Rendering parameter optimization method and system for different terminals

Abstract

According to the rendering parameter optimization method and system for different terminals, a rasterization ray tracing mixed rendering pipeline and a rasterization-based forward rendering pipeline are designed and realized respectively for a high-end light tracing terminal and a middle-low-end light tracing-free terminal, and rendering parameters in the pipelines are optimized based on an NSGA-II algorithm; by adaptively adjusting the rendering parameters, the rendering quality can be maximized under the expected equipment frame rate, or the equipment frame rate can be maximized under the expected rendering quality, so that the balance between the equipment frame rate and the rendering quality on terminals with different performances is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of graphics rendering technology, and in particular relates to a rendering parameter optimization method and system for different terminals. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of Internet technology, multi-person online virtual assembly applications have become popular. In multi-user virtual assembly applications, different users need to enter the same virtual environment for collaborative interaction. This requires the system to provide fast, low-latency feedback to ensure that every user operation can be reflected in the virtual environment in a timely manner to avoid affecting the interaction efficiency due to delays or freezes.

[0004] However, in multi-user virtual assembly applications, the types of terminal devices used by users vary, from high-performance desktops to resource-constrained mobile devices, and the computing power of each terminal device also varies. Nivida launched the Turing architecture graphics card in 2018. The graphics card includes RT Core for accelerating ray tracing intersection, making real-time ray tracing a reality. It is suitable for high-performance host terminals, while mobile terminals are limited by bandwidth. The GPU rendering architecture is mainly TBDR, and the rendering efficiency is relatively low. It can be seen that the hardware performance of different terminal devices on the market varies.

[0005] High-performance devices can handle more complex graphics rendering to provide a more realistic VR experience. Some devices can even support hardware ray tracing to render more realistic light and shadows. However, mobile devices with limited resources need to simplify the rendering process to ensure smooth real-time feedback. Therefore, how to balance the rendering of picture quality and device frame rate for terminal devices with different performance is an issue that needs to be considered urgently. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a rendering parameter optimization method and system for different terminals. The present invention designs and implements a rasterization ray tracing hybrid rendering pipeline and a rasterization-based forward rendering pipeline for high-end ray tracing terminals and mid-to-low-end non-ray tracing terminals, respectively, and optimizes the rendering parameters in the pipeline based on the NSGA-II algorithm. By adaptively adjusting the rendering parameters, the rendering quality can be maximized at the expected device frame rate, or the device frame rate can be maximized at the expected rendering quality, thereby ensuring the balance between the device frame rate and rendering quality on terminals with different performance.

[0007] According to some embodiments, the present invention adopts the following technical solutions: A rendering parameter optimization method for different terminals includes the following steps: For terminal devices that support hardware ray tracing, a rasterization ray tracing hybrid pipeline is used. For terminal devices that do not support hardware ray tracing, a rasterization-based forward rendering pipeline is used. Optimize rendering parameters based on the number of SVGF iterations and the number of SPPs in the rasterization ray tracing hybrid pipeline; Optimize rendering parameters based on planar reflection texture resolution, rendering scale, shadow map resolution, and MSAA sampling number of the rasterization-based forward rendering pipeline; The terminal device finds the rendering parameters that maximize the rendering quality under the expected device frame rate according to the optimized rendering parameters, or finds the rendering parameters that maximize the device frame rate under the expected rendering quality, and performs rendering.

[0008] As an optional embodiment, the rasterization ray tracing hybrid pipeline is configured to generate a geometry cache using deferred rendering, perform path tracing based on the geometry cache, calculate global illumination, obtain a noisy image, filter out the noise, use an anti-aliasing algorithm to reduce aliasing and enhance temporal stability, and finally add a Bloom post-processing effect and output it to a frame buffer.

[0009] As an optional implementation, the rasterization-based forward rendering pipeline is configured to generate reflection effects using planar reflections, generate shadows using percentage neighbor filtering, reduce aliasing using MSAA, and finally add Bloom post-processing effects and output to the frame buffer.

[0010] As an optional implementation, before optimizing the rendering parameters, the rendering parameters are combined and encoded according to their quality levels. The rendering parameters of the rasterization ray tracing hybrid pipeline are encoded as follows: ,in is the number of spp, is the number of SVGF filter iterations; the rendering parameters of the rasterization-based forward rendering pipeline are encoded as ,in Scaling value for rendering, is the shadow map resolution, is the planar reflection texture resolution, The number of MSAA samples.

[0011] As a further implementation method, for the rasterization ray tracing hybrid pipeline, the SSIM value between the image rendered by the current pipeline and the image after the path tracing converges is used as the comprehensive image quality indicator, and the SSIM value between the image rendered by the current pipeline and the image after the path tracing converges and is filtered with the same number of SVGF iterations is used as the image denoising quality indicator; for the rasterization-based forward rendering pipeline, the image generated with the highest rendering parameters is used as the reference image, and the SSIM value between the image rendered by the current pipeline and the reference image is used as the image quality indicator.

[0012] As an optional implementation, before rendering, important viewpoints are arranged in different directions around the center of the scene, the viewpoints are oriented toward the center of the scene, and different viewpoints are arranged in each direction according to the radius from the center point; The rendering quality under all rendering parameter combinations is pre-calculated for each important viewpoint according to the quality evaluation standard.

[0013] As an optional implementation, NSGA-II is used to optimize rendering parameters, specifically including: Generate N different rendering parameters as the initial parent population; Perform fast non-dominated sorting on the parent population and divide them into different levels of non-dominated layers according to the performance of rendering parameters in rendering quality and device frame rate; Calculate the crowding distance for individuals in each non-dominated layer; A binary tournament is used to select N individuals for the subsequent crossover mutation stage. The selection principle is to give priority to individuals with higher non-dominance levels. If the non-dominance levels are the same, individuals with larger crowding distances are given priority. The selected individuals undergo a crossover operation according to the preset crossover rate, recombining the rendering parameters of the two parents to generate new offspring; The selected individuals undergo mutation operations according to the preset mutation rate, and new offspring are generated by fine-tuning the rendering parameters of the parent generation; After the offspring population is generated, the parent and offspring populations are merged to generate a new population, and the non-dominated sorting and crowding distance are calculated for the new population. Individuals with higher non-dominated layer ranks in the new population are preferentially retained. If the ranks are consistent, individuals with larger crowding distances are preferentially retained. Through multiple iterative optimizations, the Pareto optimal rendering parameter set for each important viewpoint is obtained.

[0014] As a further step, the calculation process of crowding distance includes: ; ; ; in and They are respectively the rendering quality SSIM value and the device frame rate, max and min represent the maximum and minimum values ​​of the corresponding parameters, is the crowding distance of the i-th rendering parameter in the rendering quality direction, is the crowding distance of the i-th rendering parameter at the device frame rate, is the total crowding distance of the i-th rendering parameter.

[0015] As an optional implementation, during the rendering process, the terminal device interpolates the rendering parameters of the current camera between the parameters of each important viewpoint according to the current camera's posture, and the interpolation process is performed frame by frame.

[0016] Furthermore, the interpolation process is calculated according to the following formula: ; in is the rendering parameter of the current camera, is a set of important viewpoints, An important point of view The rendering parameters, is the contribution weight of the viewpoint to the rendering parameters under the current camera pose, and the calculation formula is as follows:

[0017] ; ; Depend on and Composed and normalized; where is the weight in the direction, is the camera orientation, Direction for important viewpoints; is the weight on the distance, is the distance from the camera to the center of the scene, is the distance from the important viewpoint to the center of the scene, is the radius closer to the center of the scene, The radius is farther from the center of the scene.

[0018] A rendering parameter optimization system for different terminals, comprising: The terminal device uses a rasterization ray tracing hybrid pipeline for terminal devices that support hardware ray tracing, and uses a rasterization-based forward rendering pipeline for terminal devices that do not support hardware ray tracing; and is used to find the rendering parameters that maximize the rendering quality at the expected device frame rate according to the optimized rendering parameters, or find the rendering parameters that maximize the device frame rate at the expected rendering quality, and perform rendering; The cloud side is configured to optimize rendering parameters based on the number of SVGF iterations and the number of SPPs of the rasterization ray tracing hybrid pipeline; and to optimize rendering parameters based on the planar reflection texture resolution, rendering scale, shadow map resolution, and MSAA sampling number of the rasterization-based forward rendering pipeline.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a rendering parameter optimization method for different terminals, which can ensure the balance between device frame rate and rendering quality regardless of whether it is a high-performance ray tracing terminal or a low-performance mobile terminal. The present invention designs and implements a rasterization ray tracing hybrid rendering pipeline and a rasterization-based forward rendering pipeline for high-end ray tracing capable terminals and mid-to-low-end ray tracing-free terminals, respectively, and optimizes the rendering parameters in the pipeline based on the NSGA-II algorithm. By adaptively adjusting the rendering parameters, the rendering quality can be maximized at the desired device frame rate, or the device frame rate can be maximized at the desired rendering quality, thereby ensuring a balance between the device frame rate and rendering quality on terminals with different performance.

[0020] The implementation of the present invention is not limited to a specific API, and Vulkan, DirectX, OpengGL and other graphics APIs can be used; it does not depend on a specific platform, and Unity, UE or other self-developed engine frameworks can be used; the scope of application is not limited to a specific application, and can be used for virtual assembly, virtual roaming, games, etc.; the supported devices include but are not limited to PCs, tablets, mobile phones, etc., and it has broad application prospects.

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 A rendering pipeline diagram for the rasterization ray tracing hybrid rendering pipeline; Figure 2A schematic diagram of a rendering pipeline based on a rasterization-based forward rendering pipeline; Figure 3 Schematic diagram for rendering parameter optimization; Figure 4 Flowchart for optimizing parameters for NSGA-II; Figure 5 A timing diagram of rendering quality and device frame rate for obtaining rendering parameters for the cloud.

[0024] Figure 6 Pipeline optimization test results for the rasterization-ray tracing hybrid pipeline.

[0025] Figure 7 Pipeline optimization test results for the rasterization-based forward rendering pipeline. DETAILED DESCRIPTION

[0026] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0029] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0030] Embodiment 1 This embodiment proposes a rendering parameter optimization method for different terminals, which ensures the balance between frame rate and rendering quality on terminal devices with different performance. First, it is necessary to match rendering pipelines of different qualities according to the hardware performance of the user device. This paper designs two types of rendering pipelines. High-performance devices can use rasterization ray tracing hybrid pipelines to provide more realistic light and shadow, while low-performance devices use a lighter forward rendering pipeline to ensure smooth real-time feedback.

[0031] Figure 1The rendering pipeline of the hybrid rendering pipeline is demonstrated. The pipeline uses deferred rendering to generate geometry cache, path tracing to calculate global illumination, SVGF to filter noise, TAA to reduce aliasing and enhance temporal stability, and finally adds Bloom post-processing effects and outputs to the frame buffer. The entire pipeline exposes the number of spp and the number of SVGF filtering iterations for adjustment.

[0032] Figure 2 The rendering pipeline of the forward rendering pipeline is demonstrated. The pipeline uses planar reflection to generate reflection effects, uses percentage neighbor filtering to generate shadows, uses MSAA to reduce aliasing, and finally adds Bloom post-processing effects and outputs them to the frame buffer. The entire pipeline exposes the planar reflection texture resolution, shadow map resolution, MSAA sampling number, and rendering scaling value for adjustment.

[0033] Figure 3 This is a schematic diagram of rendering parameter optimization. When the software is started, it will select the appropriate pipeline according to the GPU architecture of the user's device. On a macro level, it ensures that the system can run on different devices and have good rendering effects. In order to further balance the rendering quality of the pipeline and the device frame rate, this paper fine-tunes the rendering parameters in the pipeline, selects several important viewpoints in the scene, and pre-calculates the rendering quality under different rendering parameters for each viewpoint. This calculation process is offline and only needs to be executed once for each scene. The results are stored in the cloud, and all users who use the virtual scene can directly access the data; then the NSGA-II algorithm is run in the cloud to find the optimal rendering parameter set on the Pareto front for the user's device. This stage is only executed once when the user starts the software for the first time, and the NSGA-II component drives the automatic collection of device frame rate information; finally, the user device performs adaptive rendering, obtains the rendering parameters that meet the requirements from the Pareto optimal parameter set of each important viewpoint, and interpolates between the parameters of each viewpoint according to the current user's camera posture, so as to find the rendering parameters that maximize the rendering quality at a specified frame rate or maximize the frame rate at a specified rendering quality. The specific optimization process includes the following steps: Step S101: For devices that support hardware ray tracing, a rasterization ray tracing hybrid pipeline is used; for devices that do not support hardware ray tracing, a rasterization-based forward rendering pipeline is used.

[0034] Step S102: Combining and encoding the rendering parameters according to the quality level of the rendering parameters. The rendering parameters of the rasterization ray tracing hybrid pipeline are encoded as ,in is the number of spp, is the number of SVGF filter iterations, and their value range covers different quality levels, as shown in Table 1; the rendering parameters of the forward rendering pipeline based on rasterization are encoded as ,in Scaling value for rendering, is the shadow map resolution, is the planar reflection texture resolution, are the number of MSAA samples, and their value range covers different quality levels, as shown in Table 2.

[0035] Step S103: It is necessary to establish quality evaluation standards for images rendered with different parameters. For the rasterization ray tracing hybrid pipeline, the SSIM value between the image rendered by the current pipeline and the image after the path tracing converges is used as the comprehensive image quality index, and the SSIM value between the image rendered by the current pipeline and the image after the path tracing converges and is filtered with the same number of SVGF iterations is used as the image denoising quality index; for the rasterization-based forward rendering pipeline, the image generated with the highest rendering parameters is used as the reference image, and the SSIM value between the image rendered by the current pipeline and the reference image is used as the image quality index.

[0036] Step S104: important viewpoints are arranged in different directions around the center of the scene. The viewpoints need to be oriented toward the center of the scene, and two viewpoints are arranged in each direction according to the radius from the center point.

[0037] Step S105: pre-calculating the rendering quality of all rendering parameter combinations for each important viewpoint according to the quality evaluation standard.

[0038] Step S106: In order to control the rendering pipeline in more detail, the rasterization ray tracing hybrid pipeline exposes the number of SVGF iterations and the number of spp, and optimizes the rendering parameters based on the three goals of image comprehensive quality, image denoising quality, and device frame rate based on the NSGA-II algorithm. The forward rendering pipeline exposes four parameters, namely, planar reflection texture resolution, rendering scale, shadow map resolution, and MSAA sampling number, and optimizes the rendering parameters based on the two goals of image quality and device frame rate based on the NSGA-II algorithm. Figure 4 The optimization process of NSGA-II is shown, which includes the following sub-steps: Step S1061: Generate N different rendering parameters as the initial parent population.

[0039] Step S1062: Perform fast non-dominated sorting on the parent population, and divide them into non-dominated layers of different levels according to the performance of rendering parameters in rendering quality and device frame rate. Figure 5To obtain the rendering quality and device frame rate timing diagram of the rendering parameters in the cloud, the cross-variation selection of NSGA-II may produce repeated rendering parameters. The SSIM and frame rate of the repeated rendering parameters have been calculated historically and can be directly obtained from the cache; for non-repeated rendering parameters, it is necessary to query the database to obtain their SSIM values; obtaining the device frame rate is relatively time-consuming, and the server needs to send a message requesting the frame rate first. The message indicates which viewpoint and rendering parameter frame rate currently need to be measured; after receiving the message, the client will set the camera to the specified posture and adjust it to the specified rendering parameters. Since switching rendering parameters will cause the frame rate to fluctuate, in order to obtain relatively accurate data, the client will first wait for 5 frames to allow the device frame rate to stabilize, and then count the frame rates of the next 10 frames, use the 3-sigma principle to eliminate abnormal data, and finally calculate the average frame rate and return it to the server.

[0040] Step S1063: Calculate the crowding distance of each individual in the non-dominated layer. The crowding distance is calculated according to the following formula:

[0041]

[0042]

[0043] in and They are rendering quality SSIM value and device frame rate, is the crowding distance of the i-th rendering parameter in the rendering quality direction, is the crowding distance of the i-th rendering parameter at the device frame rate, is the total crowding distance of the i-th rendering parameter, which is equal to the sum of the crowding distance of the rendering quality and the crowding distance of the device frame rate.

[0044] Step S1064: In order to increase population diversity, the selected individuals are mutated according to a preset mutation rate, and new offspring are generated by fine-tuning the rendering parameters of the parent generation.

[0045] Step S1065: the selected individuals are crossovered according to a preset crossover rate, the rendering parameters of the two parent generations are reorganized, and a new child generation is generated; Step S1066: In order to increase population diversity, the selected individuals are mutated according to a preset mutation rate, and new offspring are generated by fine-tuning the rendering parameters of the parent generation; Step S1067: After the offspring population is generated, the parent and offspring populations are merged to generate , and Calculate non-dominated sorting and crowding distance, and keep priority For individuals with higher ranks in the non-dominant layer, if their ranks are the same, individuals with larger crowding distances are preferred.

[0046] Step S1068: Through multiple iterations of optimization, a Pareto optimal rendering parameter set for each important viewpoint is returned to the client. Step S107: The client performs adaptive rendering based on the Pareto parameter set, including the following sub-steps: Step S1071: Based on the returned Pareto optimal rendering parameter set, find the rendering parameters that maximize the rendering quality under the expected device frame rate, or find the rendering parameters that maximize the device frame rate under the expected rendering quality. The above operation needs to be performed for each important viewpoint to find the optimal parameters.

[0047] Step S1072: Interpolate the rendering parameters of the current camera between the parameters of each important viewpoint according to the current camera's position and posture. The interpolation process needs to be performed every frame and is calculated according to the following formula:

[0048] in is the rendering parameter of the current camera, is a set of important viewpoints, An important point of view The rendering parameters, is the contribution weight of the viewpoint to the rendering parameters under the current camera pose, and the calculation formula is as follows:

[0049]

[0050]

[0051] Depend on and Composed and normalized; where is the weight in the direction, is the camera orientation, Direction for important viewpoints; is the weight on the distance, is the distance from the camera to the center of the scene, is the distance from the important viewpoint to the center of the scene, is the radius closer to the center of the scene, The radius is farther from the center of the scene.

[0052] In summary, this paper conducts a real-machine test on the rendering parameter optimization method for different terminals. Figure 6This is the pipeline optimization test effect of the rasterization ray tracing hybrid pipeline. The test model is NVIDIA GeForce RTX 3080Laptop GPU, the CPU is AMD Ryzen 9 5900HX, the resolution is 1920x1080, the rendering frame rate with the highest denoising quality is 37 FPS, the denoising quality SSIM is 0.98576, and the comprehensive quality SSIM is 0.94237. The rendering frame rate with parameter optimization is 51 FPS, the denoising quality SSIM is 0.98199, and the comprehensive quality SSIM is 0.95868. Compared with the rendered image with the highest denoising quality, the denoising quality of the rendered image after parameter optimization is only reduced by 0.382%, but the frame rate is increased by 14fps (increased by about 37.84%), and the comprehensive quality is improved by 1.73%. It can be observed that the clarity of the ground reflection has been significantly improved.

[0053] Figure 7 This is the pipeline optimization test effect of the rasterization-based forward rendering pipeline. The test model is Xiaomi Tablet 7, the GPU is Adreno(TM) 732, the CPU architecture is ARM 64, the resolution is 3200x2136, the highest quality rendering frame rate is 49.8FPS, the quality SSIM is 1, the parameter optimized rendering frame rate is 67 FPS, and the quality SSIM is 0.99433. It can be found that the SSIM difference between the parameter optimized rendering image and the highest quality rendering image is not much. After parameter optimization, the quality is only reduced by 0.567%, but the frame rate is increased by 17.2 FPS (about 34.5% increase) compared with the highest configuration parameters.

[0054] Table 1

[0055] Table 2

[0056] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of one or more computer-usable storage media (including but not limited to disk storage, CD - ROM , optical storage, etc.).

[0057] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0058] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0059] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0060] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modifications, equivalent replacements, improvements, etc. made by those skilled in the art without creative efforts within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A rendering parameter optimization method for different terminals, characterized in that: The following steps are involved: For terminal devices that support hardware ray tracing, a rasterization ray tracing hybrid pipeline is used. For terminal devices that do not support hardware ray tracing, a rasterization-based forward rendering pipeline is used. Optimize rendering parameters based on the number of SVGF iterations and the number of SPPs in the rasterization ray tracing hybrid pipeline; Optimize rendering parameters based on planar reflection texture resolution, rendering scale, shadow map resolution, and MSAA sampling number of the rasterization-based forward rendering pipeline; The terminal device finds the rendering parameters that maximize the rendering quality under the expected device frame rate according to the optimized rendering parameters, or finds the rendering parameters that maximize the device frame rate under the expected rendering quality, and performs rendering.

2. A rendering parameter optimization method for different terminals as claimed in claim 1, characterized in that: The rasterization ray tracing hybrid pipeline is configured to generate a geometry cache using deferred rendering, perform path tracing based on the geometry cache, calculate global illumination, obtain a noisy image, filter out the noise, use an anti-aliasing algorithm to reduce aliasing and enhance temporal stability, and finally add a Bloom post-processing effect and output it to a frame buffer.

3. The rendering parameter optimization method for different terminals as claimed in claim 1, characterized in that: The rasterization-based forward rendering pipeline is configured to generate reflection effects using planar reflections, generate shadows using percentage proximity filtering, reduce aliasing using MSAA, and finally add Bloom post-processing effects and output to the frame buffer.

4. The rendering parameter optimization method for different terminals as claimed in claim 1, characterized in that: Before optimizing the rendering parameters, the rendering parameters are combined and encoded according to their quality levels. The rendering parameters of the rasterization ray tracing hybrid pipeline are encoded as follows: ,in is the number of spp, is the number of SVGF filter iterations; The rendering parameters of the rasterization-based forward rendering pipeline are encoded as ,in Scaling value for rendering, is the shadow map resolution, is the planar reflection texture resolution, The number of MSAA samples.

5. The method for optimizing rendering parameters for different terminals as claimed in claim 4, characterized in that: For the rasterization ray tracing hybrid pipeline, the SSIM value between the image rendered by the current pipeline and the image after path tracing convergence is used as the comprehensive image quality indicator, and the SSIM value between the image rendered by the current pipeline and the image after path tracing convergence and filtering with the same number of SVGF iterations is used as the image denoising quality indicator; For the rasterization-based forward rendering pipeline, the image generated by the highest rendering parameters is used as the reference image, and the SSIM value between the current pipeline rendered image and the reference image is used as the image quality indicator.

6. The method for optimizing rendering parameters for different terminals as claimed in claim 1, characterized in that: Before rendering, important viewpoints are arranged in different directions around the center of the scene. The viewpoints face the center of the scene, and different viewpoints are arranged in each direction according to the radius from the center point. The rendering quality under all rendering parameter combinations is pre-calculated for each important viewpoint according to the quality evaluation standard.

7. The rendering parameter optimization method for different terminals as claimed in claim 1, characterized in that: Use NSGA-II to optimize rendering parameters, including: Generate N different rendering parameters as the initial parent population; Perform fast non-dominated sorting on the parent population and divide them into different levels of non-dominated layers according to the performance of rendering parameters in rendering quality and device frame rate; Calculate the crowding distance for individuals in each non-dominated layer; A binary tournament is used to select N individuals for the subsequent crossover mutation stage. The selection principle is to give priority to individuals with higher non-dominance levels. If the non-dominance levels are the same, individuals with larger crowding distances are given priority. The selected individuals undergo a crossover operation according to the preset crossover rate, recombining the rendering parameters of the two parents to generate new offspring; The selected individuals undergo mutation operations according to the preset mutation rate, and new offspring are generated by fine-tuning the rendering parameters of the parent generation; After the offspring population is generated, the parent and offspring populations are merged to generate a new population, and the non-dominated sorting and crowding distance are calculated for the new population. Individuals with higher non-dominated layer ranks in the new population are preferentially retained. If the ranks are consistent, individuals with larger crowding distances are preferentially retained. Through multiple iterative optimizations, the Pareto optimal rendering parameter set for each important viewpoint is obtained.

8. The method for optimizing rendering parameters for different terminals as claimed in claim 7, characterized in that: The calculation process of crowding distance includes: ； ； ； in and They are respectively the rendering quality SSIM value and the device frame rate, max and min represent the maximum and minimum values ​​of the corresponding parameters, is the crowding distance of the i-th rendering parameter in the rendering quality direction, is the crowding distance of the i-th rendering parameter at the device frame rate, is the total crowding distance of the i-th rendering parameter.

9. The method for optimizing rendering parameters for different terminals as claimed in claim 1, characterized in that: During the rendering process, the terminal device interpolates the rendering parameters of the current camera between the parameters of each important viewpoint based on the current camera's position, and the interpolation process is performed frame by frame; The interpolation process is calculated according to the following formula: ； in is the rendering parameter of the current camera, is a set of important viewpoints, An important point of view The rendering parameters, is the contribution weight of the viewpoint to the rendering parameters under the current camera pose, and the calculation formula is as follows: ； ； Depend on and Composition and normalization; in is the weight in the direction, is the camera orientation, Direction for important viewpoints; is the weight on the distance, is the distance from the camera to the center of the scene, is the distance from the important viewpoint to the center of the scene, is the radius closer to the center of the scene, The radius is farther from the center of the scene.

10. A rendering parameter optimization system for different terminals, characterized in that: include: The terminal device uses a rasterization ray tracing hybrid pipeline for terminal devices that support hardware ray tracing, and uses a rasterization-based forward rendering pipeline for terminal devices that do not support hardware ray tracing; and is used to find the rendering parameters that maximize the rendering quality at the expected device frame rate according to the optimized rendering parameters, or find the rendering parameters that maximize the device frame rate at the expected rendering quality, and perform rendering; The cloud side is configured to optimize rendering parameters based on the number of SVGF iterations and the number of SPPs of the rasterization ray tracing hybrid pipeline; and to optimize rendering parameters based on the planar reflection texture resolution, rendering scale, shadow map resolution, and MSAA sampling number of the rasterization-based forward rendering pipeline.

Images