Three-dimensional model editing method, device, storage medium, and program product

Abstract

The application discloses a three-dimensional model editing method and device, a storage medium and a program product, relates to the technical field of image processing, and comprises the following steps: performing preliminary non-rigid editing on sparse anchor points, so as to obtain non-rigid edited anchor points, at this time, the non-rigid edited anchor points represent the characteristics of a target state; then, the non-rigid edited anchor points are used to initialize random Gaussian points; after densification, more Gaussian points can be obtained, so that the appearance of a target object can be more accurately represented; target dense Gaussian points that can accurately describe the appearance characteristics of the target object are obtained; and then, the dense three-dimensional model generated by the target dense Gaussian points is more accurate. At this time, the dense three-dimensional model is a target three-dimensional model. Since the non-rigid edited anchor points represent the change from a source state to a target state, the target object in the target three-dimensional model changes from the source state to the target state, and the dynamic change is represented.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to three-dimensional model editing methods, devices, storage media, and program products. Background Technology

[0002] Today, the construction of 3D models is widely used in film production and games, and building high-quality 3D models has become a research goal for researchers.

[0003] In related technologies, text-driven editing of objects in a source 3D model yields an edited image, thus creating the final 3D model. However, text-driven editing of objects in a source 3D model can only replace their colors, failing to meet the demands of complex and dynamic applications, and resulting in low quality and accuracy of the 3D model. Summary of the Invention

[0004] This application provides a method, device, storage medium, and program product for editing three-dimensional models, in order to at least solve the problems in related technologies that fail to meet the needs of dynamic application scenarios and that the quality and accuracy of three-dimensional models are low.

[0005] Firstly, this application provides a method for editing a three-dimensional model, including:

[0006] Obtain sparse anchor points from the source 3D model; the source 3D model includes the target object; the target object is in the source state; the sparse anchor points are anchor points that describe the geometric features of the target object;

[0007] Preliminary non-rigid editing is performed based on sparse anchor points to obtain non-rigid edited anchor points; the non-rigid edited anchor points are the anchor points from the source state to the target state of the target object after non-rigid editing.

[0008] Random Gaussian points are initialized based on the anchor points after non-rigid editing to obtain initial Gaussian points; the number of random Gaussian points is greater than the number of sparse anchor points; the random Gaussian points are a portion of the source Gaussian points in the source 3D model;

[0009] The initial Gaussian points are densed to determine the target densed Gaussian points that describe the appearance features of the target object; the densed 3D model generated based on the target densed Gaussian points is the target 3D model; after the target object in the target 3D model performs the target action, it forms the target state from the source state.

[0010] Secondly, this application also provides an editing device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described three-dimensional model editing methods.

[0011] Thirdly, this application also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described three-dimensional model editing methods.

[0012] Fourthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described three-dimensional model editing methods.

[0013] This application provides a 3D model editing method. First, sparse anchor points are obtained from a source 3D model, which includes the target object. These sparse anchor points describe the geometric features of the target object. Then, preliminary non-rigid editing is performed on the sparse anchor points to obtain non-rigidly edited anchor points. It is evident that the non-rigidly edited anchor points at this stage represent the characteristics of the target state. Next, random Gaussian points are initialized based on the non-rigidly edited anchor points to obtain initial Gaussian points. Then, the initial Gaussian points are densed. Since denserization can obtain more Gaussian points, it can more accurately represent the appearance of the target object. When a certain threshold is reached... After the process, target-dense Gaussian points are obtained. These target-dense Gaussian points can accurately describe the appearance features of the target object. Therefore, the densed 3D model generated from these target-dense Gaussian points is more accurate. This densed 3D model is the target 3D model. Since the target-dense Gaussian points are obtained by densening the initial Gaussian points, and the initial Gaussian points are based on anchor points obtained after non-rigid editing, the target object in the target 3D model changes from its source state to its target state, thus reflecting the dynamic changes in the 3D model. This allows the solution to realize changes in the state of the target object and can be used in dynamic copying application scenarios. It should be noted that the number of random Gaussian points is greater than the number of sparse anchor points, thus ensuring the number of initial Gaussian points and effectively controlling the number of Gaussian points after densening, improving the accuracy of the target 3D model. Furthermore, in this application, sparse anchor points can describe the geometric features of the target object, while target dense Gaussian points can describe the appearance features of the target object. In this application, the two features are combined to achieve decoupling of geometry and appearance, accurately representing the geometry and appearance of the target object in a sparse geometry-dense appearance manner. The use of sparse anchor points facilitates geometric transformations after preliminary non-rigid editing, while the density of target dense Gaussian points improves quality. In this application, the anchor points are densed based on the non-rigidly edited anchor points to obtain target dense Gaussian points that represent the appearance features of the target object, which can accurately capture the geometric and appearance features of the target object, making the target 3D model more accurate. Attached Figure Description

[0014] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A schematic diagram of the 3D model editing scene provided for this application;

[0016] Figure 2 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 1;

[0017] Figure 3 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 2;

[0018] Figure 4 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 4;

[0019] Figure 5 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 5;

[0020] Figure 6 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment Six;

[0021] Figure 7 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment Seven;

[0022] Figure 8 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 10;

[0023] Figure 9 This is a schematic diagram of a three-dimensional model editing method provided in Example Twelve;

[0024] Figure 10 This is a schematic diagram of a three-dimensional model editing device provided in Embodiment Thirteen;

[0025] Figure 11 This is a schematic diagram of the editing device provided in Embodiment Thirteen. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0027] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0028] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] The specific application environment architecture or specific hardware architecture on which the execution of the 3D model editing method depends is described here.

[0030] Today, the construction of 3D models is widely used in film production and games, and building high-quality 3D models has become a research goal for researchers.

[0031] In related technologies, text-driven editing of objects in a source 3D model is used to obtain a target 3D model. However, when editing objects in a source 3D model using text-driven methods, only color replacement can be achieved, which cannot meet the needs of complex and dynamic application scenarios.

[0032] In order to overcome the shortcomings of related technologies, the inventors of this solution have conducted creative research and designed a new solution. This solution provides a 3D model editing method. To address the inability of related technologies to meet dynamic change requirements, this solution performs preliminary non-rigid editing on sparse anchor points to obtain non-rigidly edited anchor points. These non-rigidly edited anchor points represent the anchor points through which the target object transforms from its source state to its target state. Further initialization and densification based on these non-rigidly edited anchor points yield denser Gaussian points. The denser 3D model generated from these denser Gaussian points is the target 3D model. The target object in the target 3D model transforms from its source state to its target state after performing a target action. Therefore, the target object undergoes a state change in the target 3D model, thus achieving non-rigid editing and meeting dynamic change requirements. Furthermore, to address the issues of poor 3D model quality and low accuracy in related technologies, this solution obtains sparse anchor points from the source 3D model. These sparse anchor points describe the geometric features of the target object. Then, the non-rigidly edited anchor points obtained from the preliminary non-rigid editing of the sparse anchor points are initialized to obtain initial Gaussian points. These initial Gaussian points are then densified to obtain more... The Gaussian points are densified, resulting in more densified Gaussian points until the target dense Gaussian points are obtained. Target dense Gaussian points increase density. Since these target dense Gaussian points can describe the appearance features of the target object, the resulting 3D model is more accurate in appearance. Furthermore, because the target dense Gaussian points are obtained based on sparse anchor points, the resulting 3D model is geometrically more accurate. Therefore, this scheme achieves a 3D model decoupled from geometry and appearance, specifically employing a sparse geometry-dense... The appearance-based approach decomposes the target object into sparse anchor points and dense Gaussian points obtained through densification. The sparse anchor points enable non-rigid editing, while the dense Gaussian points improve rendering quality due to their density. Simultaneously, this solution uses sparse anchor points as basic control points and employs densification to optimize geometry and appearance, thereby achieving non-rigid editing. In summary, this solution can accurately capture the geometry and appearance of the target object during the process of transforming the source 3D model from its source state to the target state, thus improving the accuracy and quality of the target 3D model.

[0033] Figure 1 A schematic diagram of the 3D model editing scene provided in this application, such as Figure 1 As shown, the specific application scenario of this application includes editing device 101.

[0034] The editing device 101 can be a computer or a server; there are no restrictions here.

[0035] The editing device 101 includes a three-dimensional model editing device.

[0036] In this application scenario, the editing device 101 obtains sparse anchor points and random Gaussian points from the source 3D model, and then performs preliminary non-rigid editing on the sparse anchor points to obtain the non-rigid edited anchor points.

[0037] Furthermore, the editing device 101 initializes the random Gaussian points based on the anchor points after non-rigid editing to obtain the initial Gaussian points.

[0038] Furthermore, the editing device 101 densifies the initial Gaussian points to determine the target denser Gaussian points. The denser 3D model generated from the target denser Gaussian points is the target 3D model.

[0039] In this application, the source 3D model is the source state. Preliminary non-rigid editing is performed on the sparse anchor points to execute the target action on the target object, transforming it from the source state to the target state. This yields the non-rigidly edited anchor points. The densed 3D model generated based on the dense Gaussian points is the target 3D model. Since the target object in the target 3D model transforms from the source state to the target state, dynamic changes are achieved. Furthermore, a sparse geometry-dense appearance approach is used to construct the target 3D model. During the construction process, the geometric and appearance features of the target object are fully considered, resulting in a more accurate and higher-quality target 3D model that not only changes the state of the target object but also improves its geometry and appearance.

[0040] In this application, the target 3D model is generated using target dense Gaussian points and optimized anchor points. The optimized anchor points are obtained by optimizing their attributes while keeping the anchor point positions unchanged after non-rigid editing. Thus, the optimized anchor points can describe the geometric features of the target object in the target state, while the dense target dense Gaussian points can describe the appearance features of the target object. By combining the geometric features and appearance features, the target 3D model is generated, making the combined features and appearance features of the target 3D model more accurate and of higher quality.

[0041] The embodiments of this application provide a three-dimensional model editing method, and the method is described in detail in conjunction with the execution flow of the three-dimensional model editing method.

[0042] Example 1

[0043] The execution subject of Embodiments 1 to 12 of this application is a three-dimensional model editing device (hereinafter referred to as the editing device), which is located in the editing equipment.

[0044] Figure 2 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 1. Figure 2 As shown, it includes:

[0045] S201, Obtain sparse anchor points from the source 3D model; the source 3D model includes the target object; the target object is in the source state; the sparse anchor points are anchor points that describe the geometric features of the target object.

[0046] The source 3D model refers to the 3D model of the target object to be edited.

[0047] In this context, the source state refers to the pose of the target object within the source 3D model or source image. For example, if the source 3D model includes a standing cat, then the source state is "standing".

[0048] In one approach, sparse anchor points can be obtained from the source 3D model using a farthest-distance sampling mechanism.

[0049] S202, perform preliminary non-rigid editing based on sparse anchor points to obtain non-rigid edited anchor points; the non-rigid edited anchor points are the anchor points from the source state to the target state of the target object after non-rigid editing.

[0050] The preliminary non-rigid editing involves performing coarse non-rigid editing on sparse anchor points, and a non-rigid transformation field is constructed after the preliminary non-rigid editing.

[0051] In one approach, preliminary non-rigid editing is performed based on sparse anchor points and a multilayer perception model, ultimately obtaining the non-rigidly edited anchor points.

[0052] S203, initialize random Gaussian points based on the anchor points after non-rigid editing to obtain initial Gaussian points; the random Gaussian points are a portion of the source Gaussian points in the source 3D model.

[0053] Among them, the number of random Gaussian points is greater than the number of sparse anchor points. Since the number of random Gaussian points is greater than the number of sparse anchor points, the number of initial Gaussian points has a certain advantage, which effectively controls the number of target dense Gaussian points and improves the density.

[0054] The source 3D model includes multiple source Gaussian points. For example, assuming there are 1000 source Gaussian points, a random mechanism is used to determine multiple source Gaussian points as random Gaussian points from these 1000 source Gaussian points.

[0055] It should be noted that the random Gaussian points used in this application are Gaussian points in the source 3D model, not completely random Gaussian points. This can improve the controllability of the initialization of random Gaussian points and ensure the stability of the initial Gaussian points.

[0056] S204, the initial Gaussian points are densed to determine the target densed Gaussian points that describe the appearance features of the target object; the densed 3D model generated based on the target densed Gaussian points is the target 3D model, and the target object in the target 3D model forms the target state from the source state after performing the target action.

[0057] In one approach, the initial Gaussian points can be densed by using a method of copying Gaussian points and an iterative densening method to gradually denser them, thereby obtaining the target densed Gaussian points.

[0058] It should be noted that the densed 3D model generated based on the target dense Gaussian is the target 3D model, and the target object in the target 3D model is the target state.

[0059] This embodiment provides a 3D model editing method. First, sparse anchor points are obtained from a source 3D model, which includes a target object. These sparse anchor points describe the geometric features of the target object. Then, preliminary non-rigid editing is performed on the sparse anchor points to obtain non-rigidly edited anchor points. As can be seen, the non-rigidly edited anchor points at this stage characterize the features of the target state. Next, random Gaussian points are initialized based on the non-rigidly edited anchor points to obtain initial Gaussian points. Then, the initial Gaussian points are densed. Since denserization can obtain more Gaussian points, it can more accurately represent the appearance of the target object. When a certain threshold is reached... After a certain degree of refinement, target-dense Gaussian points are obtained. These target-dense Gaussian points can accurately describe the appearance features of the target object. Consequently, the densed 3D model generated from these target-dense Gaussian points is more accurate. This densed 3D model is the target 3D model. In the target 3D model, since the target-dense Gaussian points are obtained by densifying the initial Gaussian points, and the initial Gaussian points are obtained based on anchor points after non-rigid editing, the target object in the target 3D model changes from the source state to the target state, thus reflecting the dynamic changes of the 3D model. This allows this solution to realize changes in the state of the target object and can be applied in complex and dynamic application scenarios. It should be noted that the number of random Gaussian points is greater than the number of sparse anchor points, thereby ensuring the number of initial Gaussian points and effectively controlling the number of Gaussian points after densification, improving the accuracy of the target 3D model. Furthermore, in this embodiment, sparse anchor points can describe the geometric features of the target object, while target dense Gaussian points can describe the appearance features of the target object. In this embodiment, the two features are combined to achieve decoupling of geometry and appearance, accurately representing the geometry and appearance of the target object in a sparse geometry-dense appearance manner. The use of sparse anchor points facilitates geometric transformations after preliminary non-rigid editing, while the density of target dense Gaussian points improves quality. In this embodiment, the anchor points are densed based on the non-rigidly edited anchor points to obtain target dense Gaussian points that represent the appearance features of the target object, which can accurately capture the geometric and appearance features of the target object, making the target 3D model more accurate.

[0060] Example 2

[0061] This embodiment is a further refinement of Embodiment 1, and it is an optional method for obtaining sparse anchor points from the source 3D model.

[0062] Figure 3 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 2. Figure 3 As shown, it specifically includes:

[0063] S301, determine the multiple source Gaussian points included in the source 3D model.

[0064] Specifically, a 3D Gaussian scan is used to scan the source 3D model to obtain multiple source Gaussian points.

[0065] S302, Calculate the first distance; the first distance is the distance between each source Gaussian point calculated based on the three-dimensional coordinates of the source Gaussian point.

[0066] Specifically, calculate the distance between each source Gaussian point.

[0067] Specifically, calculate the distance between any two source Gaussian points.

[0068] When calculating the first distance, the three-dimensional coordinates of each source Gaussian point are used.

[0069] S303, the source Gaussian point corresponding to the first distance that satisfies the preset farthest distance sampling mechanism is determined as a sparse anchor point; the number of sparse anchor points is less than the number of source Gaussian points; the number of sparse anchor points is greater than or equal to the first preset number.

[0070] Specifically, based on the preset farthest distance sampling mechanism, the first distance that satisfies the preset farthest distance sampling mechanism is determined from multiple calculated first distances. Then, the source Gaussian point corresponding to the first distance is determined, and then the source Gaussian point is determined as the sparse anchor point.

[0071] For example, suppose there are five source Gaussian points, namely source Gaussian point 1, source Gaussian point 2, source Gaussian point 3, source Gaussian point 4 and source Gaussian point 5. Then, calculate the first distance, specifically: calculate the distance between source Gaussian point 1 and source Gaussian point 2, denoted as L12. Similarly, calculate the other first distances according to the above method, namely L13, L14, L15, L23, L24, L25, L34, L35 and L45. From the above multiple first distances, determine the first distance that satisfies the preset farthest distance sampling mechanism, assuming it is L13 and L14. Then, determine the source Gaussian points corresponding to L13 and L14 as sparse anchor points.

[0072] It should be noted that the number of sparse anchor points also needs to meet certain requirements to ensure the accuracy of the initial non-rigid editing of the sparse anchor points. In this embodiment, the number of sparse anchor points is greater than or equal to a first preset number, where the first preset number is a pre-defined number of sparse anchor points.

[0073] It should be noted that in this embodiment, the obtained sparse anchor points are used as basic control points. Since sparse anchor points are anchor points that describe the geometric features of the target object, the geometric features of the target object can be guaranteed in subsequent processes based on these basic control points, thus improving accuracy.

[0074] This embodiment provides a three-dimensional model editing method. In this embodiment, a mechanism using a first distance and a preset maximum distance can accurately determine sparse anchor points. Furthermore, this embodiment ensures that the number of sparse anchor points is greater than or equal to a first preset number, thereby improving the accuracy of the target three-dimensional model in subsequent target three-dimensional model generation.

[0075] Example 3

[0076] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method for obtaining non-rigidly edited anchor points by performing preliminary non-rigid editing based on sparse anchor points, specifically including:

[0077] A degree-of-freedom transformation algorithm is iteratively optimized to perform preliminary non-rigid editing on sparse anchor points, obtaining the non-rigidly edited anchor points. The degree-of-freedom transformation algorithm is used to obtain the target rotation matrix and target translation vector, and the non-rigidly edited anchor points are obtained based on the target rotation matrix and target translation vector. The degree-of-freedom transformation algorithm is iteratively optimized in the preliminary non-rigid editing to obtain the target rotation matrix and target translation vector.

[0078] Specifically, the preliminary non-rigid editing of sparse anchor points is performed using the degree-of-freedom transformation algorithm in the multilayer perceptron model. The parameters included in the degree-of-freedom transformation algorithm are iteratively optimized so that the editing loss value of the edited 3D model generated based on the rotation matrix and translation vector output by the optimized degree-of-freedom transformation algorithm meets the preset editing loss conditions. The iteratively optimized anchor points corresponding to the edited 3D model that meet the preset editing loss conditions are determined as the final non-rigid edited anchor points. At the same time, the rotation matrix and translation vector that meet the preset editing loss conditions are determined as the target rotation matrix and target translation vector. The edited 3D model is generated based on the non-rigid edited anchor points, which are generated based on the non-rigid editing anchor point algorithm.

[0079] The algorithm for transforming degrees of freedom is shown in equation (1):

[0080] (1)

[0081] in, Represents the rotation matrix. ; Represents the translation vector. , This indicates sparse anchor points, i.e., anchor points before initial non-rigid editing. Represents the algorithm for transforming degrees of freedom. This represents the parameters in the degree-of-freedom transformation algorithm.

[0082] The non-rigid editing anchor point algorithm is shown in equation (2):

[0083] (2)

[0084] in, This represents the anchor point for iterative optimization.

[0085] For example, during the initial non-rigid editing of sparse anchor points in the first iteration, the parameters in the degree-of-freedom transformation algorithm are as follows: The initial value (denoted as) ),Will Enter to In the middle, we obtained ,Will Inputting into the non-rigid editing anchor point algorithm, we get At this time, based on If the editing loss value of the edited 3D model generated based on the source 3D model does not meet the preset editing loss condition, iterative optimization is performed based on sparse anchor points. For sparse anchor points. In the second iteration, the parameters are... From initial value Optimized to Then The input is fed into the degree-of-freedom transformation algorithm for the second iteration optimization, resulting in... At this time, Input into the non-rigid editing anchor point algorithm, and obtain ,based on The edit loss value of the generated edited 3D model does not meet the preset edit loss condition. Therefore, further adjustments are made to the parameters. Iterative optimization, in the third iteration, adjust the parameters from Optimized to ,Will The input is fed into the third iteration of the optimization algorithm for the transformation of degrees of freedom, resulting in... At this time Input into the non-rigid editing anchor point algorithm, and obtain Assuming based on The iteration ends when the edit loss value of the generated edited 3D model meets the preset edit loss condition, at which point the target rotation matrix is ​​obtained (i.e., ) and target translation vector (i.e. At this point, the anchor point after non-rigid editing is... Therefore, it can be seen that It is obtained based on the target rotation matrix and the target translation vector; it should be noted that, in the third iteration, based on The edit loss value of the generated edited 3D model does not meet the preset edit loss condition, so the parameters are iteratively optimized. The iteration continues until the editing loss value of the edited 3D model generated based on the iteratively optimized anchor points meets the preset editing loss condition. The iteration ends then, and the iteratively optimized anchor points that meet the preset editing loss condition are determined as the non-rigid edited anchor points.

[0086] Among them, the preset editing loss condition is the loss condition set in advance when performing preliminary non-rigid editing on sparse anchor points. The preset editing loss condition is the loss condition considered under standard resolution.

[0087] Among them, based on The generated edited 3D model can be based on The edited 3D model is generated based on the edited 3D model of the previous iteration, where the previous iteration is the first iteration.

[0088] This embodiment provides a three-dimensional model editing method. In this embodiment, a degree-of-freedom transformation algorithm is used to obtain the target rotation matrix and the target translation vector. This allows for preliminary non-rigid editing based on the target rotation matrix and the target translation vector, thereby obtaining anchor points after non-rigid editing. In this embodiment, the degree-of-freedom transformation algorithm can be iteratively optimized to obtain a better target rotation matrix and target translation vector, thus obtaining a better anchor point after non-rigid editing and improving accuracy.

[0089] Example 4

[0090] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method to initialize random Gaussian points based on anchor points after non-rigid editing in order to obtain initial Gaussian points.

[0091] Figure 4 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 4. Figure 4 As shown, it includes:

[0092] S401: For any random Gaussian point, obtain the random three-dimensional coordinates of the random Gaussian point.

[0093] The location of the random Gaussian point is represented by random three-dimensional coordinates.

[0094] S402, determine the K-value anchor points that satisfy the preset K-value nearest neighbor method from multiple non-rigidly edited anchor points; the K-value anchor points are a second preset number; the second preset number is less than or equal to the first preset number.

[0095] Specifically, the second preset number of non-rigidly edited anchor points with the closest K value can be determined as K-value anchor points.

[0096] For example, assuming the first preset number of sparse anchor points is 10 and the second preset number is 4, the four non-rigidly edited anchor points with the closest K value are determined as K-value anchor points.

[0097] S403, calculate the second distance; the second distance is the distance between the random three-dimensional coordinates and the three-dimensional coordinates of each K-value anchor point.

[0098] S404, the second distance is determined as the weight corresponding to the anchor point after non-rigid editing.

[0099] S405: Input the weights and K-value anchor points into the preset initialization algorithm to output the initial Gaussian points after random Gaussian point initialization.

[0100] The initial three-dimensional Gaussian point can be used as follows: It means that, among them, ,in , , Indicates the position of the initial center point. This represents the covariance matrix of the initial three-dimensional Gaussian points. The color of the initial three-dimensional Gaussian point. This represents the transparency of the initial three-dimensional Gaussian points. This represents the number of initial three-dimensional Gaussian points.

[0101] The preset initialization algorithm is shown in equation (3): (3)

[0103]

[0104]

[0105]

[0106]

[0107] in, as well as The anchor points after non-rigid editing are shown in the following order. The center point location, covariance matrix, color, and transparency are included. Let be the weight corresponding to the i-th initial 3D Gaussian point. Let K be the second preset quantity. If the second preset quantity is 4, then K=4.

[0108] Furthermore, as well as The input is fed into a preset initialization algorithm, which then yields the initial values ​​for each initial Gaussian point. .

[0109] This embodiment provides a three-dimensional model editing method. In this embodiment, for any random Gaussian point, a second distance is calculated based on its random three-dimensional coordinates and the three-dimensional coordinates of the K-value anchor point. The second distance is then determined as the weight corresponding to the non-rigid editing. Subsequently, a preset initialization algorithm is used to calculate the initial Gaussian point after the random Gaussian point is initialized, and the initial Gaussian point is assigned a value. In this embodiment, the initialization of each random Gaussian point can be realized. In addition, in this embodiment, K-value anchor points that satisfy the preset K-value nearest neighbor method are used for initialization, thus ensuring the accuracy and stability of the initial Gaussian point.

[0110] Example 5

[0111] This embodiment is a further refinement of any of the above embodiments. This embodiment is a further refinement of the initial Gaussian points to determine the target dense Gaussian points that describe the appearance features of the target object.

[0112] Figure 5 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 5. Figure 5 As shown, it specifically includes:

[0113] S501, the initial Gaussian points are densed in the first iteration to generate the densed 3D model of the first iteration.

[0114] It should be noted that the denser 3D model of the first iteration can be generated based on the Gaussian points and optimized anchor points of the first iteration, on top of the initial edited 3D model. The initial edited 3D model is identical to the source 3D model. Alternatively, the denser 3D model of the first iteration can be generated based on the edited 3D model that meets the preset editing loss conditions, on top of the Gaussian points and optimized anchor points of the first iteration.

[0115] S502 determines the density loss result of the 3D model after the first iteration of densification.

[0116] The density loss result refers to whether the density loss value of the densed 3D model meets the preset density loss conditions. The density loss result includes whether the preset density loss conditions are met or not.

[0117] S503, in response to the dense loss result satisfying the preset dense loss condition, the initial Gaussian point is determined as the target dense Gaussian point describing the appearance features of the target object.

[0118] Among them, the preset densening loss condition is a pre-set loss condition for densening.

[0119] S504, in response to the dense loss result not meeting the preset dense loss condition, perform at least one iteration of densening based on the Gaussian points after the first iteration of densening, until the dense loss result of the densed 3D model after iteration optimization meets the preset dense loss condition, and determine the densed Gaussian points corresponding to the preset dense loss condition as the target densed Gaussian points describing the appearance features of the target object.

[0120] It should be noted that if the density loss result does not meet the preset density loss condition, the densitying process continues iteratively based on the Gaussian points generated in the first iteration. Each iteration of densitying continues based on the Gaussian points generated in the previous iteration, generating a denser 3D model based on the densed Gaussian points. The current denser 3D model is generated based on the denser Gaussian points of the current iteration, and is based on the denser 3D model of the previous iteration. This process continues until the density loss result of the iteratively optimized denser 3D model meets the preset density loss condition, at which point the iteration ends. Specifically, the denser Gaussian points of the current iteration are generated based on the denser Gaussian points generated in the previous iteration.

[0121] Furthermore, at the end of the iteration, the densed Gaussian point that satisfies the preset densed loss condition is determined as the target densed Gaussian point describing the appearance features of the target object.

[0122] For example, suppose two more iterations of densification are needed. In the second iteration, densification is performed based on the Gaussian points obtained from the first iteration to obtain the Gaussian points obtained from the second iteration. Based on the Gaussian points obtained from the second iteration, a densified 3D model is generated. The density loss result of the 3D model obtained from the second iteration is determined. If the density loss result meets the preset density loss condition, the Gaussian points obtained from the second iteration are determined as the target density Gaussian points describing the appearance features of the target object. If the density loss result does not meet the preset density loss condition, iterative densification continues based on the Gaussian points obtained from the second iteration until the density loss result of the 3D model obtained from the iteration meets the preset density loss condition, and the iteration ends.

[0123] This embodiment provides a 3D model editing method. Firstly, the initial Gaussian points are densed in the first iteration, generating a densed 3D model for the first iteration. Then, the corresponding density loss result is determined. Further, if the density loss result satisfies a preset density condition, the iteration ends, and the initial Gaussian point is determined as the target densed Gaussian point. If the density loss result does not satisfy the preset density condition, the densed model continues to iterate until a densed 3D model is determined whose density loss result satisfies the preset density loss condition. The densed Gaussian point corresponding to the satisfied density loss condition is then determined as the target densed Gaussian point. In this embodiment, the target densed Gaussian point can describe the appearance features of the target object, thus making the appearance of the densed 3D model more accurate.

[0124] This embodiment also includes:

[0125] The densed 3D model that meets the preset densening loss condition is determined as the target 3D model; the target 3D model is generated from the target densed Gaussian points and the optimized anchor points; the optimized anchor points are generated by optimizing the properties of the non-rigidly edited anchor points while keeping their positions unchanged.

[0126] It should be noted that any densed 3D model is generated from densed Gaussian points and optimized anchor points, while the target 3D model is generated from target densed Gaussian points and optimized anchor points.

[0127] The optimized anchor points are generated based on the non-rigidly edited anchor points. It should be noted that the optimized anchor points can also be optimized in each iteration to generate the optimized anchor points corresponding to the current iteration number. Therefore, the target 3D model can be generated from the target dense Gaussian points and the optimized anchor points corresponding to the last iteration number.

[0128] Specifically, in response to the dense loss result satisfying the preset dense loss condition, the densed 3D model of the first iteration is determined as the target 3D model.

[0129] This embodiment provides a 3D model editing method. In this embodiment, the densed 3D model that meets the preset densed loss condition is determined as the target 3D model. In this embodiment, the target 3D model is generated from target densed Gaussian points and optimized anchor points. The optimized anchor points are generated by optimizing the attributes of the non-rigidly edited anchor points while keeping their positions unchanged. In this embodiment, target densed Gaussian points can describe the appearance features of the target object. The optimized anchor points are optimized from the non-rigidly edited anchor points, which are generated from sparse anchor points. Since sparse anchor points can describe the geometric features of the target object, the optimized anchor points can also describe the geometric features of the target object. Moreover, the attributes of the optimized anchor points are more consistent and accurate, thus making the target 3D model more accurate in geometry and appearance.

[0130] Example 6

[0131] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method for performing the first iteration of densification on the initial Gaussian points to generate the densified 3D model after the first iteration.

[0132] Figure 6 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment Six. Figure 6 As shown, it includes:

[0133] S601 determines the position gradient of multiple initial Gaussian points.

[0134] S602, in response to the position gradient of any initial Gaussian point being greater than or equal to the first preset gradient, determine the initial Gaussian point as the Gaussian point to be densed.

[0135] Wherein, the first preset gradient can be .

[0136] Specifically, the position gradient of each initial Gaussian point is determined, and the magnitude of each position gradient is compared with the first preset gradient. If the position gradient of any initial Gaussian point is greater than or equal to the first preset gradient, then the initial Gaussian point is determined as the Gaussian point to be densed.

[0137] For example, assuming there are 100 initial Gaussian points, determine that the position gradient of 50 of these initial Gaussian points is greater than or equal to a first preset gradient, and then determine these 50 initial Gaussian points as Gaussian points to be densed.

[0138] S603, based on the reconstruction type of the region where the Gaussian point to be densed is located, performs the first iteration of densening of the Gaussian point to be densed to generate the first iteration of densed 3D model.

[0139] Furthermore, based on the reconstruction type of the region where the Gaussian points to be densed are located, the first iteration of densening is carried out in a targeted manner, thereby generating the first iteration of densed 3D model.

[0140] This embodiment provides a three-dimensional model editing method. In this embodiment, a condition is first defined, that is, only the initial Gaussian points whose position gradient is greater than or equal to the first preset gradient are densed. The initial Gaussian point is determined as the Gaussian point to be densed. Then, according to the reconstruction type of the region where the Gaussian point to be densed is located, the first iteration of densed model is performed in a targeted manner, thereby generating the densed three-dimensional model after the first iteration.

[0141] Example 7

[0142] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method for performing the first iteration of densification of the Gaussian points to be densified based on the reconstruction type of the region where the Gaussian points to be densified are located, so as to generate the three-dimensional model after the first iteration of densification.

[0143] Figure 7 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment Seven. Figure 7 As shown, it includes:

[0144] S701, in response to the fact that the reconstruction type of the region where the Gaussian point to be densed is located is under-reconstruction type, the Gaussian point to be densed is subjected to under-reconstruction iterative densening to determine the under-reconstruction dense Gaussian point corresponding to the Gaussian point to be densed.

[0145] Based on the above example, assuming that 20 of the 50 Gaussian points to be densed are located in regions with under-reconstruction types, then under-reconstruction iterative densening is performed on these 20 Gaussian points to be densed, thereby determining the under-reconstruction dense Gaussian points corresponding to these 20 Gaussian points to be densed.

[0146] S702, in response to the fact that the reconstruction type of the region where the Gaussian point to be densed is the transitional reconstruction type, the transitional reconstruction is iteratively densed to determine the transitional reconstruction dense Gaussian point corresponding to the Gaussian point to be densed.

[0147] Based on the above example, for the remaining 30 Gaussian points to be densed, the reconstruction type of the region where they are located is the transitional reconstruction type. Then, the transitional reconstruction is iteratively densed for these 30 Gaussian points to be densed to determine the transitional reconstruction dense Gaussian points corresponding to these 30 Gaussian points to be densed.

[0148] S703 generates the first iteration of the densed 3D model based on each under-reconstructed dense Gaussian point, each transitionally reconstructed dense Gaussian point, the initial Gaussian point excluding the Gaussian point to be densed, and the optimized anchor point; each under-reconstructed dense Gaussian point, each transitionally reconstructed dense Gaussian point, and the initial Gaussian point excluding the Gaussian point to be densed are the densed Gaussian points of the first iteration.

[0149] Among them, the initial Gaussian points other than the Gaussian points to be densed refer to the remaining initial Gaussian points excluding the Gaussian points to be densed.

[0150] According to the above example, in the above steps, 20 under-reconstructed dense Gaussian points corresponding to the Gaussian points to be densed are obtained through S701, and 30 transitionally reconstructed dense Gaussian points corresponding to the Gaussian points to be densed are obtained through S702. The initial Gaussian points other than the Gaussian points to be densed are calculated to be 50. Further, the under-reconstructed dense Gaussian points corresponding to the 20 Gaussian points to be densed, the transitionally reconstructed dense Gaussian points corresponding to the 30 Gaussian points to be densed, the 50 initial Gaussian points other than the Gaussian points to be densed, and the optimized anchor points are used to generate the densed 3D model of the first iteration.

[0151] It should be noted that, in this example, the 20 Gaussian points to be densed and the corresponding under-reconstructed dense Gaussian points, the 30 Gaussian points to be densed and the corresponding transitionally reconstructed dense Gaussian points, and the 50 initial Gaussian points other than the Gaussian points to be densed and the corresponding initial Gaussian points are the densed Gaussian points of the first iteration.

[0152] In one approach, the densed 3D model of the first iteration is generated by covering the region where the corresponding Gaussian point to be densed is located on the source 3D model with each under-reconstructed dense Gaussian point, each interim reconstructed dense Gaussian point, and the initial Gaussian point (excluding the Gaussian point to be densed). Specifically, assuming that a Gaussian point to be densed is densed through under-reconstruction iterations to generate a corresponding under-reconstructed dense Gaussian point, this under-reconstructed dense Gaussian point covers the region where the aforementioned Gaussian point to be densed is located, thus accurately obtaining the densed 3D model.

[0153] This embodiment provides a 3D model editing method. In this embodiment, iterative densification is performed in a targeted manner according to the reconstruction type of the region where the Gaussian point to be densified is located, which is either under-reconstruction or transitional reconstruction. This determines the under-reconstruction dense Gaussian points and the transitional reconstruction dense Gaussian points. Then, based on each under-reconstruction dense Gaussian point, each transitional reconstruction dense Gaussian point, the initial Gaussian points other than the Gaussian point to be densified, and the optimized anchor points, the first iteration of the densified 3D model is generated. That is, the first iteration of the densified 3D model is generated based on the first iteration of the densified Gaussian points and the optimized anchor points. Because iterative densification is performed in a targeted manner according to the reconstruction type, the target 3D model can be made more accurate.

[0154] Example 8

[0155] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method for performing under-reconstruction iterative densification on the Gaussian points to be densed in order to determine the under-reconstruction dense Gaussian points corresponding to the Gaussian points to be densed. Specifically, it includes:

[0156] Step 1: During the under-reconstruction iterative densification, the Gaussian point to be densified is determined as the first Gaussian point, and the Gaussian point to be densified is copied to obtain the corresponding copied Gaussian point; the scale of the first Gaussian point is smaller than the first preset scale.

[0157] The scale of the first Gaussian point is smaller than the first preset scale, which means that the first Gaussian point is a small Gaussian.

[0158] Furthermore, by copying the first Gaussian point, a copied Gaussian point can be obtained.

[0159] Step 2: Identify the copied Gaussian points as the under-reconstructed dense Gaussian points corresponding to the Gaussian points to be densed.

[0160] Based on the above example, 20 Gaussian points to be densed are identified as small Gaussian points. For any Gaussian point to be densed, the Gaussian point to be densed is copied to obtain the copied Gaussian point corresponding to the Gaussian point to be densed. The copied Gaussian point is identified as the under-reconstructed dense Gaussian point corresponding to the Gaussian point to be densed. Thus, the under-reconstructed dense Gaussian points corresponding to each of the 20 Gaussian points to be densed can be obtained.

[0161] This embodiment provides a three-dimensional model editing method. In this embodiment, for the first Gaussian point, since the scale of the first Gaussian point is smaller than the first preset scale, it is determined that the first Gaussian point is actually a small Gaussian point. Therefore, in order to perform under-reconstruction iterative densification, the first Gaussian point is copied to obtain the copied Gaussian point. The copied Gaussian point is the under-reconstruction dense Gaussian point corresponding to the Gaussian point to be densed.

[0162] Example 9

[0163] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method for performing transitional reconstruction iterative densification on the Gaussian points to be densed in order to determine the transitional reconstructed dense Gaussian points corresponding to the Gaussian points to be densed. Specifically, it includes:

[0164] Step 1: During the transition reconstruction iteration densification, the Gaussian point to be densified is determined as the second Gaussian point. The Gaussian point to be densified is split to obtain the split Gaussian point. The scale of the split Gaussian point satisfies the preset split scale condition. The second Gaussian point is greater than or equal to the second preset scale. The first preset scale is less than the second preset scale.

[0165] The first preset scale is smaller than the second preset scale. Since the second Gaussian point is greater than or equal to the second preset scale, the second Gaussian point can be understood as a large Gaussian point, and therefore it is necessary to split the second Gaussian point.

[0166] Step 2: Determine the Gaussian points after splitting as the transitional reconstructed dense Gaussian points corresponding to the Gaussian points to be densed.

[0167] Based on the above example, 30 Gaussian points to be densed are determined as second Gaussian points. For any second Gaussian point, it is split to obtain split Gaussian points. Then, the split Gaussian points are determined as the transition reconstruction dense Gaussian points corresponding to the Gaussian points to be densed.

[0168] This embodiment provides a three-dimensional model editing method. In this embodiment, for the second Gaussian point, since the scale of the second Gaussian point is greater than or equal to the second preset scale, it is determined that the second Gaussian point is actually a large Gaussian point. Therefore, in order to perform transition reconstruction iteration densification, the second Gaussian point is split, and then the split Gaussian point is obtained. The split Gaussian point is the transition reconstruction dense Gaussian point corresponding to the Gaussian point to be densed.

[0169] Example 10

[0170] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional way to determine the density loss result of the three-dimensional model after the first iteration of densification.

[0171] Figure 8 This is a schematic diagram of a three-dimensional model editing method provided in Embodiment 10. Figure 8 As shown, it includes:

[0172] S801, render the densed 3D model from the first iteration from different perspectives to obtain the first densed rendering image set, which includes the first densed rendering images from different perspectives.

[0173] S802, the first denser rendering image set and the target multi-view image set are input into the preset denser reconstruction loss function to output the denser loss value corresponding to the first iteration; the target multi-view image set is the image set that forms the target state from the source state after the target object performs the target action in the view of the corresponding source image; the target multi-view image set is the output of multiple source images from different viewpoints and target text prompt information from the multi-view image editing model that has been trained to convergence.

[0174] The pre-defined dense reconstruction loss function is shown in equation (3):

[0175] (3)

[0176] in, It refers to the target multi-view image from the i-th viewpoint, and the target multi-view image comes from the target multi-view image set; This refers to the first densely rendered image from the i-th viewpoint, which originates from the first densely rendered image set; where M represents the number of viewpoints; where, This refers to the densification loss value.

[0177] S803, in response to the denser loss value satisfying the preset denser loss condition, determine the denser loss result as satisfying the preset denser loss condition.

[0178] Specifically, if the output is If the preset densening loss condition is met, then the densening loss result is determined to meet the preset densening loss condition.

[0179] S804, in response to the denser loss value not meeting the preset denser loss condition, the denser loss result is determined to be that the preset denser loss condition is not met.

[0180] If the output is If the preset densening loss condition is not met, the densening loss result is determined to be "not satisfied with the preset densening loss condition".

[0181] This embodiment provides a three-dimensional model editing method. In this embodiment, the densification loss value is calculated based on the target multi-view image set. Since the target multi-view image set is the output of multiple source images from different perspectives and target text prompt information from the multi-view image editing model that has been trained to convergence, the target state in the target multi-view image set is more accurate. Therefore, the densification loss result can be accurately determined based on the target multi-view image set.

[0182] Example 11

[0183] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method of performing at least one iteration of denserization on the Gaussian points after the first iteration, until the denser loss result of the optimized 3D model satisfies a preset denser loss condition. The method of determining the denser Gaussian points corresponding to the preset denser loss condition as the target denser Gaussian points describing the appearance features of the target object includes:

[0184] Step 1: When performing at least one iteration of densification based on the Gaussian points after the first iteration, in response to the current iteration number being greater than the first preset iteration number, determine the Gaussian point to be reconstructed from the Gaussian points after densification corresponding to the current iteration number; the position gradient of the Gaussian point to be reconstructed is greater than the second preset gradient.

[0185] The second preset gradient is greater than the first preset gradient.

[0186] The first preset number of iterations can be any number, for example, 300.

[0187] For example, if the current iteration number is 301, which is greater than the first preset iteration number, then the Gaussian point to be reconstructed is determined from the densed Gaussian points corresponding to the current iteration number. If the current iteration number corresponds to 1000 densed Gaussian points, then the Gaussian point to be reconstructed is determined from the 1000 densed Gaussian points.

[0188] Step 2: Continue to iterate and denser the Gaussian points to be reconstructed to optimize the densed 3D model until the densed loss value of the optimized densed 3D model meets the preset densed loss condition at a certain number of iterations. The densed Gaussian points that meet the preset densed loss condition are determined as the target densed Gaussian points.

[0189] Furthermore, the reconstructed Gaussian points are further iteratively densed using the same method as the first iterative denser method described above, which will not be repeated here.

[0190] This embodiment provides a three-dimensional model editing method. In this embodiment, when the number of iterations reaches a certain level, that is, when the current number of iterations is greater than the number of the first preset zone, the Gaussian points to be reconstructed that are greater than the second preset gradient are determined from the Gaussian points after the current number of iterations are densed. The Gaussian points to be reconstructed are then iteratively densed. Therefore, the total number of Gaussian points can be controlled in this embodiment.

[0191] Example 12

[0192] This embodiment is a further refinement of any of the above embodiments. This embodiment is an optional method to continue iteratively denserizing the Gaussian points to be reconstructed in order to optimize the denser 3D model until the denser loss value of the optimized denser 3D model meets the preset denser loss condition at a certain number of iterations. The denser Gaussian points that meet the preset denser loss condition are determined as the target denser Gaussian points.

[0193] Figure 9 This is a schematic diagram of a three-dimensional model editing method provided in Example Twelve. Figure 9 As shown, it includes:

[0194] S901, when continuing to iterate and densify, in response to the current iteration number being greater than the second preset iteration number, determine the Gaussian point to be deleted from the densed Gaussian points corresponding to the current iteration number; the scale of the Gaussian point to be deleted is greater than the third preset scale; the second preset iteration number is greater than the first preset iteration number.

[0195] Among them, the Gaussian points to be deleted refer to the Gaussian points with a scale greater than the third preset scale in the denser Gaussian points corresponding to the current iteration number when the current iteration number is greater than the second preset iteration number.

[0196] In this embodiment, when continuing to iterate and densify, if the current iteration number is 1001, where the second preset iteration number is 1000, then the current iteration number is greater than the second preset iteration number, and the Gaussian points to be deleted are determined from the densed Gaussian points corresponding to the current iteration number.

[0197] S902, delete the Gaussian point to be deleted from the densed Gaussian points of the current iteration number to obtain the densed Gaussian points after deletion corresponding to the current iteration number.

[0198] S903, continue to iterate and densify the deleted dense Gaussian points to optimize the dense 3D model until the densification loss value of the optimized dense 3D model meets the preset densification loss condition at a certain number of iterations. The dense Gaussian points that meet the preset densification loss condition are determined as the target dense Gaussian points.

[0199] This embodiment provides a 3D model editing method. In this embodiment, when the current iteration number is greater than a second preset iteration number, Gaussian points with a scale larger than the second preset scale are determined from the densed Gaussian points corresponding to the current iteration number and deleted. Then, the deleted densed Gaussian points are iteratively densed to optimize the densed 3D model. This process continues until, at a certain iteration number, the densed loss value of the optimized densed 3D model meets the preset densed loss condition. The corresponding densed Gaussian point at this time is determined as the target densed Gaussian point. In this embodiment, deleting the Gaussian points to be deleted when the iteration number reaches a higher level can effectively ensure the total number of Gaussian points, increase the number and density of target densed Gaussian points, and ensure accuracy.

[0200] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0201] Example 13

[0202] Embodiments of this application also provide a three-dimensional model editing device. Figure 10 This is a schematic diagram of a three-dimensional model editing device provided in Embodiment Thirteen. The three-dimensional model editing device 1000 includes the following modules:

[0203] Module 1001 is used to obtain sparse anchor points from the source 3D model; the source 3D model includes the target object; the target object is in the source state; the sparse anchor points are anchor points that describe the geometric features of the target object;

[0204] The editing module 1002 is used to perform preliminary non-rigid editing based on sparse anchor points to obtain non-rigid edited anchor points; the non-rigid edited anchor points are the anchor points from the source state to the target state of the target object after non-rigid editing.

[0205] Initialization module 1003 is used to initialize random Gaussian points based on the anchor points after non-rigid editing, so as to obtain initial Gaussian points; the random Gaussian points are the source Gaussian points in the source 3D model;

[0206] The densification module 1004 is used to densify the initial Gaussian points to determine the target dense Gaussian points that describe the appearance features of the target object; the densed 3D model generated based on the target dense Gaussian points is the target 3D model, and the target object in the target 3D model forms the target state from the source state after performing the target action.

[0207] Optionally, module 1001 is used specifically to obtain sparse anchor points from the source 3D model when:

[0208] Identify the multiple source Gaussian points included in the source 3D model;

[0209] Calculate the first distance; the first distance is the distance between each source Gaussian point calculated based on the three-dimensional coordinates of the source Gaussian points;

[0210] The source Gaussian point corresponding to the first distance that satisfies the preset maximum distance sampling mechanism is determined as a sparse anchor point; the number of sparse anchor points is less than the number of source Gaussian points; the number of sparse anchor points is greater than or equal to the first preset number.

[0211] Optionally, the editing module 1002, when performing preliminary non-rigid editing based on sparse anchor points to obtain non-rigid edited anchor points, is specifically used for:

[0212] A degree-of-freedom transformation algorithm is used for iterative optimization to perform preliminary non-rigid editing on sparse anchor points, obtaining non-rigidly edited anchor points. The degree-of-freedom transformation algorithm is used to obtain the target rotation matrix and target translation vector. The non-rigidly edited anchor points are optional and are obtained based on the target rotation matrix and target translation vector. Initialization module 1003 is used to initialize random Gaussian points based on the non-rigidly edited anchor points to obtain initial Gaussian points. Specifically, it is used for:

[0213] For any random Gaussian point, obtain the random three-dimensional coordinates of the random Gaussian point;

[0214] From multiple non-rigidly edited anchor points, determine the K-value anchor points that satisfy the preset K-value nearest neighbor method; the K-value anchor points are a second preset number; the second preset number is less than or equal to the first preset number;

[0215] Calculate the second distance; the second distance is the distance between the random 3D coordinates and the 3D coordinates of each K-value anchor point;

[0216] The second distance is determined as the weight corresponding to the anchor point after non-rigid editing;

[0217] Input the weights and K-value anchor points into the preset initialization algorithm to output the initial Gaussian points after random Gaussian point initialization.

[0218] Optionally, the densification module 1004, when densifying the initial Gaussian points to determine the target densed Gaussian points describing the appearance features of the target object, is specifically used for:

[0219] The initial Gaussian points are densed in the first iteration to generate the densed 3D model of the first iteration.

[0220] Determine the density loss result of the 3D model after the first iteration of densification;

[0221] In response to the dense loss result satisfying the preset dense loss condition, the initial Gaussian point is determined as the target dense Gaussian point describing the appearance features of the target object;

[0222] In response to the dense loss result not meeting the preset dense loss condition, at least one iteration of densening is performed based on the Gaussian points after the first iteration of densening until the dense loss result of the densed 3D model after iteration optimization meets the preset dense loss condition. The densed Gaussian points that meet the preset dense loss condition are determined as the target densed Gaussian points describing the appearance features of the target object.

[0223] Optionally, this embodiment provides a three-dimensional model editing device, which further includes: a determining module;

[0224] The determination module is used to determine the densed 3D model that meets the preset densed loss conditions as the target 3D model; the target 3D model is generated from the target densed Gaussian points and the optimized anchor points; the optimized anchor points are generated by optimizing the properties of the non-rigidly edited anchor points while keeping their positions unchanged.

[0225] Optionally, the densification module 1004, when performing the first iteration of densification on the initial Gaussian points to generate the first iteration of the densified 3D model, is specifically used for:

[0226] Determine the position gradients of multiple initial Gaussian points;

[0227] If the position gradient of any initial Gaussian point is greater than or equal to the first preset gradient, the initial Gaussian point is determined as the Gaussian point to be densed.

[0228] Based on the reconstruction type of the region where the Gaussian point to be densed is located, the Gaussian point to be densed is densed in the first iteration to generate the densed 3D model of the first iteration.

[0229] Optionally, the densification module 1004, when performing the first iteration of densification on the Gaussian points to be densified based on the reconstruction type of the region where the Gaussian points to be densified are located, to generate the first iteration of the densified 3D model, is specifically used for:

[0230] In response to the fact that the reconstruction type of the region where the Gaussian point to be densed is located is under-reconstruction type, the Gaussian point to be densed is subjected to under-reconstruction iterative densening to determine the under-reconstruction dense Gaussian point corresponding to the Gaussian point to be densed.

[0231] In response to the fact that the reconstruction type of the region where the Gaussian point to be densed is located is the transitional reconstruction type, the transitional reconstruction is iteratively densed to determine the transitional reconstruction dense Gaussian point corresponding to the Gaussian point to be densed.

[0232] The first iteration of the dense Gaussian model is generated based on each under-reconstructed dense Gaussian point, each transitionally reconstructed dense Gaussian point, the initial Gaussian point excluding the Gaussian point to be densed, and the optimized anchor point; each under-reconstructed dense Gaussian point, each transitionally reconstructed dense Gaussian point, and the initial Gaussian point excluding the Gaussian point to be densed are the densed Gaussian points of the first iteration.

[0233] Optionally, the densening module 1004, when performing under-reconstruction iterative densening on the Gaussian points to be densed to determine the under-reconstruction dense Gaussian points corresponding to the Gaussian points to be densed, is specifically used for:

[0234] When performing under-reconstruction iterative densification, the Gaussian point to be densified is determined as the first Gaussian point, and the Gaussian point to be densified is copied to obtain the corresponding copied Gaussian point; the scale of the first Gaussian point is smaller than the first preset scale;

[0235] The copied Gaussian points are identified as the under-reconstructed dense Gaussian points corresponding to the Gaussian points to be densed.

[0236] Optionally, the densening module 1004, when performing iterative densening of the Gaussian points to be densed through transition reconstruction to determine the corresponding transitionally reconstructed dense Gaussian points, is specifically used for:

[0237] During the transition reconstruction iteration densification, the Gaussian point to be densified is determined as the second Gaussian point. The Gaussian point to be densified is split to obtain the split Gaussian point. The scale of the split Gaussian point satisfies the preset split scale condition. The second Gaussian point is greater than or equal to the second preset scale. The first preset scale is less than the second preset scale.

[0238] The split Gaussian points are identified as the transitional reconstructed dense Gaussian points corresponding to the Gaussian points to be densed.

[0239] Optionally, the densification module 1004, when determining the density loss result of the 3D model after the first iteration of densification, is specifically used for:

[0240] The first denser 3D model is rendered from different perspectives to obtain the first denser rendered image set, which includes the first denser rendered images from different perspectives.

[0241] The first denser rendering image set and the target multi-view image set are input into the preset denser reconstruction loss function to output the denser loss value corresponding to the first iteration; the target multi-view image set is the image set that forms the target state from the source state after the target object performs the target action in the view of the corresponding source image; the target multi-view image set is the output of multiple source images from different viewpoints and target text prompt information from the multi-view image editing model that has been trained to convergence;

[0242] In response to the denser loss value satisfying the preset denser loss condition, the denser loss result is determined to satisfy the preset denser loss condition;

[0243] In response to the denser loss value not meeting the preset denser loss condition, the denser loss result is determined to be that the preset denser loss condition is not met.

[0244] Optionally, the densening module 1004, after performing at least one iteration of densening based on the Gaussian points after the first iteration, until the dense loss result of the densed 3D model obtained through iteration optimization satisfies the preset densening loss condition, and when determining the densed Gaussian points corresponding to the preset densening loss condition as the target densed Gaussian points describing the appearance features of the target object, is specifically used for:

[0245] When performing at least one iteration of densification based on the Gaussian points after the first iteration, in response to the current iteration number being greater than the first preset iteration number, the Gaussian point to be reconstructed is determined from the Gaussian points after the current iteration number; the position gradient of the Gaussian point to be reconstructed is greater than the second preset gradient;

[0246] The reconstructed Gaussian points are continuously iterated and densed to optimize the densed 3D model until the densed loss value of the optimized densed 3D model meets the preset densed loss condition at a certain number of iterations. The densed Gaussian points that meet the preset densed loss condition are determined as the target densed Gaussian points.

[0247] Optionally, the densification module 1004, while iteratively densifying the Gaussian points to be reconstructed to optimize the densified 3D model until the densification loss value of the optimized densified 3D model meets a preset densification loss condition at a certain number of iterations, determines the densified Gaussian points corresponding to the preset densification loss condition as the target densified Gaussian points. Specifically, it is used for:

[0248] When continuing the iterative densification, in response to the current iteration number being greater than the second preset iteration number, Gaussian points to be deleted are determined from the densed Gaussian points corresponding to the current iteration number; the scale of the Gaussian points to be deleted is greater than the third preset scale; the second preset iteration number is greater than the first preset iteration number.

[0249] Remove the Gaussian points to be deleted from the densed Gaussian points of the current iteration number to obtain the densed Gaussian points after deletion for the current iteration number.

[0250] Continue to iterate and densify the deleted dense Gaussian points to optimize the dense 3D model until the densification loss value of the optimized dense 3D model meets the preset densification loss condition at a certain number of iterations. The dense Gaussian points that meet the preset densification loss condition are then determined as the target dense Gaussian points.

[0251] For a description of the features in the embodiment corresponding to the 3D model editing device, please refer to the relevant description in the embodiment corresponding to the 3D model editing method, which will not be repeated here.

[0252] Embodiments of this application also provide an editing device. Figure 11 This is a schematic diagram of the editing device provided in Embodiment Thirteen. Figure 11 As shown, the editing device 1100 includes a processor 1101 and a memory 1102, wherein the processor 1101, the memory 1102, and the communication component 1103 are connected via a bus 1104. The memory 1102 stores a computer program, and the processor 1101 is configured to run the computer program to perform the steps in any of the above-described embodiments of the 3D model editing method.

[0253] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above embodiments of the three-dimensional model editing method at runtime.

[0254] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0255] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described embodiments of the 3D model editing method.

[0256] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described embodiments of the three-dimensional model editing method.

[0257] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0258] The foregoing has provided a detailed description of a three-dimensional model editing method provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for editing three-dimensional models, characterized in that, include: The source 3D model is determined to include multiple source Gaussian points, wherein the source 3D model includes a target object in the source state; Based on the three-dimensional coordinates of each source Gaussian point, calculate multiple first distances between each source Gaussian point; The source Gaussian point corresponding to the first distance that satisfies the preset farthest distance sampling mechanism is determined as the sparse anchor point; The sparse anchor points are used to describe the geometric features of the target object, and the number of sparse anchor points is less than the number of source Gaussian points and greater than or equal to a first preset number. The sparse anchor points are preliminarily non-rigidly edited by a degree-of-freedom transformation algorithm to obtain the non-rigidly edited anchor points. The degree-of-freedom transformation algorithm is used to obtain the target rotation matrix and target translation vector that satisfy the preset editing loss conditions. The anchor points after non-rigid editing are obtained based on the target rotation matrix and target translation vector and are used to characterize the change of the target object from the source state to the target state. For any random Gaussian point, based on the random 3D coordinates of the random Gaussian point in the source 3D model, select K-value anchor points from a plurality of non-rigidly edited anchor points that satisfy the preset K-value nearest neighbor method; the K-value anchor points are a second preset number, and the second preset number is less than or equal to the first preset number; The second distance between each K-value anchor point and the random Gaussian point is determined as the weight of the corresponding non-rigidly edited anchor point; The weights and K-value anchor points are input into a preset initialization algorithm to obtain the initial Gaussian points after the random Gaussian points are initialized; The number and density of the initial Gaussian points are gradually increased through an iterative densification method to determine the target dense Gaussian points; the target dense Gaussian points are used to describe the appearance features of the target object and, together with the sparse anchor points, characterize the target object; The densed 3D model generated based on the target densed Gaussian points is the target 3D model. After the target object in the target 3D model performs the target action, it forms the target state from the source state.

2. The method according to claim 1, characterized in that, The step of gradually increasing the number and density of the initial Gaussian points through an iterative densification method to determine the target densed Gaussian points describing the appearance features of the target object includes: The initial Gaussian points are densed in the first iteration to generate the densed 3D model of the first iteration. Determine the density loss result of the densed 3D model after the first iteration; In response to the dense loss result satisfying the preset dense loss condition, the initial Gaussian point is determined as the target dense Gaussian point describing the appearance features of the target object; In response to the dense loss result not satisfying the preset dense loss condition, at least one iteration of densening is performed based on the Gaussian points after the first iteration of densening until the dense loss result of the densed 3D model after iteration optimization satisfies the preset dense loss condition. The densed Gaussian points corresponding to satisfying the preset dense loss condition are determined as the target densed Gaussian points describing the appearance features of the target object. The method further includes: The densed 3D model that meets the preset densening loss condition is determined as the target 3D model; the target 3D model is generated from the target densed Gaussian points and the optimized anchor points; the optimized anchor points are generated by optimizing the properties of the non-rigidly edited anchor points while keeping their positions unchanged.

3. The method according to claim 2, characterized in that, The first iteration of densification of the initial Gaussian points to generate the first iteration of the densified 3D model includes: Determine the position gradients of the multiple initial Gaussian points; In response to the position gradient of any initial Gaussian point being greater than or equal to a first preset gradient, the initial Gaussian point is determined to be a Gaussian point to be densed. The Gaussian points to be densed are densed in the first iteration based on the reconstruction type of the region where the Gaussian points to be densed are located, so as to generate the densed 3D model after the first iteration.

4. The method according to claim 3, characterized in that, The first iteration of densification of the Gaussian points to be densed, based on the reconstruction type of the region where the Gaussian points to be densed are located, to generate the first iteration of densed 3D model, includes: In response to the fact that the reconstruction type of the region where the Gaussian point to be densed is located is under-reconstruction type, the Gaussian point to be densed is subjected to under-reconstruction iterative densening to determine the under-reconstruction dense Gaussian point corresponding to the Gaussian point to be densed. In response to the fact that the reconstruction type of the region where the Gaussian point to be densed is located is a transitional reconstruction type, the transitional reconstruction is iteratively densed for the Gaussian point to be densed in order to determine the transitional reconstruction dense Gaussian point corresponding to the Gaussian point to be densed. The first iteration of the dense Gaussian model is generated based on the under-reconstructed dense Gaussian points, the transitional reconstructed dense Gaussian points, the initial Gaussian points excluding the Gaussian points to be densed, and the optimized anchor points; the under-reconstructed dense Gaussian points, the transitional reconstructed dense Gaussian points, and the initial Gaussian points excluding the Gaussian points to be densed are the densed Gaussian points of the first iteration.

5. The method according to claim 4, characterized in that, The step of performing under-reconstruction iterative densening on the Gaussian points to be densed, to determine the under-reconstruction dense Gaussian points corresponding to the Gaussian points to be densed, includes: During under-reconstruction iterative densification, the Gaussian point to be densified is determined as the first Gaussian point, and the Gaussian point to be densified is copied to obtain the corresponding copied Gaussian point; the scale of the first Gaussian point is smaller than the first preset scale; The copied Gaussian point is determined as the under-reconstructed dense Gaussian point corresponding to the Gaussian point to be densed.

6. The method according to claim 5, characterized in that, The step of performing transitional reconstruction iterative densification on the Gaussian points to be densed, to determine the transitional reconstructed dense Gaussian points corresponding to the Gaussian points to be densed, includes: During the transition reconstruction iteration densification, the Gaussian point to be densified is determined as the second Gaussian point, and the Gaussian point to be densified is split to obtain the split Gaussian point; the scale of the split Gaussian point satisfies the preset split scale condition; the second Gaussian point is greater than or equal to the second preset scale; the first preset scale is less than the second preset scale; The split Gaussian point is determined as the transitional reconstructed dense Gaussian point corresponding to the Gaussian point to be densed.

7. The method according to claim 2, characterized in that, Determining the density loss result of the densed 3D model after the first iteration includes: The densed 3D model of the first iteration is rendered from different perspectives to obtain a first densed rendering image set, which includes first densed rendering images from different perspectives. The first denser rendering image set and the target multi-view image set are input into a preset denser reconstruction loss function to output the denser loss value corresponding to the first iteration; the target multi-view image set is the image set that forms the target state from the source state after the target object performs the target action in the view of the corresponding source image; the target multi-view image set is the output of multiple source images from different viewpoints and target text prompt information from the multi-view image editing model that has been trained to convergence; In response to the denser loss value satisfying a preset denser loss condition, the denser loss result is determined to satisfy the preset denser loss condition; In response to the fact that the density loss value does not meet the preset density loss condition, the density loss result is determined to be that the preset density loss condition is not met.

8. The method according to claim 2, characterized in that, The process of performing at least one iteration of densening based on the Gaussian points after the first iteration until the densed loss result of the optimized 3D model satisfies a preset densening loss condition, and determining the densed Gaussian points corresponding to the preset densening loss condition as target densed Gaussian points describing the appearance features of the target object, includes: When performing at least one iteration of densification based on the Gaussian points after the first iteration of densification, in response to the current iteration number being greater than a first preset iteration number, a Gaussian point to be reconstructed is determined from the Gaussian points after densification corresponding to the current iteration number; the position gradient of the Gaussian point to be reconstructed is greater than a second preset gradient; The Gaussian points to be reconstructed are further iterated and densed to optimize the densed 3D model until the densed loss value of the optimized densed 3D model meets the preset densed loss condition at a certain number of iterations. The densed Gaussian points that meet the preset densed loss condition are then determined as the target densed Gaussian points.

9. The method according to claim 8, characterized in that, The process of iteratively denserizing the Gaussian points to be reconstructed to optimize the densed 3D model continues until the densed loss value of the optimized densed 3D model meets a preset densed loss condition at a certain number of iterations. The densed Gaussian points that meet the preset densed loss condition are then identified as the target densed Gaussian points. This includes: When continuing the iterative densification, in response to the current iteration number being greater than the second preset iteration number, Gaussian points to be deleted are determined from the densed Gaussian points corresponding to the current iteration number; the scale of the Gaussian points to be deleted is greater than the third preset scale; the second preset iteration number is greater than the first preset iteration number. The Gaussian point to be deleted is removed from the dense Gaussian points after the current iteration number to obtain the dense Gaussian points after deletion corresponding to the current iteration number. Continue to iterate and densify the deleted dense Gaussian points to optimize the dense 3D model until the densification loss value of the optimized dense 3D model meets the preset densification loss condition at a certain number of iterations. The dense Gaussian points that meet the preset densification loss condition are determined as the target dense Gaussian points.

10. An editing device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the three-dimensional model editing method as described in any one of claims 1 to 9 when executing the computer program.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the three-dimensional model editing method as described in any one of claims 1 to 9.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the three-dimensional model editing method as described in any one of claims 1 to 9.

Images