parameters, thereby meeting the urgent needs of the current mobile phone photography group to quickly obtain high-quality and personalized professional photography.

[0128] It can be seen that the present invention fully understands the user's photographic habits and preferences, constructs a hierarchy of user historical photos, and integrates commercial-grade professional image data. After training and learning with the above data sets, the large model used in this application grasps the user's personalized photographic habits and preferences. The generated personalized photographic parameters can help the user quickly complete the setting of qualified photographic parameters with a distinctive personal style, thereby obtaining satisfactory photos.

[0129] The technical solution provided by the present invention is described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model, characterized in that: The steps include: Step 1: Data collection and analysis: Collect and organize publicly published commercial professional photos and historical photos taken by users' mobile phones to create a unified photo dataset; Step 2: Model construction and training: construct a low-parameter visual large model, and input the unified photo dataset into the low-parameter visual large model for training; Step 3: Personalized parameter generation: Generate personalized photography parameters based on the user's current photography scene or theme using the trained low-parameter visual large model.

2. The method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model according to claim 1, characterized in that: The specific steps to create a unified photo dataset in step 1 are as follows: Step 1.1: Collect and obtain commercial-grade professional photo data based on publicly released commercial photo datasets and create a professional-grade photo dataset; Step 1.2: Collect and organize historical photos of users based on their mobile phone photos to create a personalized photo dataset; Step 1.3: The unified photo dataset is obtained based on the professional-level photo dataset and the personalized photo dataset.

3. The method for quickly setting personalized camera parameters for mobile phones based on low-parameter large models according to claim 2, characterized in that: The process of creating the professional-level photo dataset is as follows: Obtain the shooting parameters of commercial-grade professional photo data; The professional-level photo dataset is created based on the commercial-level professional photo data and its corresponding photographing parameters.

4. The method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model according to claim 2, characterized in that: The creation process of the personalized photo dataset is as follows: Users can classify their historical photos into three levels based on their own preferences. The user classifies the user historical photos into three levels according to the personalized style, and the personalized style of each category contains at least one user historical photo of each level, so as to obtain the personalized photo data set.

5. The method for quickly setting personalized camera parameters for mobile phones based on low-parameter large models according to claim 4, characterized in that: The steps of obtaining the unified photo dataset according to the professional-level photo dataset and the personalized photo dataset are as follows: Taking the personalized photo dataset as input, a plurality of professional-level photos having similar styles to the user's historical photos in the personalized photo dataset are screened out from the professional-level photo dataset; Add the selected professional-level photos to the personalized photo dataset, and adjust the imbalance of the data scale of the three levels of photos in the new personalized photo dataset; Extracting the photographing parameters of the photos in the new personalized photo dataset; The unified photo dataset is generated based on the new personalized photo dataset and its corresponding photographing parameters.

6. The method for quickly setting personalized camera parameters for mobile phones based on low-parameter large models according to claim 1, characterized in that: The specific steps of model construction and training in step 2 are as follows: Step 2.1, model construction: build a low-parameter visual large model; Step 2.2, data preprocessing: clean and standardize the data in the unified photo dataset to construct a training dataset; Step 2.3, model training: Input the training data set into the low-parameter visual large model to train and learn it.

7. The method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model according to claim 6, characterized in that: The process of processing the training data set by the low-parameter large visual model is as follows: Using grouped convolution technology, the input training data set and convolution kernels are divided into several groups, and the convolution kernels in each group only perform convolution operations with the images of the corresponding group; Each photo is divided into a series of tiles. Based on the deep convolution technology, non-overlapping convolutions are performed on the tiles using convolution kernels of the same size as the tiles to obtain the feature values ​​of each tile. Based on the feature values ​​of each block, the feature map of each photo output by the output layer is obtained; Use point-by-point convolution technology to perform point-by-point convolution on the feature map to generate full-scale output; Calculate the full-scale similarity and compare it with the given threshold. If it is not less than the given threshold, the training ends.

8. The method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model according to claim 1, characterized in that: The process of generating personalized parameters in step 3 is as follows: Step 3.1, the user opens the camera application and selects a shooting scene or subject; Step 3.2: Extract the features of the current scene or subject through image recognition technology; Step 3.3: Use the trained low-parameter visual large model to generate personalized photography parameters based on the user's preferences and the extracted features of the current scene or subject.

9. The method for quickly setting personalized camera parameters for a mobile phone based on a low-parameter large model according to any one of claims 1 to 8, characterized in that: The personalized photography parameters include the aperture, ISO sensitivity, and shutter speed data of the mobile phone camera.

Images