Multi-dimensional AI rendering method capable of realizing continuous learning

Abstract

The invention discloses a multi-dimensional AI rendering method capable of continuous learning, and relates to the technical field of distributed rendering, and the method comprises the steps: 1, building a multi-dimensional distributed rendering model: adding the dimensions of task learning thinking on the basis of the distributed rendering model, including a hierarchical dimension and a rendering parameter model dimension; 2, rendering parameter model self-learning, wherein a rendering parameter model comprises a data capture module, a model learning module and a model application module; the data capture module is used for capturing element data, the model learning module is used for learning according to the captured data to form a rendering parameter intermediate model, then distilling the intermediate model to form a rendering parameter model capable of being applied, and storing the rendering parameter model in the model application module; and 3, expansion learning and iterative optimization of the rendering parameter model: carrying out continuous expansion learning on the application of the model on the basis of the rendering parameter model, and carrying out model iteration after learning is completed.

Description

Technical Field

[0001] This invention relates to the field of distributed rendering technology, and more specifically to a sustainable learning multi-dimensional AI rendering method. Background Technology

[0002] With the surge in demand for high-precision 3D scenes in film, games, and digital twins, single-machine rendering can no longer meet the requirements in terms of computing power and storage capacity. Distributed rendering, through task splitting and parallel computing, can significantly compress the lighting calculation time of complex models, achieving linear scaling of computing power. Distributed rendering achieves film-level effects that are difficult to achieve with traditional single-node methods by having multiple nodes collaboratively process complex lighting, particle effects, and physical simulations (such as cloth simulation and fluid dynamics).

[0003] With the development and application of AI technology, models are formed through learning and training on data. These models are then distilled into specialized models that ultimately support professional applications. The specialized models generated in this process can be viewed as different dimensions of thinking. The more specialized models used, the more dimensions of thinking there are, but this leads to increased response time and decreased response efficiency. Conversely, the fewer specialized models used, the more singular the dimensions of thinking become. However, due to path dependence and fixed mindsets, even if the background or conditions of the problem change, old models and solution patterns may still be used, leading to errors. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this invention is to propose a sustainable learning multi-dimensional AI rendering method that selects a better thinking dimension, thereby improving response accuracy while maintaining efficiency.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a sustainable learning multi-dimensional AI rendering method, comprising the following steps: Step 1: Construct a multi-dimensional distributed rendering model: Add a task learning and thinking dimension to the distributed rendering model, including a hierarchical dimension and a rendering parameter model dimension; Step Two: Rendering Parameter Model Self-Learning: The rendering parameter model includes a data capture module, a model learning module, and a model application module. The data capture module captures element data, and the model learning module learns from the captured data to form an intermediate rendering parameter model. This intermediate model is then distilled to form an applicable rendering parameter model, which is stored in the model application module. The model application module constructs a professional rendering parameter model knowledge graph, aligns the distilled professional models, and when the task's performance under the current model constraints fails to meet the user's requirements, it uses the alignment relationships between professional models to determine if other professional models offer better constraints. Step 3: Extended learning and iterative optimization of the rendering parameter model: Based on the rendering parameter model, continuously extend the learning for the application of the model, and iterate the model after the learning is completed.

[0006] As a preferred embodiment of the present invention, in step one: the layering dimension adopts layered rendering technology, which can divide the scene into different rendering layers and process them according to priority.

[0007] In a preferred embodiment of the present invention, in step one: the distributed rendering dimension is achieved by dividing the rendering task into multiple sub-tasks and distributing them to multiple computers or servers in the network for simultaneous calculation, and finally merging the results to generate a complete image.

[0008] As a preferred embodiment of the present invention, in step one: the rendering parameter model dimension constrains the model according to the design of various data entities (elements, code tables, model layers, models, etc.), and standardizes the attributes contained in each type of business entity, whether the attribute is mandatory, and the content constraints of the attribute.

[0009] In a preferred embodiment of the present invention, in step two: During the self-learning process of the rendering parameter model, an AD automatic differentiation strategy is adopted to compare the differences between new data and the original data. The intermediate model learns the differences in a targeted manner to form a professional rendering parameter model that can be applied, and saves and applies it in the model application module. The rendering parameter model is expanded and enriched to improve the accuracy of model constraints.

[0010] In a preferred embodiment of the present invention, in step two: The data captured includes user behavior, usage scenarios, user task types, task geometric properties, material parameters, posture parameters, and interaction data. User behavior includes the software used and the user's current rendering parameter configuration; usage scenarios include industrial design, game development, and home decoration design; user task types include, for example, design, rendering, and previewing; task geometric properties include normal vectors and camera view parameters; material parameters include reflectivity, roughness, and metallicity; and pose parameters include object trajectory and motion speed.

[0011] In a preferred embodiment of the present invention, the specific application method of the model application module in step two is as follows: When a user inputs an image of artistic text and wants to analyze its meaning, the system detects that the constraints of the current image recognition model cannot meet the user's needs. It then identifies the key components of the image, evaluates relevant models based on these key elements, and assesses the constraint capabilities of the text extraction model to determine if it can fulfill the user's task. If the model is determined to be able to recognize and extract the complete text, it is then used in conjunction with a semantic understanding model to interpret the extracted individual characters and understand the overall meaning of the text. If other models are found to have better constraints, the system combines multiple specialized models and integrates the constraints to ultimately complete the task.

[0012] In a preferred embodiment of the present invention, in step three: When the data capture module determines that the difference in data has reached a threshold, the rendering parameter model formed in the previous round is used as an intermediate model for updating and learning in the model learning module. This does not affect the use of the rendering parameter model at this stage. After learning is complete, the model is iterated, and the iterated model is called in the model application module for task processing. Based on changes in time and application, new elements are added to the rendering parameter model, integrating it into a multi-dimensional distributed rendering model, thus influencing and constraining the rendering results.

[0013] Compared with existing technologies, the beneficial effects of this invention are: by using multi-dimensional rendering constraints, a better thinking dimension can be selected, which can improve the response accuracy while taking efficiency into account, and avoid model path dependence or fixed mindset. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the model architecture of the multi-dimensional AI rendering method in this embodiment.

[0015] Figure 2 This is a flowchart illustrating the multi-dimensional AI rendering method in this embodiment. Detailed Implementation

[0016] The technical solution of the present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1-2 As shown in this embodiment, a multi-dimensional AI rendering method for sustainable learning is disclosed, including: 1. Construction of a multi-dimensional distributed rendering model The core of traditional distributed rendering lies in allocating rendering tasks to multiple computing nodes for parallel processing to improve rendering efficiency. Multidimensional distributed rendering models add "dimensions" for task learning and consideration (including layered dimensions, distribution dimensions, and rendering parameter model dimensions) to the traditional model. Layered rendering technology can divide the scene into different rendering layers and process them according to priority, such as processing the ground, buildings, and sky separately, improving rendering efficiency in complex scenes. The rendering parameter model dimension invokes constraint models based on the design of various data entities (elements, code tables, model layers, models, etc.) to constrain the processing results of user input tasks, standardizing the attributes contained in each type of business entity, whether the attribute is mandatory, and the content constraints of the attribute. This constraint mechanism ensures the stability and consistency of the model, avoiding rendering errors or performance degradation caused by unreasonable model design.

[0018] 2. Rendering Parametric Model Learning Process First, a data capture module, a model learning module, and a model application module are set up. The data capture module captures element data (including various elements to be rendered). Based on the captured data, it learns to form an intermediate rendering parameter model. This intermediate model is then distilled to form an applicable rendering parameter model. During the rendering parameter model learning process, an automatic differentiation strategy is used to compare the differences between new and existing data. The intermediate model learns specifically from these differences, forming an applicable professional rendering parameter model, which is then saved and applied in the model application module. The rendering parameter model is then expanded and enriched to improve the accuracy of model constraints.

[0019] The data captured includes user behavior (e.g., software used, current rendering parameter configuration), usage scenarios (e.g., industrial design, game development, home decoration design); user task types (e.g., design, rendering, preview); task geometric attributes (e.g., normal vectors, camera view parameters), material parameters (e.g., reflectivity, roughness, metallicity), pose parameters (e.g., object trajectory, speed), and interaction data. This data is used to learn about the latest technological developments and user preferences.

[0020] The model application module constructs a knowledge graph of professional rendering parameter models, aligns the distilled professional models, and improves the generalization ability of the models in tasks. When the task cannot meet the user's needs under the constraints of the current model, the alignment relationship between professional models is used to determine whether other professional models have better constraints. For example, when a user inputs an artistic text image and wants to analyze its meaning, if the system detects that the constraints of the current image recognition model (which can only process the whole image) cannot meet the user's needs, it identifies the key components of the image, such as the strokes of the characters. It evaluates relevant models based on these key elements, and then assesses the constraint capabilities of the text extraction model (such as its ability to extract complete text from an image) to determine if it can fulfill the user's task. Once it is determined that the model can recognize and extract complete text, it is used in conjunction with a semantic understanding model to interpret the overall meaning of the extracted individual characters. If other models are determined to have better constraints, the system combines multiple professional models, comprehensively considering the constraints, to ultimately complete the task, avoiding getting bogged down in a single model constraint.

[0021] 3. Extended learning and iterative optimization of rendering parameter models Based on the rendering parameter model, continuous learning is performed to expand the model's application. When the data capture module determines that the difference in data has reached a threshold, the rendering parameter model formed in the previous round is used as an intermediate model for updating and learning in the learning module. At this time, it does not affect the use of the rendering parameter model. After the learning is completed, the model iterates, and the model application module calls the iterated model to process tasks. According to changes in time and application, new elements are added to the rendering parameter model, integrating the rendering parameter model into a multi-dimensional distributed rendering model, affecting the constrained rendering results.

[0022] For example, the parametric model version is defined as M(x,y), where x is defined as the model category, such as ray constraint model, mirror constraint model, etc., and each category is represented by a corresponding number, depending on the application scenario; y is defined as the number of updates, which is the number of updates. When M(1,1) is applied, if the difference data reaches a threshold, M(1,1) will undergo iterative learning to form M(1,2) and continue to be applied, and so on iteratively.

[0023] Although the present invention has been disclosed above with reference to preferred embodiments, the embodiments and accompanying drawings are not intended to limit the invention. Any person skilled in the art can make various changes or modifications without departing from the spirit and scope of the invention, and these changes will also be within the protection scope of the invention. Therefore, the protection scope of the present invention should be defined by the scope of the claims of this application.

Claims

1. A sustainable learning-based multi-dimensional AI rendering method, characterized in that, Includes the following steps: Step 1: Construct a multi-dimensional distributed rendering model: On the basis of the distributed rendering model, add the dimension of task learning and thinking, including the layered rendering dimension, the distributed rendering dimension, and the rendering parameter model dimension; Step Two: Rendering Parameter Model Self-Learning: The rendering parameter model includes a data capture module, a model learning module, and a model application module. The data capture module captures element data, and the model learning module learns from the captured data to form an intermediate rendering parameter model. This intermediate model is then distilled to form an applicable rendering parameter model, which is stored in the model application module. The model application module constructs a professional rendering parameter model knowledge graph, aligns the distilled professional models, and when the task's performance under the current model constraints fails to meet the user's requirements, it uses the alignment relationships between professional models to determine if other professional models offer better constraints. Step 3: Extended learning and iterative optimization of the rendering parameter model: Based on the rendering parameter model, continuously extend the learning for the application of the model, and iterate the model after the learning is completed.

2. The multi-dimensional AI rendering method for sustainable learning according to claim 1, characterized in that, In step one: Layered rendering uses layered rendering technology, which can divide the scene into different rendering layers and process them according to priority.

3. A sustainable learning multi-dimensional AI rendering method according to claim 1 or 2, characterized in that, In step one: Distributed rendering dimension divides the rendering task into multiple subtasks and distributes them to multiple computers or servers in the network for simultaneous computation, and finally merges the results to generate a complete image.

4. The multi-dimensional AI rendering method for sustainable learning according to claim 3, characterized in that, In step one: The rendering parameter model dimension calls the constraint model based on the design of various data entities, and standardizes the attributes contained in each type of business entity, whether the attribute is required, and the content constraints of the attribute.

5. The sustainable learning multi-dimensional AI rendering method according to claim 4, characterized in that: The data entities include elements, code tables, model hierarchies, and models.

6. The multi-dimensional AI rendering method for sustainable learning according to claim 1, characterized in that, In step two: During the self-learning process of the rendering parameter model, an AD automatic differentiation strategy is adopted to compare the differences between new data and the original data. The intermediate model learns the differences in a targeted manner to form a professional rendering parameter model that can be applied, and saves and applies it in the model application module. The rendering parameter model is expanded and enriched to improve the accuracy of model constraints.

7. A sustainable learning multi-dimensional AI rendering method according to claim 1 or 6, characterized in that, In step two, the data captured includes user behavior, usage scenarios, user task types, geometric attributes of tasks, material parameters, posture parameters, and interaction data. User behavior includes the software used and the user's current rendering parameter configuration; usage scenarios include industrial design, game development, and home decoration design; user task types include, for example, design, rendering, and previewing; task geometric properties include normal vectors and camera view parameters; material parameters include reflectivity, roughness, and metallicity; and pose parameters include object trajectory and motion speed.

8. The multi-dimensional AI rendering method for sustainable learning according to claim 1, characterized in that, In step two, the specific application method of the model application module is as follows: When a user inputs an image of artistic text and wants to analyze its meaning, the system detects that the constraints of the current image recognition model cannot meet the user's needs. It then identifies the key components of the image, evaluates relevant models based on these key elements, and assesses the constraint capabilities of the text extraction model to determine if it can fulfill the user's task. If the model is determined to be able to recognize and extract the complete text, it is then used in conjunction with a semantic understanding model to interpret the extracted individual characters and understand the overall meaning of the text. If other models are found to have better constraints, the system combines multiple specialized models and integrates the constraints to ultimately complete the task.

9. A sustainable learning multi-dimensional AI rendering method according to claim 6, characterized in that, In step three: When the data capture module determines that the difference data has reached the threshold, the rendering parameter model formed in the previous round is used as an intermediate model to update and learn in the model learning module. At this time, the use of the rendering parameter model is not affected. After the learning is completed, the model is iterated, and the iterated model is called in the model application module to process the task.

10. A multi-dimensional AI rendering method for sustainable learning according to claim 9, characterized in that: Based on changes in time and application, new elements are added to the rendering parameter model, integrating the rendering parameter model into a multi-dimensional distributed rendering model, thus influencing the constrained rendering results.

Images