A priori based unbounded scene gaussian sputter decoupling optimization method
By using Gaussian initialization of skyboxes and sphere-aligned densification methods, the problems of view coupling and ray coupling in unbounded scene modeling are solved, improving rendering quality and modeling effect in sparse texture areas while maintaining real-time rendering performance.
Patent Information
- Application Number
- CN202610397514.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-30
- Publication Date
- 2026-06-26
AI Technical Summary
Existing 3D Gaussian splashing technology suffers from view coupling and ray coupling issues in unbounded scene modeling, resulting in floating artifacts and low rendering quality, especially in areas with sparse textures.
A densification method based on Gaussian initialization of the skybox and sphere alignment is adopted. The sky mask is generated by a pre-trained monocular depth model, the intersection of the projected ray and the skybox sphere is calculated, and the sampling variance is calculated by combining the Gaussian principal axis features to generate a new Gaussian element without depth perturbation, thereby avoiding Gaussian diffusion of the foreground into the sky region and reducing ray coupling.
It effectively solves the floating artifact problem in unbounded scene modeling, improves rendering quality and modeling effect in sparse texture areas, and maintains the real-time rendering advantage of 3D Gaussian splashing.
Smart Images
Figure CN122289537A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of scene modeling, specifically to a priori-based Gaussian sputtering decoupling optimization method for unbounded scenes. Background Technology
[0002] 3D Gaussian splashing is a 3D scene reconstruction and new perspective synthesis technology. This technology explicitly models a 3D scene as a large set of anisotropic 3D Gaussian primitives. Combined with a differentiable rasterization pipeline and an adaptive density control strategy, it achieves real-time rendering while ensuring high fidelity in scene rendering. With this core advantage, it has become a key technology for 3D reconstruction of outdoor boundless scenes in fields such as autonomous driving, augmented reality / virtual reality, and immersive media.
[0003] Because unbounded scenes contain a large number of regions with sparse textures and no actual boundaries, such as the sky, the initial point cloud generated by the motion reconstruction method has a serious problem of insufficient spatial points in such regions. This directly leads to obvious floating artifacts during the modeling process. As a result, the Gaussian in the foreground region splits and spreads to the sky region in the viewpoint, making the depth distribution of the foreground and the sky tend to be consistent, which seriously reduces the visual quality and geometric realism of the scene rendering.
[0004] To address the aforementioned issues in unbounded scene modeling, existing technologies primarily focus on improvements in three directions: First, depth regularization methods generate depth prior information through monocular depth estimation networks, providing spatial constraints for Gaussian model training. However, the inverse depth parameterization approach employed by these methods significantly reduces the optimization accuracy of far-field scenes, making it difficult to balance near-field and far-field modeling performance. Second, homogeneous coordinate optimization methods map the coordinates and scale of Gaussians to a unit sphere, achieving consistent numerical optimization of Gaussians in both near and far fields. However, these methods only adjust the optimization strategy at the numerical level and cannot fundamentally address the issue of sparse texture regions caused by initial values. The third issue is the floating artifact problem caused by insufficient data; the hybrid representation modeling method divides the scene into foreground and background parts, and models the background areas such as the sky separately. Some methods introduce neural radiation fields to model the sky area, but are limited by the rendering inefficiency of neural radiation fields, losing the core advantage of real-time rendering of 3D Gaussian splash. Other methods use hemispherical skyboxes to model the sky area, but because there are a lot of redundant Gaussian primitives in the skybox, the sky Gaussian will incorrectly absorb the color information of the foreground Gaussian during the blending stage of the rendering process, resulting in ray coupling problems, which seriously interfere with the normal rendering of the foreground area.
[0005] The core flaw of existing technology lies in its failure to address two key coupling effects in the optimization process of 3D Gaussian splashing in unbounded scenes: firstly, view coupling, where the foreground and sky are spatially adjacent in the training view projection, causing their optimization processes to become entangled, and the foreground Gaussian is prone to diffuse into the sky region during the densification stage, forming floating artifacts; secondly, ray coupling, where redundant Gaussian primitives in the background region absorb the color information of the foreground Gaussian during the alpha blending process of the optical path, causing foreground rendering distortion. Summary of the Invention
[0006] To address the aforementioned problems, this invention proposes a priori-based Gaussian sputtering decoupling optimization method for unbounded scenes. This method solves the technical problems of existing 3D Gaussian sputtering techniques in unbounded scene modeling, such as floating artifacts, low rendering quality, and poor modeling effects in sparse texture regions caused by the coupling between the foreground and sky regions. The technical solution provided by this invention is as follows:
[0007] A priori-based Gaussian sputtering decoupling optimization method for unbounded scenes includes the following steps:
[0008] Step 1: Obtain the initial point cloud and use a pre-trained monocular depth model to extract the binary sky mask of the training image; perform quantitative random sampling of the sky region; combine the camera intrinsic and extrinsic parameters to calculate the projection ray of the sampled pixel; solve for the intersection of the projection ray and the skybox sphere as the initial 3D center position of the skybox Gaussian, and inherit the color of the sampled pixel as the initial color of the skybox Gaussian.
[0009] Step 2: Introduce spherical tangential constraints, calculate the compaction direction of the Gaussian skybox, and calculate the sampling variance by combining the Gaussian principal axis features to generate a new Gaussian element without depth perturbation.
[0010] Preferably, the specific process of random sampling is as follows: calculate the pixel proportion of the sky region in each training image, set the hyperparameter to control the total number of sky boxes and Gaussians, allocate the sampling quantity to each image according to the proportion, and randomly sample the sky region to obtain a set of sampled pixels.
[0011] Preferably, the equation of the projected ray is:
[0012]
[0013] in, The location of the camera center in the world coordinate system. The direction of the projected ray in the world coordinate system. These are the ray parameters.
[0014] Preferably, the specific process for calculating the initial 3D center position of the skybox Gaussian is as follows:
[0015] To find the intersection point of the ray and the sphere, the equation of the sphere is expressed as:
[0016]
[0017] in The coordinates of a point in the world coordinate system. Let the center of the sphere be in the world coordinate system. Let be the radius of the sphere; substituting the equation of the projected ray into the equation of the sphere:
[0018]
[0019] Derive about The standard quadratic equation:
[0020]
[0021] in, , , After solving the equation, select the unique positive root. Substituting into the equation of the projected ray, we obtain the center coordinates of the Gaussian skybox in the world coordinate system.
[0022] Preferably, the specific process for calculating the compaction direction of the Gaussian skybox is as follows:
[0023] Obtain the unit vector from the Gaussian center of the skybox to the center of the sphere. The calculation formula is as follows:
[0024]
[0025] in, Let the coordinates be the center point of the skybox Gaussian. For a very small value, , These are components in three directions; construct a perpendicular to... auxiliary vector The calculation formula is as follows:
[0026]
[0027] in, The original principal axis direction of the Gaussian unit. For a perpendicular to The unit vector is calculated using the following formula:
[0028]
[0029] Calculate the splitting direction of the new Gaussian The calculation formula is as follows:
[0030]
[0031] auxiliary vector Considering the rotation axis vector as a reference, construct a rotation coordinate system and calculate the principal axis directions. and The angle between The calculation formula is as follows:
[0032]
[0033] Let the rotation quaternion of the principal axis of the original Gaussian ellipsoid relative to the coordinate axes of the world coordinate system be... Calculate the rotation increment :
[0034]
[0035] After obtaining the incremental quaternion, the target quaternion is calculated. The specific calculation process is as follows:
[0036]
[0037]
[0038]
[0039]
[0040]
[0041] Calculate based on the conversion formula between quaternions and rotation matrices:
[0042]
[0043] This is the final tangential rotation matrix.
[0044] Preferably, the sampling variance is obtained by calculating the sampling standard deviation along the extension direction:
[0045]
[0046] in, This represents the maximum axial length of the Gaussian ellipsoid.
[0047] Preferably, the center coordinates of the new Gaussian element are:
[0048]
[0049] in, The coordinates of the center of the new Gaussian after the split are: This is the offset. , The initial 3D offset in the principal axis coordinate system. , To conform to a mean of 0 and a standard deviation of The normal distribution is denoted by , where the components perpendicular to the principal axis are all set to 0.
[0050] Preferably, the initial point cloud is obtained using the structure-of-motion (SfM) method.
[0051] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: The present invention avoids generating Gaussian representations of the sky in the near area by introducing a skybox and depth constraints. At the same time, it uses a pre-trained sky segmentation network to generate a sky mask and initializes the skybox in the visible spherical region, avoiding the generation of a large number of redundant 3D Gaussian representations that incorrectly absorb the color of the foreground and form ray coupling.
[0052] This invention replaces the densification step of the skybox with a densification method that uses spherical alignment. Gaussian splitting is performed on the projection of the Gaussian ellipsoid onto the tangent plane along the principal axis of the Gaussian ellipsoid, maximizing the filling capability while minimizing depth perturbation. Attached Figure Description
[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0054] Figure 1 This is a schematic diagram of the Gaussian initialization of the skybox for spherical projection in this invention;
[0055] Figure 2 This is a schematic diagram of the Gaussian splitting mechanism of the skybox in this invention. Detailed Implementation
[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0057] To make the above-mentioned objectives, features and effects of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0058] Example 1: The general training process for a 3D Gaussian ellipsoid is as follows: Collect 100 to 500 static images of a scene from multiple perspectives. Based on these images, use the SfM algorithm to deduce the camera pose and point cloud data representing the approximate outline of the object. The point cloud data contains spatial location and color information. Dilate the point cloud into a Gaussian ellipsoid, that is, add the initial 3D Gaussian ellipsoid's 3D scale, rotation direction, and opacity information to the point cloud. Simultaneously, to simulate the effect of light transformation, the color is changed from three-dimensional RGB color to 3rd-order 45-dimensional spherical harmonic coefficients. Select 12.5% of the images as test images, and the remaining images as training images. During training, use the training images and rendered images to perform MSE and SSIM losses. Use the monocular depth estimation map of the training images and the depth map rendered by the Gaussian ellipsoid to perform L1 loss. The gradient of the loss function is used to iteratively optimize the various properties of the Gaussian ellipsoid through backpropagation. Test images are used when testing algorithm performance, but not during training, because the viewpoints of test images and training images are different. If the rendered image from the test viewpoint has a high similarity to the test image, it means that the performance is better.
[0059] This invention proposes a priori-based Gaussian sputtering decoupling optimization method for unbounded scenes. Addressing the pain points of existing technologies in unbounded scene modeling, it proposes a Gaussian initialization method for skyboxes and a Gaussian densification method for skyboxes with spherical alignment, including the following steps:
[0060] Step 1: Prior-based Gaussian skybox initialization (PBSI): Obtain the initial point cloud and extract the binary sky mask of the training image using a pre-trained monocular depth model; perform quantitative random sampling of the sky region; calculate the projection ray of the sampled pixel by combining the camera intrinsic and extrinsic parameters; solve for the intersection of the projection ray and the skybox sphere as the initial 3D center position of the skybox Gaussian, and inherit the color of the sampled pixel as the initial color of the skybox Gaussian.
[0061] This embodiment uses the SfM method (such as COLMAP) to obtain the initial point cloud, and uses the DepthAnything-3 model to generate a monocular depth map and the skybox initialization required for the sky mask.
[0062] Calculate the percentage of pixels in the sky region in each training image and set the hyperparameters. Control the total number of Gaussian skyboxes, allocate the number of samples to each image according to the proportion, randomly sample the sky region to obtain the sampled pixel set, and avoid generating redundant Gaussians.
[0063] According to camera internal parameters and the coordinates of the sampled pixels Calculate the direction of the projected ray in the camera coordinate system. The direction is transformed to the world coordinate system using the rotation matrix of the camera's extrinsic parameters. The expression for the projected ray originating from the camera center in the world coordinate system is obtained as follows:
[0064]
[0065] in, The location of the camera center in the world coordinate system. The direction of the projected ray in the world coordinate system. Let be the ray parameters. To obtain the initial coordinates of the skybox Gaussian, we need to solve for the intersection points of the ray and the sphere. The equation of the sphere is expressed as:
[0066]
[0067] in The coordinates of a point in the world coordinate system. Let the center of the sphere be in the world coordinate system. Let be the radius of the sphere. Then, substituting the ray equation into the sphere equation, we solve for the intersection of the ray and the sphere:
[0068]
[0069] This equation mathematically describes the intersection condition between the projected ray and the skybox sphere. By expanding and rearranging the terms to eliminate matrix transpose and quadratic forms, we can derive the equation regarding... The standard quadratic equation:
[0070]
[0071] in, , , .
[0072] After solving the equation, select the unique positive root. Substituting these coordinates into the ray equation yields the center coordinates of the skybox Gaussian in the world coordinate system. The initial color of the skybox Gaussian is directly inherited from the color of the sampled pixel.
[0073] like Figure 1 As shown, the ellipsoid closest to the semicircular curve represents the skybox 3D Gaussian ellipsoid, used to represent the sky, while the other ellipsoids below represent foreground objects. The dashed arrows represent the camera's line of sight; the areas between the darker rays represent the field of view; and the lighter rays in the middle represent the lowest line of sight where the sky is directly visible and not obscured by foreground objects. As can be seen, this invention initializes the 3D Gaussian ellipsoid based on the visible sky area.
[0074] It should be noted that, to prevent the sky Gaussian box from being occluded by the foreground Gaussian box during rendering, the initial opacity of the sky Gaussian box is set to 1.0 (adjusted to 0.99 during the rendering stage). Subsequently, for other initialization steps such as scaling and rotation, both the sky box Gaussian box and the foreground Gaussian box are executed in the same way.
[0075] Step 2, Skybox Gaussian Compacting with Spherical Alignment (SASD): By introducing spherical tangential constraints, the compaction direction of the skybox Gaussian is calculated. Combined with the Gaussian principal axis features, the sampling variance is calculated to generate new Gaussian elements without depth perturbation, thus realizing the compaction update of the skybox Gaussian along the spherical tangential.
[0076] The scale of the skybox Gaussian is usually large. If the angle between its principal axis and the tangent plane of the sphere is too large during optimization, the new Gaussian generated during splitting may deviate significantly from the sphere. This not only introduces unnecessary depth perturbations but also fails to make a substantial contribution to texture optimization.
[0077] To address the aforementioned problems, this invention proposes a skybox densification method with spherical alignment, constraining the generation process of the new Gaussian surface to the tangential direction of the sphere (i.e., the densification direction). The specific effects are as follows: Figure 2 As shown. Figure 2 In the diagram, the ellipsoid without a dashed edge represents the original Gaussian ellipsoid, while the Gaussian ellipsoid with a dashed border represents the new Gaussian ellipsoid generated after splitting. Arrows pointing towards the center of the scene indicate the direction of the ellipsoid's center, and arrows collinear with the longest axis of the original Gaussian ellipsoid represent the direction of the principal axis of that Gaussian ellipsoid. This method mainly comprises three core stages: compaction direction calculation, sampling variance calculation, and new coordinate synthesis.
[0078] Compacting direction calculation: The compaction direction must satisfy two core constraints: First, the direction vector must be in the tangent plane of the sphere to minimize depth perturbation; Second, the direction vector must be consistent with the principal axis of the Gaussian ellipsoid, that is, coincide with the projection direction of the principal axis on the tangent plane, to ensure that the new Gaussian generated by splitting can retain the distribution characteristics of the original Gaussian to the greatest extent.
[0079] In practice, the compaction direction is calculated through two cross product operations to minimize the error. First, the unit vector pointing from the Gaussian center of the original skybox to the center of the sphere needs to be obtained. As shown below:
[0080]
[0081] in, Represents the coordinates of the center point of the Gaussian skybox. Represents a very small value (usually 10). -6 (to ensure numerical stability) , These represent the components in three directions, respectively. Then, in order to obtain the sum... And the tangent of the sphere to the original Gaussian unit principal axis in the same plane, requires first constructing a plane perpendicular to... auxiliary vector The specific calculation formula is as follows:
[0082]
[0083] in, The original principal axis direction of the Gaussian unit. Represents a perpendicular to The unit vector is applicable when two vectors are almost collinear. The specific calculation formula is as follows:
[0084]
[0085] Calculate the splitting direction of the new Gaussian That is, with and The specific formula for calculating the tangent of coplanar spheres is as follows:
[0086]
[0087] Then the previously constructed auxiliary vectors Considering the rotation axis vector as a reference, construct a rotation coordinate system and calculate the principal axis directions. and The angle between The specific formula is as follows:
[0088]
[0089] Let the rotation quaternion of the principal axis of the original Gaussian ellipsoid relative to the coordinate axes of the world coordinate system be... Based on the Rodriguez rotation formula and the properties of quaternions, the rotation increment is calculated. :
[0090]
[0091] Once the incremental quaternion is obtained, the target quaternion can be calculated. The specific calculation process is as follows, based on its properties:
[0092]
[0093]
[0094]
[0095]
[0096]
[0097] Then, based on the conversion formula between quaternions and rotation matrices, the final tangential rotation matrix is obtained. :
[0098]
[0099] Sampling variance calculation: To ensure that the newly generated Gaussian is highly similar to the original Gaussian, the sampling standard deviation calculated along the extension direction using this method is:
[0100]
[0101] in, This represents the maximum axis length of the Gaussian ellipsoid. Subsequently, initial three-dimensional offsets are constructed in the principal axis coordinate system. , To conform to a mean of 0 and a standard deviation of The normal distribution is used, where the components perpendicular to the principal axis are all set to 0 to ensure that the sampling process is strictly carried out along the principal axis of the Gaussian unit.
[0102] New coordinate synthesis: Obtain the center coordinates of the new Gaussian after splitting: In particular, offset according to The calculation yielded, where Let the coordinates be the center coordinates of the Gaussian to be split. The coordinates are the center coordinates of the new Gaussian after the split.
[0103] Example 2: The computer-readable storage medium of this example stores a computer program that, when executed by a processor, implements the steps in the prior-based unbounded scene Gaussian sputtering decoupling optimization method of Example 1.
[0104] The computer-readable storage medium in this embodiment can be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; furthermore, the computer-readable storage medium can include both the terminal's internal storage unit and external storage devices.
[0105] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.
[0106] Example 3: The computer device of this example includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the prior-based unbounded scene Gaussian sputtering decoupling optimization method of Example 1.
[0107] In this embodiment, the processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The memory can include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.
[0108] Those skilled in the art will clearly understand that each implementation can be achieved using software plus the necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A priori-based unbounded scene Gaussian sputtering decoupling optimization method, characterized in that, Includes the following steps: Step 1: Obtain the initial point cloud and use a pre-trained monocular depth model to extract the binary sky mask of the training image; perform quantitative random sampling on the sky region. By combining the camera's intrinsic and extrinsic parameters, the projection ray of the sampled pixel is calculated; the intersection point of the projection ray and the skybox sphere is solved as the initial 3D center position of the skybox Gaussian, and the color of the sampled pixel is inherited as the initial color of the skybox Gaussian. Step 2: Introduce spherical tangential constraints, calculate the compaction direction of the Gaussian skybox, and calculate the sampling variance by combining the Gaussian principal axis features to generate a new Gaussian element without depth perturbation.
2. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 1, characterized in that, The specific process of random sampling is as follows: calculate the pixel proportion of the sky region in each training image, set the hyperparameter to control the total number of sky boxes and Gaussians, allocate the sampling number to each image according to the proportion, and randomly sample the sky region to obtain the sampled pixel set.
3. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 1, characterized in that, The equation for the projected ray is: ； in, The location of the camera center in the world coordinate system. The direction of the projected ray in the world coordinate system. These are the ray parameters.
4. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 3, characterized in that, The specific process for calculating the initial 3D center position of the skybox Gaussian is as follows: To find the intersection point of the ray and the sphere, the equation of the sphere is expressed as: ； in The coordinates of a point in the world coordinate system. Let the center of the sphere be in the world coordinate system. Let be the radius of the sphere; substituting the equation of the projected ray into the equation of the sphere: ； Derive about The standard quadratic equation: ； in, , , After solving the equation, select the unique positive root. Substituting into the equation of the projected ray, we obtain the center coordinates of the Gaussian skybox in the world coordinate system.
5. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 1, characterized in that, The specific process for calculating the compaction direction of the Gaussian skybox is as follows: Obtain the unit vector from the Gaussian center of the skybox to the center of the sphere. The calculation formula is as follows: ； in, Let the coordinates be the center point of the skybox Gaussian. For a very small value, , These are components in three directions; construct a perpendicular to... auxiliary vector The calculation formula is as follows: ； in, The original principal axis direction of the Gaussian unit. For a perpendicular to The unit vector is calculated using the following formula: ； Calculate the splitting direction of the new Gaussian The calculation formula is as follows: ； auxiliary vector Considering the rotation axis vector as a reference, construct a rotation coordinate system and calculate the principal axis directions. and The angle between The calculation formula is as follows: ； Let the rotation quaternion of the principal axis of the original Gaussian ellipsoid relative to the coordinate axes of the world coordinate system be... Calculate the rotation increment : ； After obtaining the incremental quaternion, the target quaternion is calculated. The specific calculation process is as follows: ； ； ； ； ； Calculate based on the conversion formula between quaternions and rotation matrices: ； This is the final tangential rotation matrix.
6. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 5, characterized in that, The sampling variance is obtained by calculating the sampling standard deviation along the extension direction: ； in, This represents the maximum axial length of the Gaussian ellipsoid.
7. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 6, characterized in that, The center coordinates of the new Gorsky element are: ； in, The coordinates of the center of the new Gaussian after the split are: This is the offset. , The initial 3D offset in the principal axis coordinate system. , To conform to a mean of 0 and a standard deviation of The normal distribution is denoted by , where the components perpendicular to the principal axis are all set to 0.
8. The prior-based unbounded scene Gaussian sputtering decoupling optimization method according to claim 1, characterized in that, The initial point cloud is obtained using the structure-of-motion (SfM) method.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the prior-based unbounded scene Gaussian sputtering decoupling optimization method as described in any one of claims 1-8.
10. A computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the prior-based unbounded scene Gaussian sputtering decoupling optimization method as described in any one of claims 1-8.