Animation image three-dimensional rendering analysis method based on artificial intelligence

Abstract

The invention discloses an animation image three-dimensional rendering analysis method based on artificial intelligence, and relates to the technical field of three-dimensional modeling and rendering, and the method comprises the steps: optimizing a geometry in a three-dimensional scene through a deep learning algorithm, enhancing the material of the geometry through a generative adversarial network model, and optimizing a rendering light source of the three-dimensional scene through reinforcement learning; configuring a sampling density strategy, adjusting the sampling density strategy by using Monte Carlo random sampling and a convolutional neural network model, and selecting an effective light path and a high-quality illumination map; performing preliminary image rendering by using the optimized geometry, the enhanced geometry material and the high-quality illumination chartlet; and image semantic segmentation is carried out through the U-Net, and later effect processing is carried out according to a semantic segmentation result and the SRGAN. According to the invention, three-dimensional scene rendering is optimized through a deep learning algorithm, a generative adversarial network and reinforcement learning, and efficient and intelligent three-dimensional animation image rendering is realized.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional modeling and rendering technology, and in particular to an artificial intelligence-based three-dimensional rendering analysis method for animation images. Background Art

[0002] With the development of computer graphics, 3D rendering technology for animated images has evolved from simple geometric shapes to complex scene simulations. Early technologies mainly relied on hand-made models and preset light sources, which limited the realism and detail of animated images. In recent years, breakthroughs in deep learning algorithms in the field of image processing have brought innovations to 3D rendering. In particular, deep learning has been used to optimize the geometric structure in 3D scenes, and generative adversarial networks have been used to enhance material texture, which has led to a qualitative leap in the quality of animated images. However, existing methods still face problems such as high computational cost and lack of realism when processing complex scenes. In particular, they have limitations in dynamically adjusting sampling density to adapt to the needs of different scenes.

[0003] Although existing technologies have made certain progress in the three-dimensional rendering of animated images, there are still several key issues that need to be addressed urgently. First, in terms of geometry optimization, traditional methods rely on manual adjustments or simple algorithm implementations, which make it difficult to accurately capture complex shapes and detailed features, resulting in insufficient realism and delicacy in the rendering effects. Secondly, although generative adversarial networks have shown great potential in material enhancement, their training process is complex and computationally expensive, and they perform poorly when processing multi-material mixed scenes, and are prone to texture distortion. Light source optimization is another challenge. Most current technologies use preset light source positions and intensities, lack dynamic adjustment capabilities, and are difficult to adapt to the needs of different scenes, affecting the realism and layering of the final rendering effects. Finally, with regard to sampling density strategies, existing methods usually adopt fixed or simple adaptive strategies, which cannot efficiently cope with complex three-dimensional scene changes, resulting in low rendering efficiency and unstable image quality. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an artificial intelligence-based three-dimensional rendering analysis method for animation images to solve the problems of low rendering efficiency and unstable image quality.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides an artificial intelligence-based three-dimensional rendering analysis method for animated images, which includes optimizing the geometric bodies in the three-dimensional scene through a deep learning algorithm, enhancing the material of the geometric bodies using a generative adversarial network model, and optimizing the rendering light source of the three-dimensional scene through reinforcement learning; configuring a sampling density strategy, using Monte Carlo random sampling and a convolutional neural network model, dynamically adjusting the sampling density strategy, generating and selecting effective light paths and high-quality light maps; using the optimized geometry, enhanced geometry materials and high-quality light maps to perform preliminary image rendering and generate preliminary rendered image results; performing image semantic segmentation through U-Net, and performing post-effect processing based on the semantic segmentation results and SRGAN to generate high-quality rendered images.

[0008] As a preferred solution of the artificial intelligence-based three-dimensional rendering analysis method for animated images of the present invention, the specific steps of optimizing the geometric bodies in the three-dimensional scene by using a deep learning algorithm are as follows:

[0009] Use PointNet++ to select a representative point as the center point in each layer of the 3D scene through far-point sampling;

[0010] Use spherical neighborhood query to find neighboring points around the center point and divide them into local areas;

[0011] Use a multi-layer perceptron to extract the geometric shape features, geometric normal direction features, and geometric curvature features of the neighboring points in the local area;

[0012] The geometric shape features, geometric normal direction features and geometric curvature features are aggregated into global features through maximum pooling.

[0013] As a preferred solution of the artificial intelligence-based three-dimensional rendering analysis method for animated images of the present invention, the method of enhancing the material of the geometric body by using the generative adversarial network model comprises the following steps:

[0014] Read color, reflectivity, and transparency from geometry as material data;

[0015] Input the material data into the generative adversarial network model as a training set;

[0016] Create multiple variations and adjust the rotation, scale, and brightness of material data through data augmentation;

[0017] The generator of the generative adversarial network model generates new material data based on the trained material data;

[0018] Reapplies the new material data's attributes and textures to the geometry, replacing the original material.

[0019] As a preferred solution of the artificial intelligence-based three-dimensional rendering analysis method of the present invention, the specific steps of optimizing the rendering light source of the three-dimensional scene by reinforcement learning are as follows:

[0020] Read light source data from the 3D scene through the rendering engine API;

[0021] Define the state, action, and reward of the three-dimensional scene environment;

[0022] Select a deep Q-learning model, use the light source data to generate a frame of illumination image, and render the illumination image in a 3D scene environment;

[0023] Evaluate the quality of the rendered lighting image through image processing, give rewards, and feed back to the deep Q learning model;

[0024] If the deep Q learning model detects that the shadow area of the illuminated image is too large, it will add light sources to fill the shadow area;

[0025] If the deep Q-learning model detects that the illuminated image is overexposed, it reduces the intensity of the light source;

[0026] Adjust light source position and direction based on the reflectivity and geometry layout of the rendered lit image.

[0027] As a preferred solution of the artificial intelligence-based 3D rendering analysis method for animated images of the present invention, the configuration of the sampling density strategy utilizes Monte Carlo random sampling and a convolutional neural network model to dynamically adjust the sampling density strategy, generate and select effective light paths and high-quality light maps, and specifically comprises the following steps:

[0028] Based on the optimized rendering light source, pre-rendering is performed to generate low-resolution light maps as the basis for evaluating the light intensity of different areas in the 3D scene;

[0029] Define sampling density strategy and set light intensity threshold;

[0030] In areas with high light intensity, when the light intensity exceeds the light intensity threshold, the sampling density is reduced;

[0031] In areas with low light intensity, if the light intensity does not exceed the light intensity threshold, the sampling density is increased;

[0032] Adjust the sampling density strategy based on the real-time light source data in the rendering;

[0033] Monte Carlo random sampling is used to randomly generate light paths from light source data, simulating the light path from the light source, passing through the geometric surface in the 3D scene, and reaching the camera rendering;

[0034] Analyze and simulate light paths through convolutional neural network models and select effective light paths for rendering;

[0035] Calculate the lighting contribution of light intensity in a 3D scene to obtain high-quality lightmaps.

[0036] As a preferred solution of the artificial intelligence-based three-dimensional rendering analysis method for animated images of the present invention, wherein: the light path is simulated by the convolutional neural network model analysis to select the effective light path for image rendering, the specific steps are as follows:

[0037] Calculate the total path length of light from the light source to the camera rendering;

[0038] When simulating the light path, the number of reflections of the light is counted to obtain the number of reflections;

[0039] Record the material reflectivity of each reflection of the light path;

[0040] Combine path length, number of reflections and material reflectivity into path characteristics;

[0041] Define the path features as the training set of the convolutional neural network model and input them into the convolutional neural network model;

[0042] The local features of the light path are extracted through the convolution layer as feature maps;

[0043] Reduce the dimension of the feature map through the pooling layer;

[0044] The feature map is converted into a light path importance score through a fully connected layer to select a valid light path.

[0045] As a preferred solution of the artificial intelligence-based three-dimensional rendering analysis method for animated images of the present invention, the steps of performing preliminary image rendering using the optimized geometry, enhanced geometry material, and high-quality lightmap to generate preliminary rendered image results are as follows:

[0046] Perform preliminary image rendering using optimized geometry and enhanced geometry materials;

[0047] Draw the initial image of each geometry through the graphics API;

[0048] Combines the high-quality lightmap with the color, reflectivity, and transparency of the material data to calculate the final color value for each pixel of the preliminary rendered image and applies it to the preliminary image of each geometry;

[0049] Combines the initial images of all geometries into an image frame as the preliminary rendered image result.

[0050] As a preferred solution of the artificial intelligence-based 3D rendering analysis method for animated images of the present invention, the image semantic segmentation is performed by U-Net, and post-processing is performed based on the semantic segmentation results and SRGAN to generate a high-quality rendered image. The specific steps are as follows:

[0051] Segment the preliminary rendered image using U-Net, identify the image semantics of the human area, building area, sky area, and background area in the preliminary rendered image, and generate a segmentation result;

[0052] According to the segmentation results, the character area is sharpened and the background area is blurred;

[0053] Adjust the hue, saturation, and brightness of the preliminary rendered image based on the image semantics of the human area, building area, sky area, and background area;

[0054] Use the segmentation results to simulate the depth of field of the camera;

[0055] Use the generator in SRGAN to convert the preliminary rendered image into a high-resolution image;

[0056] Use the discriminator in SRGAN to distinguish high-resolution images from real high-resolution images, and enhance the texture details and clarity of the preliminary rendered images;

[0057] Output high-quality rendered images through the generator in SRGAN.

[0058] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the artificial intelligence-based three-dimensional rendering and analysis method for animated images as described in the first aspect of the present invention is implemented.

[0059] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based three-dimensional rendering and analysis method for animated images as described in the first aspect of the present invention.

[0060] The beneficial effects of the present invention are as follows: the present invention optimizes three-dimensional scene rendering through deep learning algorithms, generative adversarial networks and reinforcement learning, thereby realizing efficient and intelligent three-dimensional animation image rendering. First, the deep learning algorithm is used to optimize geometric bodies, extract and aggregate information such as shape features, enhance the expressiveness of model details, provide high-quality data support for subsequent steps, and improve the realism of rendered images. Secondly, the generative adversarial network is used to enhance the geometric body material, create multiple variants, and make the material present natural effects under different lighting conditions, greatly enriching the expression form of objects and improving the quality of rendered images. Finally, reinforcement learning is used to dynamically adjust the light source configuration, and automatically optimize the light source position and intensity according to the rendering results, reducing manual intervention, improving shadow and overexposure problems, and enhancing the natural harmony of light and shadow effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0062] Figure 1 This is a flow chart of the artificial intelligence-based three-dimensional rendering analysis method for animated images in Example 1.

[0063] Figure 2 This is a flowchart for generating preliminary image rendering results in Example 1. DETAILED DESCRIPTION

[0064] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0065] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0066] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0067] Example 1, reference Figure 1 and Figure 2This embodiment provides an artificial intelligence-based three-dimensional rendering analysis method for animation images, comprising the following steps:

[0068] S1. Optimize the geometry in the 3D scene through deep learning algorithms, enhance the material of the geometry using generative adversarial network models, and optimize the rendering light source of the 3D scene through reinforcement learning.

[0069] Furthermore, we use PointNet++ to select representative points as center points in each layer of the 3D scene through far-point sampling.

[0070] Specifically, in each layer of the 3D scene, PointNet++ uses the far-point sampling method to select representative points. These points are selected as center points to ensure the uniform distribution of the point cloud in space and reduce redundant points.

[0071] Use spherical neighborhood query to find neighboring points around the center point and divide them into local areas;

[0072] Specifically, for each selected center point, PointNet++ uses a spherical neighborhood query to find all neighboring points within a certain radius around it. These neighboring points constitute a local area or neighborhood. In this way, PointNet++ establishes a local point set for each center point.

[0073] Use a multi-layer perceptron to extract the geometric shape features, geometric normal direction features, and geometric curvature features of the neighboring points in the local area;

[0074] Specifically, in each local area, PointNet++ applies a multi-layer perceptron to process these neighboring points. The multi-layer perceptron is responsible for extracting key geometric features from information such as the position and relative position of the points.

[0075] For each center point, PointNet++ has obtained a set of neighboring points through spherical neighborhood queries. Each point contains its coordinates (x, y, z) in three-dimensional space. These coordinates can be used directly as input to the multi-layer perceptron, or the input can contain additional attributes of the point such as color and normal vector.

[0076] The multi-layer perceptron accepts the coordinates of each point as input, and may also add the relative position information of the point, that is, the distance or vector from the point to the center point, which can better capture the local structure;

[0077] To enhance feature representation, the multi-layer perceptron embeds the original point coordinates into a higher-dimensional space. For example, the position vector (3D) of a point may be mapped to a higher-dimensional feature vector (such as 64 or 128). This process is achieved through linear transformation and nonlinear activation functions (such as ReLU).

[0078] The multilayer perceptron processes these points through multiple layers (usually 3 layers) of fully connected neural network layers. Each layer contains weights and biases. By learning to optimize these parameters, the multilayer perceptron can identify which input features are most useful for the shape, normal direction and curvature of the geometric body.

[0079] Specifically, through deep learning of the network, the multi-layer perceptron can capture the overall shape of the point cloud, such as the distribution and density of points, that is, the geometric shape features, which describe the overall distribution and structure of the points;

[0080] The multi-layer perceptron learns the pattern of normal vectors and obtains normal information, which reflects the orientation of the surface in space, that is, the normal direction characteristics of the geometric body;

[0081] By analyzing the relationship between points using a multi-layer perceptron, the multi-layer perceptron can infer the curvature of the surface and obtain the geometric curvature feature, which indicates the degree of surface change, such as whether the surface is flat or curved.

[0082] After processing each layer or three layers of fully connected neural network, the obtained geometric feature vector contains not only the local information of each point, but also the relationship with its neighboring points. The geometric features provide a rich geometric description for each local area.

[0083] For each point in the local area, after being processed by the multi-layer perceptron, a new feature vector is output, which represents the geometric features of the area;

[0084] The geometric shape features, geometric normal direction features and geometric curvature features are aggregated into global features through maximum pooling;

[0085] Specifically, PointNet++ uses maximum pooling to aggregate the features of all local regions (geometric shape features, geometric normal direction features, and geometric curvature features). For each feature (shape features, normal direction features, curvature features, etc.), maximum pooling selects the maximum value in each feature dimension. If a geometric feature vector has N dimensions, then maximum pooling selects the maximum value in the same dimension from all local regions, ensuring that the most significant geometric features are retained in the final global feature vector.

[0086] The advantage of this is that it can generate a global feature vector of fixed length regardless of how the number or distribution of local regions changes, enabling PointNet++ to process point clouds with a variable number of points;

[0087] Read color, reflectivity, and transparency from geometry as material data;

[0088] Color, reflectivity, and transparency are extracted directly, since each geometry has these materials attached to it;

[0089] Specifically, color is the color information extracted from the surface of each geometric object, usually in the form of RGB values or material maps;

[0090] Reflectivity is the reflection coefficient or reflection map of the geometry material, describing how light reflects on the surface;

[0091] Transparency reads the transparency value or transparency map to define the transparent or translucent properties of the material;

[0092] Input the material data into the generative adversarial network model as a training set;

[0093] Specifically, the material data is converted into image format and input into the generative adversarial network model;

[0094] All the extracted material data is organized into a training dataset, which contains the color, reflectivity and transparency information of all geometric bodies;

[0095] Create multiple variations and adjust the rotation, scale, and brightness of material data through data augmentation;

[0096] Specifically, the material data is rotated to generate a new perspective or form of expression;

[0097] Adjust the size of material data and create material samples of different resolutions;

[0098] Change the brightness of the material to simulate different lighting conditions and enrich the training material samples;

[0099] The generator of the generative adversarial network model generates new material data based on the trained material data;

[0100] Specifically, the generator learns material characteristics through trained material data, generates new texture details from the original material data, adjusts reflection and refraction characteristics, etc., to make the material look more natural and realistic;

[0101] Reapply the properties and textures of the new material data to the geometry, replacing the original material;

[0102] Directly read light source data (including point light, directional light, and ambient light) from the 3D scene through the rendering engine API;

[0103] Define the state, action, and reward of the three-dimensional scene environment;

[0104] Specifically, the state defines the configuration of all current light sources (position, direction, intensity), the position and material properties of geometric objects, etc.

[0105] Actions are defined such as moving a light source, changing the direction of a light source, increasing or decreasing the intensity of a light source, increasing or decreasing the number of light sources, etc.

[0106] The reward is a set reward mechanism. For example, improving the brightness of the 3D scene, reducing shadow areas, and enhancing the visibility of material details will receive positive rewards, while overexposure or excessive shadows will receive negative rewards.

[0107] Select a deep Q-learning model, use the light source data to generate a frame of illumination image, and render the illumination image in a 3D scene environment;

[0108] Specifically, the current light source data is used to render the 3D scene and generate a frame of lighting image. This is done directly in the rendering engine.

[0109] If iterative training is required, render once after each light source adjustment;

[0110] Evaluate the quality of the rendered illuminated image through image processing (such as contrast analysis, edge detection, brightness distribution, etc.), give rewards, and feed back to the deep Q learning model;

[0111] Specifically, image processing, such as connected domain analysis, is used to identify and measure the size of shadow areas. Pixel counting or shadow-based percentages are used to quantify shadow areas. Reducing shadow areas ensures that details and materials in the 3D scene are properly illuminated, improving visibility.

[0112] Evaluate the clarity of object details through edge detection (such as Canny), comparing the edge intensity or detail texture in the rendered illuminated image with the reference image or expected detail level. High detail visibility indicates that the details of the material and geometry are fully displayed in the rendering, enhancing the realism and beauty of the illuminated image;

[0113] By analyzing the brightness distribution of pixels in the illuminated image's histogram, pixels in overexposed areas are concentrated at the high brightness end (close to 255), indicating that these areas are too bright. On the contrary, pixels in underexposed areas are concentrated at the low brightness end (close to 0), indicating that these areas are too dark. In addition, HDR information or dynamic range indicators can be used to ensure that the image is not overexposed or underexposed, maintaining the natural lighting effect of the 3D scene and avoiding brightness distortion.

[0114] To calculate the histogram of the illuminated image, create an array called the histogram array. The length of the array is equal to the number of brightness values (such as 256 for an 8-bit grayscale image). Each position represents a brightness level from 0 to 255, and the array is initially filled with zeros.

[0115] Read the brightness value of each pixel in the illumination image. If it is a color image, convert the color image into grayscale value first;

[0116] Find the position of the corresponding brightness value in the histogram array and increase the counter of the position by one;

[0117] Look at the right end of the histogram (close to 255). If there are a lot of pixels, some parts of the illuminated image may be overexposed.

[0118] Observe the left end of the histogram (close to 0). If there are many pixels, the illuminated image may be underexposed.

[0119] Rewards are distributed based on the evaluation results, with high-quality renderings receiving high rewards and vice versa, low rewards or negative rewards;

[0120] Feedback the reward to the deep Q learning model and update the Q value table of the deep Q learning model;

[0121] If the deep Q learning model detects that the shadow area of the illuminated image is too large, it will add light sources to fill the shadow area;

[0122] If the deep Q-learning model detects that the illuminated image is overexposed, it reduces the light intensity to avoid overexposure and achieve a natural exposure effect;

[0123] Adjust the light source position and direction based on the reflectivity and geometry layout of the rendered lighting image;

[0124] Reflectivity is the reflectivity of a material under current lighting conditions, not an inherent property of the material itself;

[0125] Specifically, move or rotate the light source to better capture the reflectivity or reflection properties of the material;

[0126] Adjust the light source direction to distribute light more evenly on key geometry and highlight details;

[0127] It should be noted that by optimizing geometry through deep learning algorithms, enhancing materials using generative adversarial networks, and optimizing light sources through reinforcement learning, the present invention achieves comprehensive improvements in three-dimensional scene rendering. First, the precise extraction and optimization of geometric features improves the realism and detail expression of the model. Second, GAN enhances the authenticity and diversity of materials, allowing objects to display natural effects under various lighting conditions. Finally, reinforcement learning dynamically adjusts the light source configuration, reduces shadow and overexposure problems, and improves the natural harmony of light and shadow effects. Overall, these not only significantly improve the quality and visual effects of rendering, but also improve rendering efficiency.

[0128] S2. Configure the sampling density strategy, use Monte Carlo random sampling and convolutional neural network models to dynamically adjust the sampling density strategy, generate and select effective light paths and high-quality light maps.

[0129] Furthermore, a quick pre-rendering is performed based on the optimized rendering light source to generate a low-resolution light map as the basis for evaluating the light intensity of different areas in the 3D scene;

[0130] Define sampling density strategy and set light intensity threshold;

[0131] For example, light intensity above 90% can be considered as a high light area, and below 30% as a low light area;

[0132] In areas with high light intensity, when the light intensity exceeds the light intensity threshold, the sampling density is reduced. This is because the contribution of highlight areas to the light image is usually more stable and does not require high-density sampling to capture details. Adjustments such as reducing the number of sampling paths or sampling resolution can be made.

[0133] For example, it is stipulated that in places where the light intensity exceeds 80%, only 5 samples are taken per square pixel instead of the standard 10 times;

[0134] In areas with low light intensity, when the light intensity does not exceed the light intensity threshold, the sampling density is increased to ensure that the details of low-light or shadow areas can be accurately captured by increasing the number of light paths or improving the resolution of path sampling;

[0135] For example, in areas where the light intensity is less than 30%, 20 samples per square pixel might be taken to ensure that details in the shadows are captured;

[0136] Adjust the sampling density strategy based on real-time light source data during rendering. For example, if an area begins to have unexpected highlights, the sampling density in that area can be dynamically reduced. Conversely, if the lighting conditions in an area change and cause detail to be lost, the sampling density can be increased. A controller is used to automatically adjust the sampling density based on the current rendering results (such as through histogram analysis or brightness evaluation);

[0137] Monte Carlo random sampling is used to randomly generate light paths from light source data, simulating the light path from the light source, passing through the geometric surface in the 3D scene, and reaching the camera rendering;

[0138] Specifically, Monte Carlo random sampling is used to randomly generate a large number of light paths from the light source data. These paths simulate how light starts from the light source, passes through the three-dimensional scene, and undergoes reflection and refraction until it reaches the position rendered by the camera;

[0139] Analyze and simulate light paths through convolutional neural network models and select effective light paths for rendering;

[0140] Specifically, calculate the total path length of light from the light source to the camera rendering; the expression is:

[0141]

[0142] Among them, L total is the total path length from the light source to the camera, total represents the total path, d y is the length of the yth segment in the ray path, n is the total number of ray path segments, y represents starting from 1, representing the first segment or first point in the ray path, and increasing to n;

[0143] When simulating the light path, the number of reflections of the light is counted to obtain the number of reflections;

[0144] Specifically, every time a ray of light interacts with the surface of an object geometry in the 3D scene and reflects, the value of the reflection counter is increased to record the total number of reflections in the ray path, which reflects the complexity of the ray path and the number of times the light propagates;

[0145] Record the material reflectivity of each reflection of the light path;

[0146] Specifically, at each reflection or refraction, the material properties of the point or segment are queried and the reflectivity of the material is recorded;

[0147] Combine path length, number of reflections and material reflectivity into path characteristics;

[0148] Define the path features as the training set of the convolutional neural network model and input them into the convolutional neural network model;

[0149] Specifically, the path features (length, number of reflections, and reflectivity) are converted into structured data and standardized. The standardized path features are used as a training set and input into the convolutional neural network model.

[0150] The local features of the light path are extracted through the convolution layer as feature maps;

[0151] Specifically, a convolutional layer is used to scan the input path features to capture local spatial features, such as the direction change of the path and the distribution of reflectivity. Multiple convolutional layers are used to generate multiple feature maps, each of which represents a different level of abstraction of a feature.

[0152] Reduce the dimension of the feature map through the pooling layer;

[0153] Specifically, apply maximum pooling or average pooling to reduce the size of the feature map while retaining the most important features, such as the most significant number of reflections or material reflectivity areas;

[0154] The feature map is converted into a light path importance score through a fully connected layer to select a valid light path;

[0155] Specifically, the feature map processed by the convolutional and pooling layers is flattened into a one-dimensional feature vector, integrating information such as the total path length, number of reflections, and material reflectivity. The fully connected layer converts this feature vector into a score (between 0 and 1) through weight calculation, indicating the contribution of the light path to the preliminary rendered image. Based on the score, the light path that contributes most to the quality of the preliminary rendered image is selected for rendering.

[0156] Calculate the lighting contribution of light intensity in a 3D scene to obtain high-quality light maps;

[0157] Specifically, each light path starts from the light source, passes through the 3D scene, undergoes reflection and refraction, and finally reaches the camera rendering;

[0158] Through the attenuation function, the light intensity is obtained, and the expression is:

[0159]

[0160] Among them, I is the light intensity, which represents the intensity of the light source at a specific distance at that point, I0 is the initial light intensity of the light source, which represents the light intensity of the light source when there is no distance attenuation, d 2 The square of the distance is used because light intensity spreads in three-dimensional space, following the relationship of increasing area and increasing distance squared (for example, a light source emits rays, and the area covered by these rays increases at a square rate as the distance increases);

[0161] According to the reflectivity of the material, the light intensity reaching the surface of the geometry is calculated. The expression is:

[0162] I surf =I×R;

[0163] Among them, I surf is the intensity of light when it reaches the surface of the geometric body. It represents the intensity of light on a specific surface after attenuation. surf is the surface of the geometric body. R is the reflectivity of the surface of the geometric body, which represents the proportion or ability of the surface of the geometric body to reflect light. Its value is usually between 0 and 1. R = 1 means complete reflection (such as a mirror), R = 0 means no reflection (such as a completely absorbing black surface), and values between 0 and 1 represent partial reflection and partial absorption.

[0164] For each lighting path, the illumination intensity of each lighting path is calculated through the attenuation function to obtain the reflection and refraction characteristics of the material;

[0165] Monte Carlo simulation is used to simulate multiple reflections and accumulate the contribution of light to obtain the final light intensity, which is expressed as:

[0166] I path =I×∏R i ;

[0167] Among them, I path is the final illumination intensity after all reflections or refractions in the illumination path. Path is the entire light path. ∏ is the product operator, which means multiplying all these reflectivities or transmittances. R i is the reflectivity of each reflection, i is the index of the i-th reflection in the ray path;

[0168] For each pixel or geometric surface point of each high-quality light map, one or a series of high-quality light maps are generated by weighted accumulation of contributions from all lighting paths for subsequent rendering;

[0169] It should be noted that by configuring the sampling density strategy and using Monte Carlo random sampling and convolutional neural network models for dynamic adjustment, efficient and high-quality light path selection and high-quality light map generation are achieved. First, the light intensity threshold is set according to the pre-rendering results, and the sampling density of highlight and low-light areas is intelligently adjusted to ensure detail capture while reducing computing resource consumption. Then, the Monte Carlo method is used to simulate the light path, and the effective light path is selected through convolutional neural network analysis, which improves rendering efficiency and rendered image quality. Finally, high-quality light maps are generated based on the precisely calculated light contribution, which enhances the realism and visual effects of the scene, making three-dimensional animation rendering more efficient and realistic.

[0170] S3. Perform preliminary image rendering using the optimized geometry, enhanced geometry materials, and high-quality lightmaps to generate preliminary rendered image results.

[0171] Furthermore, preliminary image rendering is performed using the optimized geometry and enhanced geometry materials;

[0172] It should be noted that the preliminary rendered image here is a comprehensive effect of geometry, materials, and lighting. It is not equal to the lighting image. The lighting image is an evaluation of the lighting distribution and does not have the details of the material.

[0173] Draw the initial image of each geometry through the graphics API;

[0174] Specifically, the optimized geometry vertex data is fed into the vertex buffer of the graphics API;

[0175] Use a graphics API (such as OpenGL or DirectX) to issue drawing commands, rendering the vertex data into an initial image for each geometry using a triangle list or other primitive type;

[0176] Applying model matrix, view matrix, and projection matrix transformations to convert geometry from local coordinates to world coordinates, ensuring that the geometry is drawn in the correct position and perspective;

[0177] The model matrix transforms the geometry from its defined local space to the world space, including transformation operations such as translation, rotation, and scaling. For example, if there is a chair model, it can be placed in a certain position in the room through the model matrix and rotated to face the table.

[0178] The view matrix transforms all objects in the world space into the camera view space. It is created by the camera's position, direction, and up vector to perform position transformation and rotation of the camera.

[0179] The projection matrix projects the geometry in the view space into the clip space. There are two types of projection: perspective projection and orthographic projection. Perspective projection is used to simulate the visual effect of the human eye, and distant objects will become smaller. Orthographic projection keeps the size of the object unchanged regardless of the distance.

[0180] Use vertex and fragment shaders to handle the basic appearance of geometry, but do not consider complex lighting;

[0181] The high-quality light map is combined with the color, reflectivity, and transparency of the material data to calculate the final color value of each pixel of the preliminary rendered image and apply it to the initial image of each geometry (that is, in the fragment shader of the rendering pipeline, the intensity I of the light map is combined with the material data of the geometry, and these final color values are applied to the pixels of the initial image, replacing or overwriting the original preliminary color to form an image with lighting and material effects). The expression is:

[0182] C f =(C b ×I×R)×A+C g ×(1-A);

[0183] Among them, C f is the final color value, C b is the base color, the color sampled from the material map, A is the transparency, C g The background color is the background color at the current pixel or fragment position, and is the color that needs to be mixed when considering transparency.

[0184] Merge the initial images of all geometric bodies into an image frame as the preliminary rendered image result;

[0185] Specifically, the rendered initial images of all geometries are blended into one image frame through the frame buffer, and basic anti-aliasing techniques (such as MSAA-multi-sampling anti-aliasing) are applied to smooth the image edges;

[0186] Use multiple render targets to draw in layers, and then composite the final image as the preliminary rendered image result;

[0187] It should be noted that by using optimized geometry and enhanced materials for preliminary rendering, combined with high-quality light maps, the realism and visual effects of the three-dimensional scene can be significantly improved. The graphics API is used to efficiently draw geometry, and the model, view and projection matrices are applied to ensure correct spatial transformation and perspective performance. The color, reflectivity, transparency and light intensity are comprehensively considered in the shader, and the final color value of each pixel is accurately calculated to achieve delicate material performance and light and shadow effects. The image frames are merged through the frame buffer and anti-aliasing technology is used to smooth the edges to generate clear and realistic preliminary rendering image results.

[0188] S4. Perform image semantic segmentation through U-Net, and perform post-processing based on the semantic segmentation results and SRGAN to generate high-quality rendered images.

[0189] Furthermore, the preliminary rendered image is segmented through U-Net to identify the image semantics of the human area, building area, sky area and background area in the preliminary rendered image and generate the segmentation result;

[0190] Specifically, use the pre-trained U-Net model;

[0191] Convert the format of the preliminary rendered image to a size and format acceptable to the U-Net model (such as normalization). Usually, the U-Net model input is a 3-channel (RGB) image;

[0192] The input preliminary rendered image passes through a series of convolutional layers (encoders) to gradually reduce the resolution of the preliminary rendered image and extract features;

[0193] At the deepest part of the encoder, feature information is compressed;

[0194] The features are gradually restored to the resolution of the original preliminary rendered image through the upsampling layer (decoder).

[0195] The unique feature of the U-Net model is the "skip connection", which directly passes the high-resolution feature map of the encoding stage to the decoding stage to ensure that details are preserved;

[0196] Finally, the class label segmentation map of each pixel is output, converted into a probability map through the softmax layer, and then a threshold or argmax is applied to determine the category of each pixel (such as human area, building area, sky area, background area, etc.);

[0197] According to the segmentation results, the character area is sharpened and the background area is blurred to enhance the depth of field;

[0198] Specifically, a sharpening filter is applied to the segmented character area to enhance edge details, making the character appear clearer and more prominent in the image;

[0199] Apply a blur filter (such as Gaussian blur) to the background area to simulate a natural depth of field effect, blurring the background area relative to the foreground and increasing the contrast between the foreground and background.

[0200] Adjust the hue, saturation, and brightness of the preliminary rendered image based on the image semantics of the human area, building area, sky area, and background area;

[0201] Specifically, different tone mapping is applied according to the semantics of each area (people area, building area, sky area, and background area). For example, the color tone of the sky is adjusted to simulate the lighting effects of different time periods.

[0202] Increase or decrease the color saturation of specific areas, such as increasing the color saturation of buildings to make them more vivid, or reducing the saturation of the background to highlight the foreground;

[0203] Adjust brightness to make specific areas more or less visible, such as making architectural details stand out more, or darkening the background to enhance the sense of depth.

[0204] Use the segmentation results to simulate the depth of field of the camera;

[0205] Specifically, the segmentation results are combined with the previous blur and sharpening processing to accurately control the distribution of focus, ensuring that foreground objects (such as people) are clear, while the background (such as buildings and part of the sky) gradually becomes blurred, simulating the depth of field effect of a real camera;

[0206] Use the generator in SRGAN to convert the preliminary rendered image (i.e., low-resolution image) into a high-resolution image;

[0207] Use the discriminator in SRGAN to distinguish high-resolution images from real high-resolution images, and enhance the texture details and clarity of the preliminary rendered images;

[0208] Output high-quality rendered images through the generator in SRGAN;

[0209] Specifically, SRGAN consists of a generator and a discriminator. The generator generates a high-resolution image from a preliminary rendered image (i.e., a low-resolution image), while the discriminator evaluates whether the generated preliminary rendered image looks real.

[0210] The image to be enhanced (which can be a preliminary rendered image or a segmented portion of the preliminary rendered image) is fed into SRGAN as input, and the preliminary rendered image (i.e., a low-resolution image) is converted into a high-resolution image.

[0211] The generator of SRGAN learns to extract image features from the input preliminary rendered image through a deep convolutional network learning model, and uses the image features to generate a higher resolution version;

[0212] Specifically, the convolutional layer is used to extract the image features of the input preliminary rendered image;

[0213] Increase the initial rendered image resolution through upsampling layers while maintaining or enhancing details;

[0214] Using residual learning to gradually increase detail and clarity;

[0215] The discriminator distinguishes the generated high-resolution images from the real high-resolution images, helping the generator to generate more realistic details. Through this adversarial learning, the generator continuously improves its generation quality.

[0216] Input the rendered preliminary image or the rendered image with insufficient details into the trained SRGAN generator to directly generate a high-resolution and detailed high-quality rendered image;

[0217] The generator outputs a high-quality rendered image with significantly improved texture detail and clarity.

[0218] It should be noted that U-Net is used for image semantic segmentation to accurately identify and distinguish different areas (such as people, buildings, sky, etc.), and the sharpness, blur, hue and saturation of each area are adjusted in a targeted manner to enhance the sense of depth of field and visual hierarchy. The SRGAN technology is used to convert preliminary rendered images or rendered images with insufficient details into high-resolution images, significantly improving texture details and clarity, making the preliminary rendered images more realistic and delicate. This process not only enhances the overall visual effect of the preliminary rendered image, but also ensures the superiority of the preliminary rendered image in detail expression through meticulous local processing and high-quality amplification, and ultimately generates a high-quality rendered image that is both artistically beautiful and highly realistic.

[0219] This embodiment also provides a computer device suitable for the case of an artificial intelligence-based three-dimensional rendering and analysis method for animated images, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the artificial intelligence-based three-dimensional rendering and analysis method for animated images proposed in the above embodiment.

[0220] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0221] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for implementing the three-dimensional rendering and analysis of animated images based on artificial intelligence as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0222] In summary, the present invention optimizes three-dimensional scene rendering through deep learning algorithms, generative adversarial networks and reinforcement learning, and realizes efficient and intelligent three-dimensional animation image rendering. First, deep learning algorithms are used to optimize geometric bodies, extract and aggregate information such as shape features, enhance the expressiveness of model details, provide high-quality data support for subsequent steps, and improve the realism of rendered images. Secondly, generative adversarial networks are used to enhance geometric body materials and create multiple variants, so that the materials present natural effects under different lighting conditions, greatly enriching the expression of objects and improving the quality of rendered images. Finally, reinforcement learning is used to dynamically adjust the light source configuration, and automatically optimize the light source position and intensity according to the rendering results, reducing manual intervention, improving shadow and overexposure problems, and enhancing the natural harmony of light and shadow effects.

[0223] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based three-dimensional rendering analysis method for animated images, characterized by: include, Optimize the geometry in the 3D scene through deep learning algorithms, enhance the material of the geometry using generative adversarial network models, and optimize the rendering light source of the 3D scene through reinforcement learning; Configure sampling density strategies, use Monte Carlo random sampling and convolutional neural network models to dynamically adjust sampling density strategies, generate and select effective light paths and high-quality light maps; Perform preliminary image rendering using optimized geometry, enhanced geometry materials, and high-quality lightmaps to generate preliminary rendered image results; Image semantic segmentation is performed through U-Net, and post-processing is performed based on the semantic segmentation results and SRGAN to generate high-quality rendered images.

2. The artificial intelligence-based 3D rendering analysis method for animated images according to claim 1, wherein: The specific steps of optimizing the geometry in the three-dimensional scene through deep learning algorithms are as follows: Use PointNet++ to select a representative point as the center point in each layer of the 3D scene through far-point sampling; Use spherical neighborhood query to find neighboring points around the center point and divide them into local areas domain; Use a multi-layer perceptron to extract the geometric shape features, geometric normal direction features, and geometric curvature features of the neighboring points in the local area; The geometric shape features, geometric normal direction features and geometric curvature features are aggregated into global features through maximum pooling.

3. The artificial intelligence-based 3D rendering analysis method for animated images according to claim 2, wherein: The specific steps of using the generative adversarial network model to enhance the material of the geometric body are as follows: Read color, reflectivity, and transparency from geometry as material data; Input the material data into the generative adversarial network model as a training set; Create multiple variations and adjust the rotation, scale, and brightness of material data through data augmentation; The generator of the generative adversarial network model generates new material data based on the trained material data; Reapplies the new material data's attributes and textures to the geometry, replacing the original material.

4. The method for analyzing three-dimensional rendering of animated images based on artificial intelligence according to claim 3, wherein: The specific steps of optimizing the rendering light source of the three-dimensional scene through reinforcement learning are as follows: Read light source data from the 3D scene through the rendering engine API; Define the state, action, and reward of the three-dimensional scene environment; Select a deep Q-learning model, use the light source data to generate a frame of illumination image, and render the illumination image in a 3D scene environment; Evaluate the quality of the rendered lighting image through image processing, give rewards, and feed back to the deep Q learning model; If the deep Q learning model detects that the shadow area of the illuminated image is too large, it will add light sources to fill the shadow area; If the deep Q-learning model detects that the illuminated image is overexposed, it reduces the intensity of the light source; Adjust light source position and direction based on the reflectivity and geometry layout of the rendered lit image.

5. The method for analyzing three-dimensional rendering of animated images based on artificial intelligence according to claim 4, wherein: The configuration sampling density strategy uses Monte Carlo random sampling and convolutional neural network models to dynamically adjust the sampling density strategy, generate and select effective light paths and high-quality light maps. The specific steps are: Based on the optimized rendering light source, pre-rendering is performed to generate low-resolution light maps as the basis for evaluating the light intensity of different areas in the 3D scene; Define sampling density strategy and set light intensity threshold; In areas with high light intensity, when the light intensity exceeds the light intensity threshold, the sampling density is reduced; In areas with low light intensity, if the light intensity does not exceed the light intensity threshold, the sampling density is increased; Adjust the sampling density strategy based on the real-time light source data in the rendering; Monte Carlo random sampling is used to randomly generate light paths from light source data, simulating the light path from the light source, passing through the geometric surface in the 3D scene, and reaching the camera rendering; Analyze and simulate light paths through convolutional neural network models and select effective light paths for rendering; Calculate the lighting contribution of light intensity in a 3D scene to obtain high-quality lightmaps.

6. The method for analyzing three-dimensional rendering of animated images based on artificial intelligence according to claim 5, wherein: The convolutional neural network model is used to analyze and simulate the light path, and the effective light path for image rendering is selected. The specific steps are: Calculate the total path length of light from the light source to the camera rendering; When simulating the light path, the number of reflections of the light is counted to obtain the number of reflections; Record the material reflectivity of each reflection of the light path; Combine path length, number of reflections and material reflectivity into path characteristics; Define the path features as the training set of the convolutional neural network model and input them into the convolutional neural network model; The local features of the light path are extracted through the convolution layer as feature maps; Reduce the dimension of the feature map through the pooling layer; The feature map is converted into a light path importance score through a fully connected layer to select a valid light path.

7. The method for analyzing three-dimensional rendering of animated images based on artificial intelligence according to claim 6, wherein: The optimized geometry, enhanced geometry material and high-quality light map are used to perform preliminary image rendering to generate preliminary rendered image results. The specific steps are: Perform preliminary image rendering using optimized geometry and enhanced geometry materials; Draw the initial image of each geometry through the graphics API; Combines the high-quality lightmap with the color, reflectivity, and transparency of the material data to calculate the final color value for each pixel of the preliminary rendered image and applies it to the preliminary image of each geometry; Combines the initial images of all geometries into an image frame as the preliminary rendered image result.

8. The artificial intelligence-based 3D rendering analysis method for animated images according to claim 7, wherein: The image semantic segmentation is performed by U-Net, and post-processing is performed based on the semantic segmentation results and SRGAN to generate high-quality rendered images. The specific steps are: Segment the preliminary rendered image using U-Net, identify the image semantics of the human area, building area, sky area, and background area in the preliminary rendered image, and generate a segmentation result; According to the segmentation results, the character area is sharpened and the background area is blurred; Adjust the hue, saturation, and brightness of the preliminary rendered image based on the image semantics of the human area, building area, sky area, and background area; Use the segmentation results to simulate the depth of field of the camera; Use the generator in SRGAN to convert the preliminary rendered image into a high-resolution image; Use the discriminator in SRGAN to distinguish high-resolution images from real high-resolution images, and enhance the texture details and clarity of the preliminary rendered images; Output high-quality rendered images through the generator in SRGAN.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based three-dimensional rendering analysis method for animation images according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based three-dimensional rendering analysis method for animation images according to any one of claims 1 to 8 are implemented.

Images