Intelligent agent autonomous evolutionary algorithm for business scene and application system of intelligent agent autonomous evolutionary algorithm

Abstract

The invention discloses a business scene-oriented agent autonomous evolutionary algorithm and an application system thereof, which are applied to the field of gymnasium equipment arrangement and can perform intelligent post-processing on an effect picture generated by AI, realize seamless joint of virtual design and real products, provide a WYSIWYG solution for users and improve the user experience. The method comprises the following implementation steps: S1, generating a virtual effect picture; s2, acquiring equipment elements: acquiring types and corresponding pixel coordinates of the elements by adopting a machine vision algorithm; s3, product matching; s4, a real effect picture is obtained, wherein virtual elements are cleared through an image erasing algorithm; subsequently adjusting the model size, angle and light setting of the real element to obtain an effect picture of the real scene; and S5, user feedback: collecting evaluation and modification suggestions of the user on a real scheme, recording sales volume of recommended products, and realizing autonomous learning and sustainable evolution of the system.

Description

Technical Field

[0001] The present invention relates to the field of large models, and specifically to an intelligent agent autonomous evolution algorithm and its application system for business scenarios. Background Art

[0002] The selection and layout of gym equipment directly affects user experience and operational efficiency. Traditional equipment configuration relies on manual experience, and has problems such as high cost and poor adaptability. In recent years, large image models based on generative AI have made great progress. They can perform space planning based on user input instructions (such as venue size, equipment requirements and style preferences), and automatically generate equipment layout plans, which greatly reduces the time and labor costs of traditional gym planning. However, the renderings obtained by generative AI often have problems such as inconsistency with physical laws and difficulty in realizing design elements, resulting in a significant gap between virtual design and actual implementation. In response to this pain point, the technical solution provided by the present invention can perform intelligent post-processing of AI-generated renderings, achieve seamless connection between virtual design and real products, and provide customers with a what-you-see-is-what-you-get solution.

[0003] Currently, there are also some technologies related to the field of AI image generation on the market. For example, Chinese patent CN202411603990.8 discloses a training method for an image generation model, an image generation method, and an apparatus thereof. Based on a set of style reference images and a first prompt word, a pre-trained base large model is fine-tuned and trained to finally obtain a target image generation model. However, the technology provided by the above patent focuses on the pre-training and parameter fine-tuning of the large image generation model, and is less adapted and optimized for specific application scenarios. As a result, the image generation quality cannot meet user needs and cannot be directly applied to equipment selection and space layout in the gym.

[0004] In order to solve the above problems, the present invention provides an intelligent agent autonomous evolution algorithm and its application system for commercial scenarios, which can perform intelligent post-processing on AI-generated renderings, realize the seamless connection between virtual design and real products, and provide users with a what-you-see-is-what-you-get equipment layout plan. Summary of the Invention

[0005] To address the above issues, the present invention provides an intelligent agent autonomous evolution algorithm for commercial scenarios and its application system, which is applied to the field of gym equipment layout. It can perform intelligent post-processing on AI-generated renderings, achieve seamless integration of virtual design and real products, and provide users with a what-you-see-is-what-you-get equipment layout solution. The specific implementation steps are as follows:

[0006] S1: Virtual rendering generation: Based on the keywords input by the user, a virtual rendering is generated using an AI model; the keywords include: venue area, equipment requirements, style, and budget; the AI ​​model uses Stable Diffusion;

[0007] S2: Equipment element acquisition: using a machine vision algorithm to identify the type and corresponding pixel coordinates of elements in the virtual rendering; the machine vision algorithm uses an image recognition algorithm based on deep learning; the elements include: a treadmill, a basketball stand, a badminton court, a table tennis table, a bench press, a spinning bike, and a gymnastics ball;

[0008] S3: Product Matching: Segment the virtual elements obtained in S2 and save them as separate jpg images. Retrieve detailed information about the virtual elements from the product library, obtain 3D data on price, model, and color, and store it in a MySQL database. Provide three real product options for each virtual element based on appearance consistency, model compatibility, and price.

[0009] S4: Product Replacement: Remove virtual elements through image erasure algorithms; use KeyShot to embed the real product model into the virtual rendering, and then adjust the model size, angle and lighting settings to obtain the rendering of the real scene;

[0010] S5: User Feedback: Collect user evaluations and modification suggestions on real solutions, record the sales of recommended products, and enable autonomous learning and continuous evolution of the system;

[0011] The image erasure algorithm is expressed as a deep neural network algorithm, which can erase the elements within the mask without damaging the original basic information of the image; the mask adopts the pixel coordinates obtained by the image recognition algorithm;

[0012] The real three-dimensional model adopts the obj format. The construction method of the three-dimensional model can use a three-dimensional laser scanner to obtain high-precision three-dimensional point cloud data, and then obtain a three-dimensional image of the element; the construction of the three-dimensional model can generate a three-dimensional model based on two-dimensional images from multiple perspectives.

[0013] The advantages of the present invention are:

[0014] 1. High efficiency: Traditional equipment configuration relies on manual experience, resulting in high costs and poor adaptability. This solution generates preliminary virtual renderings using large image models, offering excellent generalization performance and providing users with a wider range of options.

[0015] 2. Realistic AI: Renderings generated by generative AI often exhibit issues such as discrepancies with physical laws and difficulty implementing design elements, leading to significant discrepancies between virtual designs and actual implementation. This solution utilizes AI and image processing technology to seamlessly integrate virtual designs with real-world products, providing users with a "what you see is what you get" equipment layout solution.

[0016] 3. Increase product sales: This solution can provide users with an integrated equipment purchasing solution and increase the sales of fitness equipment by providing real product links. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the present invention, and do not constitute a limitation of the embodiments of the present invention.

[0018] Figure 1 : Framework diagram of the autonomous evolution algorithm of intelligent agents for business scenarios and its application system implementation.

[0019] Figure 2 : Image erasing algorithm and mask production process.

[0020] Figure 3 : Obtain three-dimensional models based on laser scanner.

[0021] Figure 4 : Obtaining three-dimensional models based on two-dimensional images from multiple perspectives.

[0022] Figure 5 : Implementation steps for product addition. DETAILED DESCRIPTION

[0023] The present invention will be further described below using drawings and examples. The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention.

[0024] Example 1

[0025] Attachment Figure 1 This paper presents an autonomous evolutionary algorithm for intelligent agents in commercial scenarios and its application system implementation framework. This algorithm, applied to the field of gym equipment layout, can intelligently post-process AI-generated renderings, achieve seamless integration between virtual design and real products, and provide users with a WYSIWYG equipment layout solution. The specific implementation steps are as follows:

[0026] S1: Virtual rendering generation: Based on the keywords input by the user, a virtual rendering is generated using an AI model; the keywords include: venue area, equipment requirements, style, and budget; the AI ​​model uses Stable Diffusion;

[0027] S2: Equipment element acquisition: using a machine vision algorithm to identify the type and corresponding pixel coordinates of elements in the virtual rendering; the machine vision algorithm uses an image recognition algorithm based on deep learning; the elements include: a treadmill, a basketball stand, a badminton court, a table tennis table, a bench press, a spinning bike, and a gymnastics ball;

[0028] S3: Product Matching: Segment the virtual elements obtained in S2 and save them as separate jpg images. Retrieve detailed information about the virtual elements from the product library, obtain 3D data on price, model, and color, and store it in a MySQL database. Provide three real product options for each virtual element based on appearance consistency, model compatibility, and price.

[0029] S4: Product Replacement: Remove virtual elements through image erasure algorithms; use KeyShot to embed the real product model into the virtual rendering, and then adjust the model size, angle and lighting settings to obtain the rendering of the real scene;

[0030] S5: User Feedback: Collect user evaluations and modification suggestions on real solutions, record the sales of recommended products, and enable autonomous learning and continuous evolution of the system;

[0031] The image erasure algorithm is expressed as a deep neural network algorithm, which can erase the elements within the mask without damaging the original basic information of the image; the mask adopts the pixel coordinates obtained by the image recognition algorithm;

[0032] The real three-dimensional model adopts the obj format. The construction method of the three-dimensional model can use a three-dimensional laser scanner to obtain high-precision three-dimensional point cloud data, and then obtain a three-dimensional image of the element; the construction of the three-dimensional model can generate a three-dimensional model based on two-dimensional images from multiple perspectives.

[0033] Example 2

[0034] LAMA (Locally Aware Meta-learning Algorithm) is an advanced image erasure algorithm widely used in fields such as digital photography, medical imaging, artistic restoration, and video editing. The LAMA model utilizes meta-learning and locally aware techniques to rapidly adapt to new tasks and data, ensuring that the model more accurately captures both local details and global structure in an image, resulting in more natural and coherent erasure effects than traditional methods. Therefore, in this project, the LAMA algorithm was selected to inpaint the intermediate image after masking.

[0035] After using the YOLO algorithm to locate the target, you need to write a Python program to create a mask to define the position of the virtual equipment elements. This can then remove the target from the original image, achieving the clearing and separation of the virtual equipment elements.

[0036] The production of the mask frame is a key step, which aims to convert the target area detected by the YOLO algorithm into a mask image with the same size as the original image.

[0037] Attachment Figure 2 The image erasing algorithm and mask production process are demonstrated. The specific implementation steps are as follows:

[0038] S1. Target Boundary Location: Using the output of the YOLO algorithm, we obtain the target's bounding box information, that is, the target's position and size. During this process, we need to modify the detect function of the YOLO algorithm as follows:

[0039] with open("output.txt", "a", encoding="utf-8") as file:

[0040] file.write(str(int(10)) + " " )

[0041] with open('output.txt', 'a') as file: # Format the information and write it to the file

[0042] log_string = '%g %g ' % im.shape[2:]

[0043] file.write(log_string)

[0044] with open("output.txt", "a", encoding="utf-8") as file:

[0045] file.write(str(int(cls)) + " " + str(int(d.xyxy[0,0])) + " " + str(int(d.xyxy[0,1])) + " " + str(int(d.xyxy[0,2])) + " " + str(int(d.xyxy[0,3])) + " ")

[0046] with open("output.txt", "a", encoding="utf-8") as file:

[0047] file.write("\n")

[0048] S2. Generate mask image: Extract the image size information, create a blank mask image, and then use the target bounding box information output in step S1 to obtain the mask;

[0049] S3. In the mask image, set the pixels within the bounding box of the target to non-zero values ​​and the remaining pixels to zero, and finally separate the virtual equipment elements from the original image through the LAMA algorithm.

[0050] We then output the file path, image size, detected category, and coordinates of the rectangle to a TXT file. The specific format is as follows: 1100334067 1500 1500 0 250 830 544 1121 0 549 97 845 390 0 799 8281095 1121 0 650 1195 945 1484 0 400 463 695 756 0 99 1194 394 1485 0 948 4641247 757 0 1099 99 1396 392 0 1 98 297 392 0 1200 1195 1495 1489 2 1264 13681489 1478 0 1350 900 1500 1121 3 1 463 146 622 0 0 465 144 755 1 1350 9571500 1121

[0051] Among them, the first data is the image name, the second data is the image width, the third data is the image height, and the fourth to eighth data are the target type and the rectangular frame position respectively; and so on. In the entire algorithm, we wrote Python program code and used TXT files as a bridge for communication with the AI ​​algorithm to realize the recognition of the target position and finally obtain the mask.

[0052] S4. The original image and the obtained mask are transmitted to the LAMA algorithm to clear specific targets in the virtual rendering. The entire environment is based on torch-1.13.1.

[0053] Example 3 Acquiring a 3D model of a real product is crucial and a prerequisite for obtaining high-quality, real-world renderings. During system implementation, the real 3D model is constructed using the obj format. A 3D laser scanner can be used to acquire high-precision 3D point cloud data, which in turn creates a 3D image of the element. This 3D model can be generated from 2D images from multiple perspectives.

[0054] Attachment Figure 3 The three-dimensional model is obtained based on the laser scanner. The specific implementation steps are as follows:

[0055] S1: Equipment calibration: Calibrate the scanner after powering on to ensure modeling accuracy;

[0056] S2: Product Scan: Scan around the treadmill to ensure that all external surfaces of the product are covered and there are no blind spots;

[0057] S3: Post-processing: Use the product's supporting software to stitch and denoise the point cloud data, and then convert the point cloud into a mesh to obtain the surface contour of the actual treadmill;

[0058] S4: Model output: Output the acquired 3D model in obj format.

[0059] Attachment Figure 4 It demonstrates how to obtain a 3D model based on 2D images from multiple perspectives. The specific implementation steps are:

[0060] S1: Software installation and environment configuration: Install agisoft metashape software on the computer, and choose version 2.2;

[0061] S2: Photo acquisition: Use a camera to acquire a two-dimensional image of the chair, taking 25 photos around the chair to ensure that all exterior surfaces of the product are covered without blind spots.

[0062] S3: Photo alignment: Use the built-in Workflow → Align Photos command in the software to align the pictures.

[0063] S4: Generate dense point cloud: Use the built-in Workflow → Build Dense Cloud command in the software to obtain the 3D point cloud data of the real chair;

[0064] S5: Generate mesh model: Use the built-in Workflow → Build Mesh command in the software to obtain the mesh model of the real chair;

[0065] S6: Model export: Export the chair 3D model in obj format and store it on the computer.

[0066] Example 4

[0067] KeyShot is a professional 3D visualization software based on physical rendering, widely used in product design presentations, industrial engineering verification, animation production, education and scientific research. In product design, it can quickly generate high-fidelity renderings for client proposals and marketing promotions. In industrial design, it can verify the feasibility of product appearance through material simulation and generate exploded views to assist in structural review. It also supports the production of product animations and VR / AR content, and can also be used for teaching demonstrations and scientific research visualization. This system uses this software to implement the addition of real products based on actual needs. The specific process is as follows:

[0068] S1: Software Installation: Install KeyShot 2023 software on your computer;

[0069] S2: Importing the real product 3D model: Import the 3D model in obj format obtained in Example 3 into the KeyShot software, ensuring that the coordinate system of the model is consistent with the world coordinate system;

[0070] S3: Background addition: Erase the virtual equipment elements through Example 2, then save the image in jpg format and import it into KeyShot software;

[0071] S4: Model adjustment: Adjust the model size, angle and lighting settings of the real product to obtain the effect of the real scene;

[0072] S5: Rendering: Set the resolution to 7680 pixels in width and 4046 pixels in height, output a high-definition image, and save the image data in jpg format.

[0073] The specific implementation method described above provides a detailed description of the objectives, technical solutions and beneficial effects of the present invention. It should be understood that the above is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The autonomous evolutionary algorithm and its application system for intelligent agents in commercial scenarios are applied to the field of gym equipment layout. They can intelligently post-process AI-generated renderings, achieve seamless integration between virtual designs and real products, and provide users with a what-you-see-is-what-you-get equipment layout solution. The specific implementation steps are as follows: S1: Virtual rendering generation: Based on the keywords input by the user, a virtual rendering is generated using an AI model; the keywords include: venue area, equipment requirements, style, and budget; the AI ​​model uses Stable Diffusion; S2: Equipment element acquisition: using a machine vision algorithm to identify the type and corresponding pixel coordinates of elements in the virtual rendering; the machine vision algorithm uses an image recognition algorithm based on deep learning; the elements include: a treadmill, a basketball stand, a badminton court, a table tennis table, a bench press, a spinning bike, and a gymnastics ball; S3: Product Matching: Segment the virtual elements obtained in S2 and save them as separate jpg images. Retrieve detailed information about the virtual elements from the product library, obtain 3D data on price, model, and color, and store it in a MySQL database. Provide three real product options for each virtual element based on appearance consistency, model compatibility, and price. S4: Product Replacement: Remove virtual elements through image erasure algorithms; use KeyShot to embed the real product model into the virtual rendering, and then adjust the model size, angle and lighting settings to obtain the rendering of the real scene; S5: User Feedback: Collect user evaluations and modification suggestions on real solutions, record the sales of recommended products, and enable autonomous learning and continuous evolution of the system; The image erasing algorithm is described as a deep neural network algorithm, which can erase the elements in the mask without damaging the original basic information of the image; the mask adopts the pixel coordinates obtained by the image recognition algorithm, The real three-dimensional model adopts the obj format. The construction method of the three-dimensional model can use a three-dimensional laser scanner to obtain high-precision three-dimensional point cloud data, and then obtain a three-dimensional model of the real product; the construction of the three-dimensional model can generate a three-dimensional model based on multi-perspective two-dimensional images.

Images