Method and system for generating empty room decoration effect picture based on AI large model

Abstract

The invention discloses a method and system for generating an empty room decoration effect picture based on an AI large model, and the method comprises the steps: S1, obtaining data information corresponding to a decoration effect image, inputting the obtained data information into the AI large model, and carrying out the training, an empty room decoration model is obtained; s2, obtaining a to-be-decorated empty room picture, and analyzing the obtained empty room picture to generate an empty room frame diagram; and S3, inputting the empty room frame diagram and pre-selected decoration effect information into the empty room decoration model for processing, and generating a decoration effect diagram of the to-be-decorated empty room picture.

Description

Technical Field

[0001] The present invention relates to the field of interior decoration technology, and in particular to a method and system for generating a decoration rendering of an empty room based on an AI large model. Background Art

[0002] At present, there are many pain points in the design and production of renderings for empty rooms or rough houses in the soft decoration design industry. The most prominent problems include: the long production cycle of renderings, the single design effect, and the inability to quickly meet customer needs. These problems not only affect the customer experience, but also restrict the development of the soft decoration design industry.

[0003] First of all, the long production cycle of renderings is a common problem in the design of empty room renderings. Since designers need to spend a lot of time on hand-drawing or modeling, the production cycle of renderings is long and inefficient. In addition, since the production of renderings requires a lot of time and energy, it also increases the work pressure and cost of designers, which in turn affects the efficiency and quality of the entire industry.

[0004] Secondly, meeting the personalized needs of customers is also a major pain point in the design of empty room renderings. There are already many soft decoration design software on the market, but customer needs are getting higher and higher and more personalized. Therefore, designers need to customize the design according to customer needs, which takes more time and effort, and also increases the communication cost and time between customers and designers.

[0005] Although there are some quick design software for empty room renderings on the market, their implementation principle is to search for similar content from the existing database through the text or photos provided by the user, and paste the found pictures onto the renderings. The effect is poor, and the style of the pictures found is not uniform. At the same time, due to the different ideas and styles of different designers, it is difficult to guarantee the effect and quality of the pasted pictures based on the existing pictures. In addition, due to the low quality and precision of the mapping, color difference, blur, incoordination and other problems are prone to occur, which not only affects the customer experience, but also affects the quality and reputation of the entire soft decoration design.

[0006] However, AIGC (artificial intelligence generated content) as a new content creation method, AIGC inherits the advantages of professional generated content and user generated content, and gives full play to the technical advantages to create a new form of digital content generation and interaction. With the continuous development of AIGC, technologies such as AI writing, AI music, and AI painting have caused extensive discussion in the field of creation. Since AIGC mainly uses deep learning, reinforcement learning and other technologies to achieve content generation through training models, models occupy an important position. With the emergence of large models (with a huge number of model parameters), more realistic and lifelike content generation becomes possible, allowing AIGC to play a role in practical applications. Therefore, the present invention proposes a method and system for generating empty room decoration renderings based on AI large models. Summary of the invention

[0007] The purpose of the present invention is to address the defects of the prior art and provide a method and system for generating decoration renderings of empty rooms based on an AI large model.

[0008] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0009] A method for generating a decoration rendering of an empty room based on an AI large model, comprising:

[0010] S1. Obtain data information corresponding to the decoration effect image, and input the obtained data information into the AI ​​large model for training to obtain an empty room decoration model;

[0011] S2. Obtaining an image of an empty room to be renovated, and analyzing the obtained image of the empty room to generate an empty room framework diagram;

[0012] S3. Input the empty room framework diagram and the pre-selected decoration effect information into the empty room decoration model for processing, and generate a decoration effect diagram of the empty room to be decorated.

[0013] Furthermore, the step S1 specifically includes:

[0014] S11. Obtain multiple pictures of complete renderings;

[0015] S12. Crop the obtained multiple renderings to obtain renderings of the same size, and divide the obtained renderings into a training image set and a test image set;

[0016] S13. Label the training images according to their contents, rename and save the training images and the corresponding label data, and input the saved images into the Stable Diffusion model for training;

[0017] S14. Determine whether the number of iterations reaches the preset value. If not, continue to execute step S13; if yes, output the AI ​​decoration model;

[0018] S15. Input the test picture set into the AI ​​decoration model for processing, and the AI ​​decoration model outputs the decoration effect picture;

[0019] S16. Determine whether the output decoration rendering is the same as the input test picture. If not, re-execute step S13; if so, obtain the final empty room decoration model.

[0020] Furthermore, the step S2 specifically includes:

[0021] S21. Obtain a picture of the room to be decorated;

[0022] S22. Check whether the acquired room image is an empty room. If not, no processing is performed; if so, an empty room framework image is generated.

[0023] Furthermore, the step S3 specifically includes:

[0024] S31. Input the empty room frame diagram into the empty room decoration model for processing, and select the overall style of the generated rendering and the data information corresponding to the rendering;

[0025] S32. The empty room decoration model outputs decoration effect pictures.

[0026] Furthermore, the data information corresponding to the effect diagram in step S31 includes the image size of the effect diagram, the prompt words of the required effect diagram, and the number of effect diagram images generated.

[0027] Accordingly, a system for generating decoration renderings of empty rooms based on an AI large model is also provided, including

[0028] An input module is used to obtain data information corresponding to the decoration effect image, and input the obtained data information into the AI ​​large model for training to obtain an empty room decoration model;

[0029] An analysis module is used to obtain pictures of empty rooms to be renovated, analyze the obtained pictures of empty rooms, and generate a framework diagram of the empty rooms;

[0030] The generation module is used to input the empty room framework diagram and the pre-selected decoration effect information into the empty room decoration model for processing, and generate a decoration effect diagram of the empty room picture to be decorated.

[0031] Furthermore, the input module includes:

[0032] A first acquisition module is used to acquire multiple pictures of complete renderings;

[0033] The first processing module is used to perform cropping processing on the acquired multiple rendering images to obtain rendering images of the same size, and divide the obtained rendering images into a training image set and a test image set;

[0034] The training module is used to label the training images according to their contents, rename and save the training images and the corresponding label data, and input the saved images into the Stable Diffusion model for training.

[0035] A judgment module is used to judge whether the number of iterations reaches a preset value;

[0036] The test module is used to input the test picture set into the AI ​​decoration model for processing, and the AI ​​decoration model outputs the decoration effect picture;

[0037] The verification module is used to determine whether the output decoration rendering is the same as the input test picture.

[0038] Furthermore, the analysis module includes:

[0039] The second acquisition module is used to acquire the picture of the room to be decorated;

[0040] The detection module is used to detect whether the acquired room image is an empty room.

[0041] Furthermore, the generation module includes:

[0042] The second processing module is used to input the empty room frame diagram into the empty room decoration model for processing, and select the overall style of the generated rendering and data information corresponding to the rendering;

[0043] Output module, used to output decoration effect diagram of empty room decoration model.

[0044] Furthermore, the data information corresponding to the effect graph in the second processing module includes the image size of the effect graph, the prompt words of the required effect graph, and the number of effect graph images generated.

[0045] Compared with the prior art, the present invention integrates various design styles into the painting AI large model as the basic model through training and learning of existing design schemes, and then analyzes the empty room photos uploaded by users to obtain their structural information, and identifies the specific frame position of the space and the unique horizontal coefficient and vertical coefficient of each model according to its actual scene. At the same time, combined with the user's demand prompt word description, and then combined with the model's original training basic images and soft decoration image data combination arrangement, the present invention accurately generates empty room decoration renderings for users. The entire rendering generation time is extremely short, the generated styles are diverse, the effects are realistic, and the generated single renderings are unified in style and high in quality, which can be directly used as a reference for decoration renderings, thereby providing powerful auxiliary role for designers in the field of interior design in the design of empty room effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a method for generating an empty room decoration rendering based on an AI large model provided in Example 1;

[0047] Figure 2 It is a system structure diagram of the empty room decoration effect diagram provided in the first embodiment. DETAILED DESCRIPTION

[0048] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0049] The purpose of the present invention is to address the defects of the prior art and provide a method and system for generating decoration renderings of empty rooms based on an AI large model.

[0050] Embodiment 1

[0051] This embodiment provides a method for generating a decoration rendering of an empty room based on an AI large model. Figure 1 As shown, including:

[0052] S1. Obtain data information corresponding to the decoration effect image, and input the obtained data information into the AI ​​large model for training to obtain an empty room decoration model;

[0053] S2. Obtaining an image of an empty room to be renovated, and analyzing the obtained image of the empty room to generate an empty room framework diagram;

[0054] S3. Input the empty room framework diagram and the pre-selected decoration effect information into the empty room decoration model for processing, and generate a decoration effect diagram of the empty room to be decorated.

[0055] The method of the empty room decoration effect diagram of this embodiment is based on the application system. Figure 2 As shown in the figure, the application system is divided into four layers based on its functions, including infrastructure layer, data storage layer, application platform layer, and business application layer. The basic equipment layer and data storage layer are used as the underlying support of the system to provide basic functions such as data storage, model storage, and application hosting; the application platform layer is mainly used for training verification and model deployment of large AI models; the business application layer implements the main business functions, and then is used by designers and end users in the system user layer.

[0056] The infrastructure layer is mainly responsible for providing computing, storage and network resources to ensure the stable operation of the system. These resources include servers, storage devices, protection systems, network equipment and related hardware and software facilities. The role of this layer is to provide a stable, reliable and efficient operating environment for the entire system. When building the empty room rendering system, this layer used 2 application servers, 1 data server, and 5 AI computing servers, all hosted in a professional cloud computer room to provide reliable computing power and services.

[0057] The data storage layer is mainly responsible for the storage, management and protection of data to ensure the availability, integrity and security of data. The data storage layer includes a database management system, a file system and related data caching technology. The role of this layer is to provide the system with reliable data storage and model file management, and to ensure the security and consistency of data. The database used in this layer uses MySQL, which is a commonly used relational database management system with the characteristics of stability and high performance. It can store and manage various data of the system, such as model file information, parameter information, rendering generation records, etc. The file management system uses Minio, which is an open source object storage system developed based on the Go language. It has high performance, high reliability and easy scalability. Minio provides a simple and reliable object storage solution for fields such as massive data storage, artificial intelligence, and big data analysis. The model used in the present invention and the original image and rendering before generation are stored by Minio.

[0058] The application platform layer is mainly responsible for providing public services and functional support for business applications to ensure the rapid development and deployment of business applications. The application platform layer includes various data collection and processing tools, data screening and cleaning tools, and model training and verification tools, such as data collection, data management, AI model training, etc. AI model training includes data enhancement, model training, model verification, model management parameter configuration, etc. Data collection and cleaning processing tools are developed based on Python. As a high-level programming language, Python has the advantages of being easy to learn and highly efficient. It is also widely used in the field of artificial intelligence and provides a large number of easy-to-use libraries to facilitate AI program development. The empty room rendering system uses OpenCV for image processing and uses TensorFlow and PyTorch for model training.

[0059] The business application layer is mainly responsible for the specific business logic processing and implementation, and is one of the most important parts of the entire system. The business application layer includes various business application modules and components, such as photo upload, prompt word input, rendering exhibition hall, rendering generation, style model selection, historical generation management, and management background interfaces. The management background includes model management, model parameters, generation record management, prompt word classification management, prompt word management, and user management; the role of this layer is to provide the UI layer with various business interfaces. In the present invention, this layer mainly uses Python's FastAPI library as a development framework. FastAPI is a modern Web framework for building APIs using Python. It combines the asynchronous characteristics and type annotations of Python 3.7+ to provide excellent performance and development experience. In the present invention, FastAPI is used to build a RESTful API to provide service interfaces such as training and reasoning of machine learning models and other business interfaces. Because FastAPI has the characteristics of fast response and low latency, it can help achieve rapid deployment and efficient calling of AI models, and improve the response speed and concurrent processing capabilities of the model. At the same time, the asynchronous nature of FastAPI also enables it to better adapt to high-concurrency request scenarios and provide stronger support for AI applications.

[0060] The system user layer is mainly responsible for providing users with an interactive experience, including front-end frameworks for PC and mobile terminals, so that users can easily interact with the system. The interactive interface of the present invention is developed in Vue as a whole. Vue is a popular front-end framework with the characteristics of simplicity, flexibility and efficiency. In the present invention, Vue receives images or text data uploaded by users, and then uses the back-end FastAPI interface to obtain the AI ​​model processing results and display them to users on the front end. Based on Vue's responsive data binding and componentized development methods, the slower execution process can be displayed in real time during AI reasoning. At the same time, it can ensure that when the reasoning results of the AI ​​model change, the front-end interface can reflect these changes in real time, thereby ensuring smooth communication between the front-end interface and the back-end AI model.

[0061] This embodiment is based on the above framework and is further divided into two modules: model training and rendering generation. Step S1 is the model training module, and steps S2-S3 are the rendering generation module.

[0062] In step S1, data information corresponding to the decoration effect image is obtained, and the obtained data information is input into the AI ​​large model for training to obtain an empty room decoration model.

[0063] Model training is the whole process of a model trainer (such as a design user) importing image data through the training interface provided by the system, merging various design styles into the painting AI model as the basic model, setting various parameters required for training, starting training to generate a model, and releasing the model online after verification. It specifically includes the following steps:

[0064] S11. Obtain multiple pictures of complete renderings.

[0065] Before training, you need to prepare a training data set. Specifically, according to the needs of the soft furnishing style, collect complete renderings as training data; the quality of the collected renderings needs to be high-definition, and the pictures of the same style should be unified, but the content needs to be diverse. At the same time, the number of samples for each training should be reasonably set, as the number of samples will affect the final fitting results. Too few samples may lead to underfitting of the model, while too many samples may lead to overfitting of the model.

[0066] It should be noted that in this embodiment, a model is trained for each style of renderings, so there are multiple models for renderings of multiple styles, and then the output of renderings of different styles is achieved according to the multiple models.

[0067] S12. Crop the obtained multiple renderings to obtain renderings of the same size, and divide the obtained renderings into a training image set and a test image set.

[0068] Image cropping: The images in the training set are cropped into images of the same size, and the commonly used size is 512*512. Among them, cropping images can be implemented using existing solutions, and this embodiment will not be described in detail.

[0069] S13. Label the training images according to their different contents, rename and save the training images and the corresponding label data, and input the saved images into the Stable Diffusion model for training.

[0070] Each training image is labeled according to its content. After manual labeling, several labels for different image contents are obtained. Different training images and their corresponding labels are renamed, and the renamed data is saved. The saved image information is input into the Stable Diffusion model in the AI ​​model for training.

[0071] Labels are manually annotated based on the content in the picture, mainly describing the objects and environment in the picture, such as: bedroom, sofa, bed, bathroom and other label words.

[0072] Renaming is for easy reading, that is, the image name is consistent with the name of the text file storing the image label. During training, the label and image can be matched one by one after reading by name.

[0073] Before training, the parameters of the Stable Diffusion model need to be set, such as setting appropriate learning rate, batch size and other hyperparameters, number of iterations, etc., and using appropriate optimizers (such as Adam, etc.) to optimize the model; during the training process, the model needs to be evaluated regularly, including calculating loss functions, accuracy and other indicators to measure the performance of the model. This method mainly uses loss functions to evaluate the model.

[0074] The loss function is used to evaluate the model: the loss is the difference between the original image and the generated image, and the loss function is a mathematical formula used to calculate the size of the difference. This method uses the commonly used L2 loss function (Euclidean distance method), that is, the similarity between data points is measured by image pixels. By minimizing the difference between images, the generated image can achieve a high degree of similarity with the training image. In the calculation process, the L2 loss function is used to calculate the content loss and style loss respectively, and finally the total loss is calculated to achieve the purpose of evaluating the model. The ultimate goal is to make the loss between the training image and the generated image converge to a relatively low value, which is usually between 0.1 and 0.2 based on experience and training conditions. If it is not reached, the parameters need to be adjusted and retrained.

[0075] S14. Determine whether the number of iterations reaches the preset value. If not, continue to execute step S13; if yes, output the AI ​​decoration model;

[0076] S15. Input the test picture set into the AI ​​decoration model for processing, and the AI ​​decoration model outputs the decoration effect picture;

[0077] Model testing: After training is completed, the model is tested using a test data set to evaluate the accuracy of the model and the final image quality.

[0078] S16. Determine whether the output decoration rendering is the same as the input test picture. If not, re-execute step S13; if so, obtain the final empty room decoration model.

[0079] If the output decoration rendering is different from the input test picture, the model needs to be tuned. Specifically, the model needs to be adjusted and optimized according to the evaluation results, including modifying the network structure, increasing or decreasing the number of layers, adjusting the learning rate, increasing the number of samples, etc.

[0080] If the output decoration rendering is the same as the input test picture, the model that passes the test will be released for users to use.

[0081] In this embodiment, the model trainer is trained through the application platform layer tool. The training method is to perform style transfer learning based on the existing StableDiffusion large model, and take the modern Nordic style bedroom generation model as an example. The training steps are as follows (models of other styles are similar):

[0082] (1) Prepare 200 pictures of modern Nordic-style bedrooms;

[0083] (2) Use the image processing function to resize the images to 512x512 pixels, select 150 of them as the training image set and 50 as the verification image set;

[0084] (3) Select the basic large model stablediffusion 1.5 version, and perform transfer learning on this basis. Specifically, import the training set image set into the stablediffusion large model for training, label the images according to their content in the stable diffusion large model, rename the label data and image data for saving, set the required parameters in the training interface, including the number of images is 150, the image size is 512x512, the number of training iterations is 10, the number of training steps is 20, and the input prompt word is configured as "modern Nordic bedroom". Then start training, check the loss function value during training, save the model obtained when the loss function value is less than 0.1, and obtain the modern Nordic style bedroom model after saving.

[0085] The prompt word is a type of image label, but it is quite special. It is a general term for a type of image, that is, all current training images contain this label.

[0086] (4) After training, the modern Nordic-style bedroom model is evaluated. Specifically, 50 training pictures are selected, the AI ​​image generation function is used to upload the training pictures, the model for this training is selected, and the label "modern Nordic bedroom" is entered. Other labels such as "bed" and "bedside table" can also be entered. Click "Generate". After generating the renderings in turn, observe whether the style of the generated renderings is consistent with the style of the training pictures used to train the model. At the same time, observe whether the generated content items are consistent with the input prompt labels. If the content is inconsistent with the prompt label, and the style is inconsistent with the training picture style, modify the parameters and training pictures, and re-train.

[0087] (5) If the generated rendering meets the requirements, the verification is successful. The verified model is uploaded to the system model library and enabled for online use.

[0088] According to the method of this embodiment, a plurality of models of different styles are obtained, and the plurality of models of different styles are stored in the model library.

[0089] In step S2, a picture of an empty room to be renovated is obtained, and the obtained picture of the empty room is analyzed to generate a framework diagram of the empty room.

[0090] The user accesses the corresponding interface of the system business application layer of this embodiment through a browser, selects a photo of his own room or takes a photo and uploads it, and the system performs a structural analysis on the uploaded photo; specifically, it includes:

[0091] S21. Obtain a picture of the room to be decorated;

[0092] Upload pictures of empty rooms: Users enter the page for generating empty room renderings and use the picture upload function to take photos or select photos to upload pictures of empty rooms.

[0093] S22. Check whether the acquired room image is an empty room. If not, no processing is performed; if so, an empty room framework image is generated.

[0094] Empty room image analysis: After the empty room image is uploaded, the trained spatial analysis model is called to analyze the spatial framework in the image to determine whether the uploaded room image is an empty room. If it is an empty room, the ControlNet Depth model (open source model) is called to generate an empty room framework image.

[0095] It should be noted that the training method of the spatial analysis model is consistent with the training method of the above-mentioned style model. The difference is that the spatial analysis model obtains the classification labels of the training images (manually labeled during training), that is, whether the image shows information such as a bedroom, living room, or empty room.

[0096] In step S3, the empty room framework diagram and the pre-selected decoration effect information are input into the empty room decoration model for processing to generate a decoration effect diagram of the empty room to be decorated.

[0097] After generating the framework diagram of the empty room, the system provides multiple available AI decoration models of different styles and corresponding prompt label suggestions. The user selects the AI ​​decoration model of the desired style, and then generates a rendering based on the AI ​​decoration model by selecting or entering the prompt label.

[0098] In this embodiment, the purpose of inputting the prompt tag is to accurately control the generated content in the rendering. Since the empty room picture itself has no objects, the content is relatively random when it is generated purely based on the model, so it is necessary to input the prompt tag to control the generation of the object.

[0099] Step S3 specifically includes:

[0100] S31. Input the empty room frame diagram into the empty room decoration model for processing, and select the overall style of the generated rendering and the data information corresponding to the rendering;

[0101] Effect picture style parameter settings: select the overall style model of the effect picture to be generated, select the size of the generated image, set the prompt words for personalized needs, and select the number of effect pictures to be generated.

[0102] The overall style model includes a base model and a style model. The base model is the Stable Diffusion 1.5 large model, and the style model is the pre-trained model mentioned above. The style model is a model obtained by fine-tuning the large model, so it needs to be used with the corresponding large model.

[0103] S32. The empty room decoration model outputs decoration effect pictures.

[0104] Rendering generation and inspection: After setting the parameters, the rendering is generated according to the uploaded original picture of the empty room and the set parameters. The user checks the generated rendering and selects the appropriate picture to save and use. It can also be generated repeatedly.

[0105] This embodiment takes the selection of a modern Nordic bedroom model as an example. After the desired AI decoration model is selected, other parameters can be set. Other parameters include: setting the image width and height to 1024x768, the prompt words to "bed, bedside table, bedside lamp", and the number of generated images to 4; then start the generation to obtain 4 decoration effect pictures, and select the ones with good effects to save and use.

[0106] The decoration rendering generated in this embodiment is a brand new rendering generated according to the model style, empty room framework and prompt tags input by the user. It is based on the characteristics of Stable Diffusion 1.5 (diffusion model) to diffuse the style and content of the existing picture, thereby achieving the purpose of creating a new decoration rendering.

[0107] This embodiment, through training and learning of existing design solutions, merges various design styles into the painting AI large model as the basic model, and then analyzes the empty room photos uploaded by the user to obtain their frame information, identifies the specific frame position of the space according to its actual scene, and combines the user's demand prompt word description with the model's original training basic images and soft decoration image data combination and arrangement to accurately generate empty room decoration renderings for users. The entire rendering generation time is extremely short, the generated styles are diverse, the effects are realistic, and the generated single renderings are unified in style and high in quality, which can be directly used as a reference for decoration renderings, thereby providing powerful auxiliary role for designers in the field of interior design when designing empty room effects.

[0108] Embodiment 2

[0109] This embodiment provides a system for generating decoration renderings of empty rooms based on an AI large model, including

[0110] An input module is used to obtain data information corresponding to the decoration effect image, and input the obtained data information into the AI ​​large model for training to obtain an empty room decoration model;

[0111] An analysis module is used to obtain pictures of empty rooms to be renovated, analyze the obtained pictures of empty rooms, and generate a framework diagram of the empty rooms;

[0112] The generation module is used to input the empty room framework diagram and the pre-selected decoration effect information into the empty room decoration model for processing, and generate a decoration effect diagram of the empty room picture to be decorated.

[0113] Furthermore, the input module includes:

[0114] A first acquisition module is used to acquire multiple pictures of complete renderings;

[0115] The first processing module is used to perform cropping processing on the acquired multiple rendering images to obtain rendering images of the same size, and divide the obtained rendering images into a training image set and a test image set;

[0116] The training module is used to label the images in the training image set according to their contents, rename and save the training images and the corresponding label data, and input the saved images into the Stable Diffusion model for training;

[0117] A judgment module is used to judge whether the number of iterations reaches a preset value;

[0118] The test module is used to input the test picture set into the AI ​​decoration model for processing, and the AI ​​decoration model outputs the decoration effect picture;

[0119] The verification module is used to determine whether the output decoration rendering is the same as the input test picture.

[0120] Furthermore, the analysis module includes:

[0121] The second acquisition module is used to acquire the picture of the room to be decorated;

[0122] The detection module is used to detect whether the acquired room image is an empty room.

[0123] Furthermore, the generation module includes:

[0124] The second processing module is used to input the empty room frame diagram into the empty room decoration model for processing, and select the overall style of the generated rendering and data information corresponding to the rendering;

[0125] Output module, used to output decoration effect diagram of empty room decoration model.

[0126] Furthermore, the data information corresponding to the effect graph in the second processing module includes the image size of the effect graph, the prompt words of the required effect graph, and the number of effect graph images generated.

[0127] It should be noted that the system for generating decoration renderings of empty rooms based on an AI large model provided in this embodiment is similar to the first embodiment, and will not be elaborated in detail in this embodiment.

[0128] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for generating decoration renderings of empty rooms based on AI large models, characterized in that: include: S1. Obtain data information corresponding to the decoration effect image, and input the obtained data information into the AI ​​large model for training to obtain an empty room decoration model; S2. Obtaining an image of an empty room to be renovated, and analyzing the obtained image of the empty room to generate an empty room framework diagram; S3. Input the empty room framework diagram and the pre-selected decoration effect information into the empty room decoration model for processing, and generate a decoration effect diagram of the empty room to be decorated.

2. According to the method of claim 1, the method is characterized in that: The step S1 specifically includes: S11. Obtain multiple pictures of complete renderings; S12. Crop the obtained multiple renderings to obtain renderings of the same size, and divide the obtained renderings into a training image set and a test image set; S13. Label the training images according to their contents, rename and save the training images and the corresponding label data, and input the saved images into the Stable Diffusion model for training; S14. Determine whether the number of iterations reaches the preset value. If not, continue to execute step S13; if yes, output the AI ​​decoration model; S15. Input the test picture set into the AI ​​decoration model for processing, and the AI ​​decoration model outputs the decoration effect picture; S16. Determine whether the output decoration rendering is the same as the input test picture. If not, re-execute step S13; if so, obtain the final empty room decoration model.

3. The method for generating a decoration rendering of an empty room based on an AI large model according to claim 2, characterized in that: The step S2 specifically includes: S21. Obtain a picture of the room to be decorated; S22. Check whether the acquired room image is an empty room. If not, no processing is performed; if so, an empty room framework image is generated.

4. The method for generating a decoration rendering of an empty room based on an AI large model according to claim 3 is characterized in that: The step S3 specifically includes: S31. Input the empty room frame diagram into the empty room decoration model for processing, and select the overall style of the generated rendering and the data information corresponding to the rendering; S32. The empty room decoration model outputs decoration effect pictures.

5. The method for generating a decoration rendering of an empty room based on an AI large model according to claim 4 is characterized in that: The data information corresponding to the effect diagram in step S31 includes the image size of the effect diagram, the prompt words of the required effect diagram, and the number of effect diagram images generated.

6. A system for generating decoration renderings of empty rooms based on AI large models, characterized in that: It includes an input module, which is used to obtain data information corresponding to the decoration effect image, and input the obtained data information into the AI ​​big model for training to obtain the empty room decoration model; An analysis module is used to obtain pictures of empty rooms to be renovated, analyze the obtained pictures of empty rooms, and generate a framework diagram of the empty rooms; The generation module is used to input the empty room framework diagram and the pre-selected decoration effect information into the empty room decoration model for processing, and generate a decoration effect diagram of the empty room picture to be decorated.

7. The system for generating decoration renderings of empty rooms based on AI large models according to claim 6 is characterized in that: The input module comprises: A first acquisition module is used to acquire multiple pictures of complete renderings; The first processing module is used to perform cropping processing on the acquired multiple rendering images to obtain rendering images of the same size, and divide the obtained rendering images into a training image set and a test image set; The training module is used to label the training images according to their contents, rename and save the training images and the corresponding label data, and input the saved images into the Stable Diffusion model for training. A judgment module is used to judge whether the number of iterations reaches a preset value; The test module is used to input the test picture set into the AI ​​decoration model for processing, and the AI ​​decoration model outputs the decoration effect picture; The verification module is used to determine whether the output decoration rendering is the same as the input test picture.

8. The system for generating decoration renderings of empty rooms based on AI large models according to claim 7 is characterized in that: The analysis module comprises: The second acquisition module is used to acquire the picture of the room to be decorated; The detection module is used to detect whether the acquired room image is an empty room.

9. The system for generating decoration renderings of empty rooms based on AI large models according to claim 8 is characterized in that: The generation module comprises: The second processing module is used to input the empty room frame diagram into the empty room decoration model for processing, and select the overall style of the generated rendering and data information corresponding to the rendering; Output module, used to output decoration effect diagram of empty room decoration model.

10. The system for generating decoration renderings of empty rooms based on AI large models according to claim 9 is characterized in that: The data information corresponding to the effect graph in the second processing module includes the image size of the effect graph, the prompt words of the required effect graph, and the number of effect graph images generated.

Images