Spatial data structure optimization system and method based on artificial intelligence
Through an artificial intelligence-based system, the parameters of spatial data structures are automatically adjusted using machine learning and deep learning technology, the problem of insufficient performance of spatial data structure construction methods in the existing technology is solved, and efficient ray tracing rendering is achieved to adapt to different rendering scenarios.
Patent Information
- Application Number
- CN202411762231.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-03
- Publication Date
- 2025-03-25
AI Technical Summary
The existing spatial data structure construction methods rely on heuristic algorithms, which are difficult to achieve optimal performance and cannot be automatically adjusted to adapt to different rendering scenarios and needs.
Using an artificial intelligence-based system, including training modules, inference modules and rendering modules, optimized models are generated through machine learning and deep learning technologies, and the parameters of spatial data structures are automatically adjusted to optimize the calculation process of intersecting light and objects.
It has achieved automated optimization, which can adapt to different rendering scenarios and significantly improves the performance of ray tracing rendering.
Smart Images
Figure CN119693547A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of rendering technology, and in particular to an artificial intelligence-based spatial data structure optimization system and method. Background Art
[0002] In real-time rendering and ray tracing, spatial data structures such as bounding volume hierarchies (BVH) and octrees are key to accelerating the intersection calculation between rays and objects. However, existing methods for constructing spatial data structures often rely on heuristic algorithms, which makes it difficult to achieve optimal performance.
[0003] Therefore, the prior art has defects and needs to be improved. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a spatial data structure optimization system and method based on artificial intelligence to realize automatic adjustment of the construction process of the spatial data structure to adapt to different rendering scenarios and requirements.
[0005] The technical solution of the present invention is as follows: Provide a spatial data structure optimization system based on artificial intelligence, comprising:
[0006] Training module: includes at least one processor, which is used to process training data and generate an optimization model;
[0007] Reasoning module: comprising at least one processor, configured to receive data of a rendered scene and output optimization parameters of a spatial data structure according to an optimization model;
[0008] Rendering module: comprising at least one processor, configured to perform ray tracing rendering according to the optimized spatial data structure;
[0009] Storage unit: used to store data;
[0010] The training module, reasoning module and rendering module interact with each other through a communication interface.
[0011] Furthermore, there are three storage units, and the training module, the reasoning module, and the rendering module are each allocated one storage unit.
[0012] The processor is a CPU or a GPU or a specialized AI chip (such as a TPU).
[0013] The present invention also provides an artificial intelligence-based spatial data structure optimization method, which adopts the aforementioned artificial intelligence-based spatial data structure optimization system and includes the following steps.
[0014] S1: Training module processes training data and collects data through machine learning and deep learning techniques.
[0015] S2: The training module optimizes the generation of the model, uses machine learning and deep learning techniques to process the collected data and generate an optimized model.
[0016] S3: The inference module uses the optimization model to infer the received rendering scene data. After the inference is completed, the optimization parameters are output to the rendering module for optimizing the construction and configuration of the spatial data structure.
[0017] S4: The rendering module receives the optimization parameters, scene data and rendering instructions from the inference module.
[0018] S5: According to the optimization parameters output by the inference module, the rendering module adjusts the spatial data structure, such as a bounding volume hierarchy (BVH) or an octree, to optimize the calculation process of the intersection of light and objects.
[0019] S6: The rendering module uses the optimized spatial data structure for rendering.
[0020] S7: After rendering is completed, output the rendering results.
[0021] In step S1, specifically:
[0022] Data collection: Training data includes image data of rendered scenes, object properties, lighting conditions, etc. These data are collected through sensors, cameras or other data acquisition devices.
[0023] Data preprocessing: preprocess the collected data and increase the diversity of the data to improve the generalization ability of the model in preparation for model training; preprocessing includes processes such as cleaning, labeling, and formatting; increase the diversity of the data through operations such as rotation, scaling, and cropping.
[0024] Feature extraction: Extract features that are helpful for model training from preprocessed data, such as object boundaries, material type, light intensity, etc.
[0025] Build a model: Use linear regression to model the relationship between object properties and ray tracing performance.
[0026] In step S2, the following steps are specifically included:
[0027] Model selection: According to the characteristics of the rendering scene and the optimization goals, select a suitable machine learning or deep learning model, process image data such as screenshots of the rendering scene through a convolutional neural network (CNN), and extract useful features for optimizing the spatial data structure.
[0028] Model training: Use training data to train the selected model; the training process involves steps such as parameter initialization, loss function definition, and optimization algorithm selection. Through deep reinforcement learning Q-learning, the model is trained to select the optimal spatial data structure configuration to obtain the highest rendering performance reward.
[0029] Model Validation: Evaluate the performance of the model on an independent validation dataset to ensure that the model has good generalization capabilities; use recurrent neural networks (RNNs) to analyze time series data, such as changing trends in dynamic scenes, and predict real-time adjustments to spatial data structures.
[0030] Model optimization: Adjust model parameters and structure based on validation results to improve model accuracy and efficiency.
[0031] In step S3, the specific steps are:
[0032] Data reception: The inference module receives the current rendering scene data, including object position, attributes, lighting conditions, etc.
[0033] Data preprocessing: Perform necessary preprocessing on the input data to meet the input format requirements of the optimization model; this preprocessing includes steps such as data standardization and feature encoding.
[0034] Model reasoning: Use the optimization model generated by the training module to infer the preprocessed data and generate the optimization parameters of the spatial data structure.
[0035] Result output: The optimization parameters are output to the rendering module for optimizing the construction and configuration of the spatial data structure.
[0036] In step S6, the specific steps are: the rendering module executes the ray tracing algorithm, including the steps of ray generation, ray intersection with objects, light intensity calculation, color synthesis, rendering result output, etc.
[0037] The light generating step comprises:
[0038] Viewpoint determination: Determine the viewpoint of the rendering view based on the position and orientation of the virtual camera.
[0039] Pixel Sampling: Divides the view into a grid of pixels, and each pixel is further subdivided into smaller sampling points to increase the resolution of the rendering.
[0040] Ray emission: Rays are emitted from the viewpoint through each sampling point into the scene, simulating the interaction between sight and objects in the real world.
[0041] The step of intersecting the light and the object comprises:
[0042] Spatial data structure traversal: Use optimized spatial data structures (such as BVH or Octree) to speed up the intersection test between rays and objects.
[0043] Nearest neighbor search: For each ray, find the nearest intersection point to determine the surface that the ray encounters first.
[0044] Intersection calculation: Calculate the exact intersection point between the light and the surface of the object to obtain the intersection position and normal vector.
[0045] The steps for calculating light intensity include:
[0046] Light source detection: Determine all light sources in the scene, including point light sources, surface light sources, ambient light, etc.
[0047] Shadow calculation: Check if the intersection point is in the shadow of any light source, i.e. if there is another object blocking the light source.
[0048] Application of illumination model: Apply illumination models (such as Phong model and Blinn-Phong model) to calculate the illumination intensity of the intersection, including diffuse reflection, specular reflection and ambient light components.
[0049] The color synthesis steps include:
[0050] Material sampling: Calculate the color of the intersection based on the material properties of the object surface, such as diffuse reflectance, roughness, reflectivity, etc.
[0051] Texture Mapping: If the object has a texture, map the texture onto the intersection points to increase detail and realism.
[0052] Color Accumulation: If there are multiple light sources in the scene, the contributions of all light sources to the intersection color are accumulated.
[0053] Alpha synthesis: If the intersection point belongs to a transparent or semi-transparent object, calculate its color synthesis with the object behind it, and handle transparency and refraction effects.
[0054] The rendering result output step comprises:
[0055] Pixel color calculation: Average the colors of all sampling points within a pixel to obtain the final pixel color.
[0056] Image compositing: combining the colors of all pixels into the final rendered image.
[0057] Post-processing effects: Apply post-processing effects such as blur, glow, tone mapping, etc. to enhance the visual appeal of your images.
[0058] In step S7, the rendering module outputs the final rendered image, which is displayed on the screen or saved as an image file.
[0059] The present invention improves the efficiency and performance of spatial data structures in ray tracing rendering through machine learning and deep learning techniques. The system includes a training module for generating an optimization model; an inference module for adjusting the parameters of the spatial data structure according to the optimization model; and a rendering module for performing ray tracing rendering. The present invention realizes automatic adjustment of the construction process of the spatial data structure to adapt to different rendering scenarios and requirements.
[0060] By adopting the above scheme, the present invention provides a spatial data structure optimization system and method based on artificial intelligence, which has the following technical effects:
[0061] (1) Automated optimization: The system can automatically adjust the parameters of the spatial data structure without human intervention.
[0062] (2) Strong adaptability: Able to adapt to different rendering scenarios, including static and dynamic scenes.
[0063] (3) Performance improvement: The optimized spatial data structure can significantly improve the performance of ray tracing rendering. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0065] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0066] The present invention provides a spatial data structure optimization system based on artificial intelligence, comprising:
[0067] Training module: includes at least one processor, which is used to process training data and generate an optimization model;
[0068] Reasoning module: comprising at least one processor, configured to receive data of a rendered scene and output optimization parameters of a spatial data structure according to an optimization model;
[0069] Rendering module: comprising at least one processor, configured to perform ray tracing rendering according to the optimized spatial data structure;
[0070] Storage unit: used to store data;
[0071] The training module, reasoning module and rendering module interact with each other through a communication interface.
[0072] Furthermore, there are three storage units, and the training module, the reasoning module, and the rendering module are each allocated one storage unit.
[0073] The processor is a CPU or a GPU or a specialized AI chip (such as a TPU).
[0074] See also Figure 1 The present invention also provides an artificial intelligence-based spatial data structure optimization method, which uses the aforementioned artificial intelligence-based spatial data structure optimization system and includes the following steps.
[0075] S1: Training module processes training data and collects data through machine learning and deep learning techniques.
[0076] S2: The training module optimizes the generation of the model, uses machine learning and deep learning techniques to process the collected data and generate an optimized model.
[0077] S3: The inference module uses the optimization model to infer the received rendering scene data. After the inference is completed, the optimization parameters are output to the rendering module for optimizing the construction and configuration of the spatial data structure.
[0078] S4: The rendering module receives the optimization parameters, scene data and rendering instructions from the inference module.
[0079] S5: According to the optimization parameters output by the inference module, the rendering module adjusts the spatial data structure, such as a bounding volume hierarchy (BVH) or an octree, to optimize the calculation process of the intersection of light and objects.
[0080] S6: The rendering module uses the optimized spatial data structure for rendering.
[0081] S7: After rendering is completed, output the rendering results.
[0082] In step S1, specifically:
[0083] Data collection: Training data includes image data of rendered scenes, object properties, lighting conditions, etc. These data are collected through sensors, cameras or other data acquisition devices.
[0084] Data preprocessing: preprocess the collected data and increase the diversity of the data to improve the generalization ability of the model in preparation for model training; preprocess the data through cleaning, labeling and formatting; increase the diversity of the data through operations such as rotation, scaling and cropping.
[0085] Feature extraction: Extract features that are helpful for model training from preprocessed data, such as object boundaries, material type, light intensity, etc.
[0086] Build a model: Use linear regression to model the relationship between object properties and ray tracing performance.
[0087] In step S2, the following steps are specifically included:
[0088] Model selection: According to the characteristics of the rendering scene and the optimization goals, select a suitable machine learning or deep learning model, process image data such as screenshots of the rendering scene through a convolutional neural network (CNN), and extract useful features for optimizing the spatial data structure.
[0089] Model training: Use training data to train the selected model; the training process involves steps such as parameter initialization, loss function definition, and optimization algorithm selection. Through deep reinforcement learning Q-learning, the model is trained to select the optimal spatial data structure configuration to obtain the highest rendering performance reward.
[0090] Model Validation: Evaluate the performance of the model on an independent validation dataset to ensure that the model has good generalization capabilities; use recurrent neural networks (RNNs) to analyze time series data, such as changing trends in dynamic scenes, and predict real-time adjustments to spatial data structures.
[0091] Model optimization: Adjust model parameters and structure based on validation results to improve model accuracy and efficiency.
[0092] In step S3, the specific steps are:
[0093] Data reception: The inference module receives the current rendering scene data, including object position, attributes, lighting conditions, etc.
[0094] Data preprocessing: Perform necessary preprocessing on the input data to meet the input format requirements of the optimization model; this preprocessing includes steps such as data standardization and feature encoding.
[0095] Model reasoning: Use the optimization model generated by the training module to infer the preprocessed data and generate the optimization parameters of the spatial data structure.
[0096] Result output: The optimization parameters are output to the rendering module for optimizing the construction and configuration of the spatial data structure.
[0097] In step S6, the specific steps are: the rendering module executes the ray tracing algorithm, including the steps of ray generation, ray intersection with objects, light intensity calculation, color synthesis, rendering result output, etc.
[0098] The light generating step comprises:
[0099] Viewpoint determination: Determine the viewpoint of the rendering view based on the position and orientation of the virtual camera.
[0100] Pixel Sampling: Divides the view into a grid of pixels, and each pixel is further subdivided into smaller sampling points to increase the resolution of the rendering.
[0101] Ray emission: Rays are emitted from the viewpoint through each sampling point into the scene, simulating the interaction between sight and objects in the real world.
[0102] The step of intersecting the light and the object comprises:
[0103] Spatial data structure traversal: Use optimized spatial data structures (such as BVH or Octree) to speed up the intersection test between rays and objects.
[0104] Nearest neighbor search: For each ray, find the nearest intersection point to determine the surface that the ray encounters first.
[0105] Intersection calculation: Calculate the exact intersection point between the light and the surface of the object to obtain the intersection position and normal vector.
[0106] The steps for calculating light intensity include:
[0107] Light source detection: Determine all light sources in the scene, including point light sources, surface light sources, ambient light, etc.
[0108] Shadow calculation: Check if the intersection point is in the shadow of any light source, i.e. if there is another object blocking the light source.
[0109] Application of illumination model: Apply illumination models (such as Phong model and Blinn-Phong model) to calculate the illumination intensity of the intersection, including diffuse reflection, specular reflection and ambient light components.
[0110] The color synthesis steps include:
[0111] Material sampling: Calculate the color of the intersection based on the material properties of the object surface, such as diffuse reflectance, roughness, reflectivity, etc.
[0112] Texture Mapping: If the object has a texture, map the texture onto the intersection points to increase detail and realism.
[0113] Color Accumulation: If there are multiple light sources in the scene, the contributions of all light sources to the intersection color are accumulated.
[0114] Alpha synthesis: If the intersection point belongs to a transparent or semi-transparent object, calculate its color synthesis with the object behind it, and handle transparency and refraction effects.
[0115] The rendering result output step comprises:
[0116] Pixel color calculation: Average the colors of all sampling points within a pixel to obtain the final pixel color.
[0117] Image compositing: combining the colors of all pixels into the final rendered image.
[0118] Post-processing effects: Apply post-processing effects such as blur, glow, tone mapping, etc. to enhance the visual appeal of your images.
[0119] In step S7, the rendering module outputs the final rendered image, which is displayed on the screen or saved as an image file.
[0120] In summary, the present invention provides a spatial data structure optimization system and method based on artificial intelligence, which has the following technical effects:
[0121] (1) Automated optimization: The system can automatically adjust the parameters of the spatial data structure without human intervention.
[0122] (2) Strong adaptability: Able to adapt to different rendering scenarios, including static and dynamic scenes.
[0123] (3) Performance improvement: The optimized spatial data structure can significantly improve the performance of ray tracing rendering.
[0124] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A spatial data structure optimization system based on artificial intelligence, characterized in that: include: Training module: includes at least one processor, which is used to process training data and generate an optimization model; Reasoning module: comprising at least one processor, configured to receive data of a rendered scene and output optimization parameters of a spatial data structure according to an optimization model; Rendering module: comprising at least one processor, configured to perform ray tracing rendering according to the optimized spatial data structure; Storage unit: used to store data; The training module, reasoning module and rendering module interact with each other through a communication interface.
2. The spatial data structure optimization system based on artificial intelligence according to claim 1, characterized in that: There are three storage units, and the training module, the reasoning module, and the rendering module are each allocated a storage unit.
3. The spatial data structure optimization system based on artificial intelligence according to claim 1 is characterized in that: The processor is a CPU or a GPU or a dedicated AI chip.
4. A spatial data structure optimization method based on artificial intelligence, characterized in that: The artificial intelligence-based spatial data structure optimization system according to any one of claims 1 to 3 comprises the following steps: S1: Processing of training data in the training module, collecting data through machine learning and deep learning techniques; S2: The training module optimizes the generation of the model, using machine learning and deep learning techniques to process the collected data and generate an optimized model; S3: The inference module uses the optimization model to infer the received rendering scene data. After the inference is completed, the optimization parameters are output to the rendering module for optimizing the construction and configuration of the spatial data structure; S4: The rendering module receives the optimization parameters, scene data and rendering instructions from the reasoning module; S5: The rendering module adjusts the spatial data structure according to the optimization parameters output by the reasoning module; S6: The rendering module uses the optimized spatial data structure for rendering; S7: After rendering is completed, output the rendering results.
5. The method for optimizing spatial data structure based on artificial intelligence according to claim 4, characterized in that: In step S1, specifically: Data collection: Training data includes image data of rendered scenes, object properties, and lighting conditions; Data preprocessing: preprocess the collected data and increase the diversity of the data to improve the generalization ability of the model in preparation for model training; Feature extraction: extract features from preprocessed data that are helpful for model training; Build a model: Use linear regression to model the relationship between object properties and ray tracing performance.
6. The method for optimizing spatial data structure based on artificial intelligence according to claim 4, characterized in that: In step S2, the following steps are specifically included: Model selection: According to the characteristics of the rendering scene and the optimization goal, select the appropriate machine learning or deep learning model, process the image data through the convolutional neural network (CNN), and extract useful features for the optimization of the spatial data structure; Model training: Use the training data to train the selected model; through deep reinforcement learning Q-learning, train the model to select the optimal spatial data structure configuration to obtain the highest rendering performance reward; Model Validation: Evaluate the performance of the model on an independent validation dataset to ensure that the model has good generalization capabilities; analyze time series data through recurrent neural networks (RNNs) to predict real-time adjustments to spatial data structures; Model optimization: Adjust model parameters and structure based on validation results to improve model accuracy and efficiency.
7. The method for optimizing spatial data structure based on artificial intelligence according to claim 4, characterized in that: In step S3, the specific steps are: Data reception: The inference module receives the current rendering scene data, including object positions, attributes, and lighting conditions; Data preprocessing: Perform necessary preprocessing on the input data to meet the input format requirements of the optimization model; Model reasoning: Use the optimization model generated by the training module to reason about the preprocessed data and generate optimization parameters for the spatial data structure; Result output: The optimization parameters are output to the rendering module for optimizing the construction and configuration of the spatial data structure.
8. The method for optimizing spatial data structure based on artificial intelligence according to claim 4, characterized in that: In step S6, the specific steps are: the rendering module executes the ray tracing algorithm, including the steps of ray generation, ray intersection with objects, light intensity calculation, color synthesis, and rendering result output.
9. The method for optimizing spatial data structure based on artificial intelligence according to claim 8, characterized in that: The light generating step comprises: Viewpoint determination: Determine the viewpoint of the rendering view based on the position and orientation of the virtual camera; Pixel sampling: Divide the view into a pixel grid, and each pixel is further subdivided into smaller sampling points to increase the resolution of the rendering; Light emission: Light is emitted from the viewpoint through each sampling point into the scene, simulating the interaction between sight and objects in the real world; The step of intersecting the light and the object comprises: Spatial data structure traversal: Use optimized spatial data structures to speed up the intersection test between light and objects; Nearest neighbor search: Find the nearest intersection point for each ray to determine the surface that the ray encounters first; Intersection calculation: calculate the exact intersection point between the light and the surface of the object to obtain the intersection position and normal vector; The steps for calculating light intensity include: Light source detection: determine all light sources in the scene, including point light sources, surface light sources, and ambient light; Shadow calculation: check whether the intersection point is in the shadow of any light source, that is, whether there are other objects blocking the light source; Lighting model application: Apply the lighting model to calculate the lighting intensity of the intersection, including diffuse reflection, specular reflection and ambient light components; The color synthesis steps include: Material sampling: Calculate the color of the intersection based on the material properties of the object surface; Texture mapping: If the object has texture, map the texture to the intersection to increase detail and realism; Color accumulation: If there are multiple light sources in the scene, the contribution of all light sources to the intersection color is accumulated; Alpha synthesis: If the intersection point belongs to a transparent or semi-transparent object, calculate its color synthesis with the object behind it, and handle transparency and refraction effects; The rendering result output step comprises: Pixel color calculation: average the colors of all sampling points within a pixel to get the final pixel color; Image compositing: combining the colors of all pixels into the final rendered image; Post-processing effects: Apply post-processing effects such as blur, glow, tone mapping, etc. to enhance the visual appeal of your images.
10. The method for optimizing spatial data structure based on artificial intelligence according to claim 8, characterized in that: In step S7, the rendering module outputs the final rendered image, which is displayed on the screen or saved as an image file.