Dynamic scene reconstruction method and device, equipment, medium and product

Abstract

The invention discloses a dynamic scene reconstruction method and device, equipment, a medium and a product, and relates to the technical field of multimedia, and the method comprises the steps: responding to a received dynamic scene reconstruction instruction, and determining the rendering time and the rendering visual angle according to the dynamic scene reconstruction instruction; based on a pre-trained 4D Gaussian representation, performing rendering according to the rendering time and the rendering view angle to obtain a target rendering view; wherein the 4D Gaussian representation is obtained by performing adaptive optimization training on Gaussian primitives in a 4D space based on a scale perception residual field, and an area occupied by the Gaussian primitives is coded through the scale perception residual field, so that more accurate features are generated and aligned with self-splitting behaviors of the Gaussian primitives; through an adaptive optimization strategy, the capability of reconstructing high-frequency time information in the dynamic scene by the obtained 4D Gaussian representation is enhanced, the 4D Gaussian representation obtained through training can be rendered at any time and any view angle, and the rendering quality of the dynamic scene is improved.

Description

Technical Field

[0001] The present application relates to the field of multimedia technology, and in particular to a dynamic scene reconstruction method, device, equipment, medium and product. Background Art

[0002] Dynamic scene reconstruction is essential for immersive imaging and has driven the development of multiple multimedia technologies such as VR, AR, and the Metaverse. Rendering dynamic scenes from arbitrary time, position, and viewpoint is essential for improving user experience multimedia products such as free viewpoint video and bullet time effects.

[0003] At present, dynamic scene reconstruction methods based on Neural Radiance Fields (NeRF) and 3DGS have made significant progress. NeRF uses implicit fields to simulate static scenes and achieves realistic viewpoint synthesis. Many NeRF extension methods for dynamic scenes either use deformation fields and gauge fields to simulate the motion of objects relative to the gauge frame over time, or decompose the four-dimensional volume into only space and spatiotemporal space, and represent the space by combining reduced dimensionality features. Despite significant progress in rendering quality, these methods have obvious disadvantages in rendering speed. The emergence of 3DGS makes real-time rendering of dynamic scenes possible. Some methods attempt to model dynamic scenes based on 3DGS. However, they either have difficulty simulating temporally complex scenes, such as the appearance and disappearance of objects, or ignore the spatiotemporal information in the scene, and have disadvantages in processing temporally complex dynamic scenes.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a dynamic scene reconstruction method, device, equipment, medium and product, aiming to improve the rendering quality of dynamic scenes.

[0006] To achieve the above object, the present application provides a dynamic scene reconstruction method, the method comprising:

[0007] In response to receiving a dynamic scene reconstruction instruction, determining a rendering time and a rendering viewing angle according to the dynamic scene reconstruction instruction;

[0008] Based on the pre-trained 4D Gaussian representation, rendering is performed according to the rendering time and the rendering perspective to obtain a target rendering view;

[0009] The 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in the 4D space based on the scale-aware residual field.

[0010] In one embodiment, before the step of rendering based on the pre-trained 4D Gaussian representation according to the rendering time and the rendering perspective to obtain the target rendering view, the step further includes:

[0011] Acquire a calibrated multi-view video sequence, wherein the calibrated multi-view video sequence includes a sampling time and a corresponding real image;

[0012] Initializing a 4D Gaussian basis and a scale-aware residual field in a 4D space according to the calibrated multi-view video sequence;

[0013] Given the sampling time, determining a residual feature of the 4D Gaussian basis element in the scale-aware residual field;

[0014] Calculate the residual of the survival state and attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature;

[0015] combining the initial attributes of the 4D Gaussian primitives with the residuals of the attributes to obtain a 3D Gaussian representation in 3D space;

[0016] Projecting the 3D Gaussian representation to obtain a rendered image;

[0017] Gradient inversion is performed according to the loss between the rendered image and the real image, and adaptive optimization training is performed on the 4D Gaussian basis to obtain the 4D Gaussian representation.

[0018] In one embodiment, the step of calculating the residual of the survival state and attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature comprises:

[0019] Calculate the survival state of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature;

[0020] The residual feature of the 4D Gaussian primitive at the sampling time is decoded by a multi-layer perceptron, and combined with the survival state to obtain the residual of the attribute.

[0021] In an embodiment, the scale-aware residual field is represented by six planes, including a spatial-only plane and a spatiotemporal plane.

[0022] In one embodiment, the step of determining the residual features of the 4D Gaussian primitives in the scale-aware residual field comprises:

[0023] A multi-scale feature representation framework is established through pyramid texture mapping stack technology;

[0024] Stores and optimizes the lowest resolution level by dynamically calculating pyramid texture map levels;

[0025] Determining a scaling level of the 4D Gaussian primitive in the multi-scale feature representation framework according to a projection size and a basic spatial scale of the 4D Gaussian primitive;

[0026] Scale-aware feature extraction is performed according to the minimum resolution level and the zoom level to obtain the residual feature.

[0027] In one embodiment, the step of performing adaptive optimization training on the 4D Gaussian basis element to obtain the 4D Gaussian representation includes:

[0028] Obtaining the sampling probability of the 4D Gaussian basis element;

[0029] Calculate the definite integral of the 4D Gaussian basis element in the time domain according to the sampling probability to obtain a cumulative distribution function;

[0030] The densification threshold and learning rate of the 4D Gaussian basis element are dynamically adjusted according to the cumulative distribution function, and iterative training is performed until a preset number of iterations is reached to obtain the 4D Gaussian representation.

[0031] In addition, to achieve the above-mentioned purpose, the present application also proposes a dynamic scene reconstruction device, the dynamic scene reconstruction device comprising:

[0032] A response module, configured to, in response to receiving a dynamic scene reconstruction instruction, determine a rendering time and a rendering viewing angle according to the dynamic scene reconstruction instruction;

[0033] A rendering module, configured to perform rendering based on a pre-trained 4D Gaussian representation according to the rendering time and the rendering viewing angle to obtain a target rendering view;

[0034] The 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in the 4D space based on the scale-aware residual field.

[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes a dynamic scene reconstruction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the dynamic scene reconstruction method described above.

[0036] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the dynamic scene reconstruction method described above are implemented.

[0037] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the dynamic scene reconstruction method described above are implemented.

[0038] One or more technical solutions proposed in this application have at least the following technical effects:

[0039] In response to receiving a dynamic scene reconstruction instruction, a rendering time and a rendering perspective are determined according to the dynamic scene reconstruction instruction; based on a pre-trained 4D Gaussian representation, rendering is performed according to the rendering time and the rendering perspective to obtain a target rendering view; wherein the 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in a 4D space based on a scale-aware residual field, and the area occupied by the Gaussian primitives is encoded through the scale-aware residual field, thereby generating more accurate features and aligning with the self-splitting behavior of the Gaussian primitives; an adaptive optimization strategy is used to enhance the ability of the obtained 4D Gaussian representation to reconstruct high-frequency time information in dynamic scenes, and the 4D Gaussian representation obtained through training can be rendered at any time and any perspective, thereby achieving real-time rendering and ensuring high-quality reconstruction, thereby improving the rendering quality of dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0042] Figure 1 A schematic diagram of a flow chart provided for the first embodiment of the dynamic scene reconstruction method of the present application;

[0043] Figure 2 A schematic diagram of a flow chart provided for Embodiment 2 of the dynamic scene reconstruction method of the present application;

[0044] Figure 3 It is a schematic diagram of the overall process according to the second embodiment of the present application;

[0045] Figure 4 A flowchart of the third embodiment of the dynamic scene reconstruction method of the present application is provided;

[0046] Figure 5 A flowchart of the fourth embodiment of the dynamic scene reconstruction method of the present application is provided;

[0047] Figure 6 This is a schematic diagram for comparing the effects shown in the fourth embodiment of the present application;

[0048] Figure 7 This is a schematic diagram of the module structure of the dynamic scene reconstruction device according to an embodiment of the present application;

[0049] Figure 8 Schematic diagram of the device structure of the hardware operating environment involved in the dynamic scene reconstruction method in the embodiment of the present application.

[0050] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0051] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0052] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0053] The main solution of the embodiment of the present application is: in response to receiving a dynamic scene reconstruction instruction, determining the rendering time and rendering perspective according to the dynamic scene reconstruction instruction; based on a pre-trained 4D Gaussian representation, rendering is performed according to the rendering time and rendering perspective to obtain a target rendering view; wherein the 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in 4D space based on a scale-aware residual field, and the area occupied by the Gaussian primitives is encoded through the scale-aware residual field, thereby generating more accurate features and aligning with the self-splitting behavior of the Gaussian primitives; through an adaptive optimization strategy, the ability of the obtained 4D Gaussian representation to reconstruct high-frequency time information in dynamic scenes is enhanced, and the 4D Gaussian representation obtained through training can be rendered at any time and any perspective, thereby achieving real-time rendering and ensuring high-quality reconstruction, thereby improving the rendering quality of dynamic scenes.

[0054] In this embodiment, for the convenience of description, the following description is made by taking the dynamic scene reconstruction device as the execution subject.

[0055] In the prior art, dynamic scene reconstruction is crucial for immersive imaging, which has driven the development of various multimedia technologies such as VR, AR, and Metaverse. Rendering dynamic scenes from arbitrary time, position, and viewpoint is crucial to improving the user experience of multimedia products such as free viewpoint video and bullet time effects. Our goal is to reconstruct a continuous four-dimensional space from discrete temporal video sequences. However, this effort faces several challenges. First, reconstruction quality, as a bottleneck for widespread adoption, requires accurate capture of spatial dimensions and temporal changes in dynamic scenes. In addition, the growing demand for real-time interactivity in multimedia products to improve user engagement highlights the importance of achieving real-time rendering. However, existing methods have difficulty in achieving high-quality reconstruction and real-time rendering at the same time, which is exactly the problem our method aims to solve.

[0056] At present, dynamic scene reconstruction methods based on NeRF and 3DGS have made significant progress. NeRF uses implicit fields to simulate static scenes and achieve realistic viewpoint synthesis. Many NeRF extension methods, used for dynamic scenes, either use deformation fields and gauge fields to simulate the motion of objects relative to the gauge frame over time, or decompose the four-dimensional volume into only space and spatiotemporal space, and represent the space by combining reduced-dimensional features. Despite significant progress in rendering quality, these methods have obvious disadvantages in rendering speed. The emergence of 3DGS makes real-time rendering of dynamic scenes possible. Some methods attempt to model dynamic scenes based on 3DGS. However, some methods either have difficulty simulating temporally complex scenes, such as the appearance and disappearance of objects, or ignore the spatiotemporal information in the scene, and have disadvantages in processing temporally complex dynamic scenes.

[0057] The present application provides a solution, specifically a dynamic scene reconstruction method based on Gaussian projection, which can reconstruct high-quality dynamic scenes with complex temporal information from monocular or multi-camera video sequences.

[0058] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a dynamic scene reconstruction device, etc. The following takes the dynamic scene reconstruction device as an example to illustrate this embodiment and the following embodiments.

[0059] Based on this, the present application embodiment provides a dynamic scene reconstruction method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the dynamic scene reconstruction method of the present application.

[0060] In this embodiment, the dynamic scene reconstruction method includes steps S10 to S20:

[0061] Step S10, in response to receiving a dynamic scene reconstruction instruction, determining a rendering time and a rendering viewing angle according to the dynamic scene reconstruction instruction;

[0062] Step S20, based on the pre-trained 4D Gaussian representation, rendering is performed according to the rendering time and rendering perspective to obtain a target rendering view, wherein the 4D Gaussian representation is obtained by adaptively optimizing the Gaussian primitives in the 4D space based on the scale-aware residual field.

[0063] Specifically, in an embodiment of the present application, a 4D Gaussian representation is obtained by training a calibrated multi-view video sequence, which can be used to perform view rendering at any time or any viewpoint corresponding to the video sequence.

[0064] Optionally, in an embodiment of the present application, a set of four-dimensional Gaussian primitives and a scale-aware residual field M are simultaneously optimized. When combined with M, each Gaussian primitive generates a residual feature and a lifetime o'. They all represent the temporal characteristics of the Gaussian primitives. Given a sampling time t0, in an embodiment of the present application, the survival state g(t0) of the Gaussian can be calculated, and the Gaussian residual features at time t0 can be decoded using MLP to obtain residual attributes. In addition, in an embodiment of the present application, these residuals are combined with the initial attributes of the Gaussian in four-dimensional space to obtain a three-dimensional Gaussian representation. After obtaining the three-dimensional Gaussian representation, the Gaussian projection method can be used to generate a rendered image.

[0065] Optionally, a scale-aware residual field is proposed in an embodiment of the present application for encoding the area occupied by the Gaussian primitives, so as to obtain more accurate features and align with the self-splitting behavior of the Gaussian primitives. Specifically, the 4D space is decomposed into three planes Pso containing only spatial dimensions and three spatiotemporal planes Pst. Since the size of the Gaussian primitives only affects its projection within the spatial plane Pso, the embodiment of the present application specifically considers using scale-aware encoding only within these planes. For each plane P containing only spatial dimensions, a MipMap stack can be used to represent features of different spatial scales in the scene. The 0th layer of the MipMap stack is a feature map of shape M×N×N, and its spatial scale is the smallest among all layers. The remaining layers are obtained by calculating thumbnails based on the features of the previous layer, where the width and height are reduced by half respectively. After the MipMap stack is established, appropriate features can be obtained for a Gaussian primitive according to its geometric size.

[0066] Optionally, an adaptive optimization strategy is introduced in the present application to enhance the model's ability to reconstruct high-frequency time information in dynamic scenes. Specifically, the strategy dynamically adjusts the learning rate and gradient density threshold according to the sampling probability of the Gaussian primitive within the observable time range. Specifically, in the embodiments of the present application, its survival function can be used to calculate the time integral within the observable range to represent its sampling probability. The larger the integral, the more intersections the life cycle of the Gaussian primitive has with the observable range, and thus it is more likely to be sampled. Further, in the embodiments of the present application, different learning rates and thresholds for densification are assigned to it according to its sampling probability.

[0067] This embodiment adopts the above scheme, specifically in response to receiving a dynamic scene reconstruction instruction, determines the rendering time and rendering perspective according to the dynamic scene reconstruction instruction; based on the pre-trained 4D Gaussian representation, renders according to the rendering time and rendering perspective to obtain a target rendering view; wherein the 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in 4D space based on a scale-aware residual field, and the area occupied by the Gaussian primitives is encoded by the scale-aware residual field, so as to generate more accurate features and align with the self-splitting behavior of the Gaussian primitives; through an adaptive optimization strategy, the ability of the obtained 4D Gaussian representation to reconstruct high-frequency time information in dynamic scenes is enhanced, and the 4D Gaussian representation obtained through training can be rendered at any time and any perspective, so as to achieve real-time rendering and ensure high-quality reconstruction, thereby improving the rendering quality of dynamic scenes.

[0068] Based on the first embodiment of the present application, the second embodiment of the present application is proposed. In the second embodiment of the present application, the same or similar contents as those of the above-mentioned embodiment 1 can be referred to the above introduction and will not be repeated later. Figure 2 Before step S20, the dynamic scene reconstruction method further includes steps S01 to S07:

[0069] Step S01, obtaining a calibrated multi-view video sequence, wherein the calibrated multi-view video sequence includes a sampling time and a corresponding real image;

[0070] Step S02, initializing a 4D Gaussian basis element and a scale-aware residual field in a 4D space according to the calibrated multi-view video sequence;

[0071] Step S03, given the sampling time, determining the residual feature of the 4D Gaussian primitive in the scale-aware residual field;

[0072] Step S04, calculating the residual of the survival state and attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature;

[0073] Step S05, combining the initial attributes of the 4D Gaussian primitives with the residuals of the attributes to obtain a 3D Gaussian representation in a 3D space;

[0074] Step S06, projecting the 3D Gaussian representation to obtain a rendered image;

[0075] Step S07, performing gradient inversion according to the loss between the rendered image and the real image, performing adaptive optimization training on the 4D Gaussian basis to obtain the 4D Gaussian representation.

[0076] Reference Figure 3 , Figure 3 FIG. 1 is a schematic diagram of an overall process according to the second embodiment of the present application. Figure 3 As shown, where σ: Lσ represents the life cycle of 4D Gaussian, τ: represents the position of 4D Gaussian in the time domain, P SO : represents the space plane P St represents the space-time plane, Δμ, Δ∑, Δc: represents the residual of Gaussian attribute.

[0077] Optionally, first determine the set of Gaussian primitives to be optimized in four-dimensional space (three-dimensional space plus time dimension), and the scale-aware residual field M associated with these Gaussian primitives, and then calculate and define the residual characteristics and life span o' for each Gaussian primitive, which will describe the dynamic change characteristics of the Gaussian primitive over time. For a specific sampling time point t0, calculate the survival probability t(t0) of each Gaussian primitive, which reflects the activity level of the Gaussian primitive at that time point.

[0078] Optionally, the scale-aware residual field is represented by six planes, including a spatial-only plane and a spatiotemporal plane.

[0079] Optionally, the step of calculating the residual of the survival state and attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature includes:

[0080] Calculate the survival state of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature;

[0081] The residual feature of the 4D Gaussian primitive at the sampling time is decoded by a multi-layer perceptron, and combined with the survival state to obtain the residual of the attribute.

[0082] Optionally, in the embodiment of the present application, a multi-layer perceptron (MLP) is used to decode the residual features of each Gaussian primitive at time point t0 to predict its feature state at that time point. The decoded residual attributes are combined with the initial three-dimensional attributes of the Gaussian primitive to form a complete three-dimensional Gaussian representation, which will be used for subsequent image rendering. The obtained three-dimensional Gaussian representation is used to generate the final rendered image by techniques such as the Gaussian projection method.

[0083] This embodiment adopts the above scheme, specifically by obtaining a calibrated multi-view video sequence, wherein the calibrated multi-view video sequence includes a sampling time and a corresponding real image; initializing a 4D Gaussian primitive and a scale-aware residual field in a 4D space according to the calibrated multi-view video sequence; given the sampling time, determining the residual features of the 4D Gaussian primitive in the scale-aware residual field; calculating the residual of the survival state and the attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual features; combining the initial attributes of the 4D Gaussian primitive with the residual of the attribute to obtain a 3D Gaussian representation in a 3D space; The 3D Gaussian representation is projected to obtain a rendered image; gradient inversion is performed according to the loss between the rendered image and the real image, and the 4D Gaussian primitives are adaptively optimized and trained to obtain the 4D Gaussian representation, and the area occupied by the Gaussian primitives is encoded through a scale-aware residual field, so as to generate more accurate features and align with the self-splitting behavior of the Gaussian primitives; an adaptive optimization strategy is used to enhance the ability of the obtained 4D Gaussian representation to reconstruct high-frequency time information in dynamic scenes. The 4D Gaussian representation obtained through training can be rendered at any time and any perspective, realizing real-time rendering and ensuring high-quality reconstruction, thereby improving the rendering quality of dynamic scenes.

[0084] Based on any of the above embodiments of the present application, a third embodiment of the present application is proposed. In the third embodiment of the present application, the same or similar contents as any of the above embodiments can be referred to the above introduction and will not be repeated later. Figure 4 The step of determining the residual features of the 4D Gaussian primitives in the scale-aware residual field comprises steps S031 to S034:

[0085] Step S031, establishing a multi-scale feature representation framework through pyramid texture mapping stack technology;

[0086] Step S032, storing and optimizing the lowest resolution level by dynamically calculating the pyramid texture mapping level;

[0087] Step S033, determining a scaling level of the 4D Gaussian primitive in the multi-scale feature representation framework according to the projection size and basic spatial scale of the 4D Gaussian primitive;

[0088] Step S034, performing scale-aware feature extraction according to the minimum resolution level and the zoom level to obtain the residual feature.

[0089] Optionally, in order to fully integrate the spatiotemporal information of the scene and save computing resources, six planes are used in the embodiment of the present application to represent the scale-aware residual field M in the embodiment of the present application, which consists of only a spatial plane and a spatiotemporal plane. However, ignoring the size of the Gaussian primitives and projecting them onto the plane for feature extraction based only on their 4D positions will result in incorrect residual features. First, the Gaussian primitives can be approximated as ellipsoids. Therefore, when the Gaussian primitives are projected onto the spatial plane only, an elliptical area is obtained, rather than a single point in the current NeRF-based method. Therefore, the corresponding features of the Gaussian primitives should be a combination of all the areas it occupies. Secondly, if the self-splitting strategy of 3DGS is followed and the large Gaussian primitives are split into smaller primitives, they will have different residual features, which deviate significantly from the features of their parent primitives, which contradicts the original intention. Therefore, it is crucial to find an appropriate method to encode the area of ​​the Gaussian projection.

[0090] Optionally, in an embodiment of the present application, a scale-aware residual field is proposed to solve the above problem, which decomposes the 4D space into three spatial planes Pso and three spatiotemporal planes Pst. Given that the size of the Gaussian primitives only affects their projections in the spatial planes Pso, the embodiment of the present application specifically considers using scale-aware coding only therein.

[0091] Optionally, for each spatial plane Pi,j, a MipMap stack is used in the embodiment of the present application to represent features of different spatial scales in the scene. Level 0 of the MipMap stack is a feature map with a shape of M×N×N and has the smallest spatial scale s among all levels. 0 The remaining levels in the mipmap stack are obtained by calculating thumbnails based on the features of the previous level, where the width and height are reduced by a factor of 2 each time. Taking Pxy as an example, the relationship between their spatial scales is as follows:

[0092]

[0093] Optionally, variable B max and B min Represents the maximum and minimum values ​​of the scene bounding box, s l is the spatial scale of level l in the MipMap stack. In practice, only features in level 0 MipMap are stored and optimized, while the remaining levels are dynamically calculated and generated during forward inference, thus having the ability to encode features at different spatial scales in the scene.

[0094] Alternatively, for a Gaussian primitive with scale s in 4D space, when projected onto the spatial plane only, it produces a vector with axes (s x ,s y ). Therefore, based on the projection size of the Gaussian primitive on Pxy and the corresponding basic spatial scale, the zoom level l associated with the Gaussian primitive can be determined.

[0095]

[0096] Optionally, in order to maintain the highest accuracy, in the embodiment of the present application, the minimum value among them is selected as the final spatial level l=min(l x ,l y ). From this, we can get the two MipMap features closest to its spatial level: can be obtained with 4D position μ in Px,y 4d The Gaussian representation of :

[0097]

[0098] in, Indicated by The final representation of the scale-aware residual for a Gaussian primitive is:

[0099]

[0100] C so ={(x,y),(x,z),(y,z)}C st ={(x,t),(t,z),(y,t)}

[0101] Through the above scheme, this embodiment establishes a multi-scale feature representation framework through pyramid texture mapping stack technology; stores and optimizes the lowest resolution level by dynamically calculating the pyramid texture mapping level; determines the scaling level of the 4D Gaussian primitive in the multi-scale feature representation framework according to the projection size and basic spatial scale of the 4D Gaussian primitive; performs scale-aware feature extraction according to the lowest resolution level and the scaling level to obtain the residual feature, and encodes the area occupied by the Gaussian primitive through the scale-aware residual field, thereby generating more accurate features and aligning with the self-splitting behavior of the Gaussian primitive.

[0102] Based on any of the above embodiments of the present application, a fourth embodiment of the present application is proposed. In the fourth embodiment of the present application, the same or similar contents as any of the above embodiments can be referred to the above introduction and will not be repeated in the following. Figure 5The step of performing adaptive optimization training on the 4D Gaussian basis element to obtain the 4D Gaussian representation includes steps S071 to S073:

[0103] Step S071, obtaining the sampling probability of the 4D Gaussian basis element;

[0104] Step S072, calculating the definite integral of the 4D Gaussian basis element in the time domain according to the sampling probability to obtain a cumulative distribution function;

[0105] Step S073, dynamically adjusting the densification threshold and learning rate of the 4D Gaussian basis element according to the cumulative distribution function, and performing iterative training until a preset number of iterations is reached to obtain the 4D Gaussian representation.

[0106] Optionally, Gaussian primitives have different temporal positions and lifetimes in this 4D space, each sampled with different probabilities at observation time. To represent the temporal complexity of the scene, dynamic primitives usually have shorter lifetimes and lower sampling probabilities than static primitives. Primitives with less exposure throughout the temporal domain will have smaller gradients during the backpropagation of the loss function. The gradient value is crucial in the 3DGS framework because it needs to exceed a threshold to densify the corresponding primitives and optimize the currently imperfectly reconstructed regions. Therefore, directly applying the same optimization and densification strategy to each primitive may lead to an unbalanced optimization.

[0107] In order to solve the above problems, an adaptive optimization strategy is proposed in the embodiment of the present application. 4d The sampling probability within the observable time range dynamically adjusts the learning rate and the density gradient threshold. Specifically, in the embodiment of the present application, g 4d The γ(t) of is used to calculate the time integral within the observable range, which represents its sampling probability. The larger the integral, the more parts of the Gaussian basis element's lifetime intersect with the observable range, and the more likely it is to be sampled.

[0108] (a) Definite integral of time domain distribution: based on g 4d The state function γ(t) of , in the embodiment of the present application, its definite integral (t) in the time domain can be calculated to represent the CDF (cumulative distribution function) of each Gaussian basis element.

[0109]

[0110] I=F(t end )-F(t start )

[0111] Optionally, I represents the entire time domain from t start to end The definite integral of start and tend Normalized to 0 and 1 respectively.

[0112] Optionally, since γ(t) is a Gaussian-like function, it is challenging to calculate the exact definite integral value. In the embodiment of the present application, an approximate cumulative distribution function is derived:

[0113]

[0114] Alternatively, a = 0.070565992, b = 1.5976. Therefore, the definite integral of the Gaussian basis in the time domain can be obtained with minimal computational complexity.

[0115] (b) Integral-based Gaussian optimization strategy: After obtaining the time integral of each Gaussian, the learning rate and gradient threshold can be dynamically adjusted according to each primitive to achieve fast reconstruction of dynamic areas:

[0116]

[0117] Optionally, k i and lr i Respectively represent g 4d The density threshold and learning rate, while Ii and Imax represent g 4d The time integral of the Gaussian basis element and the maximum time integral of all Gaussian basis elements. Each time when densification control is required, the densification threshold is adjusted. In addition, in the embodiment of the present application, each time the preset number of iterations (for example, 50 times) is exceeded, according to the g 4d The learning rate is dynamically adjusted by parameters related to the 4D position, scale, rotation and zero-order SH coefficient.

[0118] Reference Figure 6 , Figure 6 Schematic diagram showing the effect comparison according to the fourth embodiment of the present application. Figure 6 As shown, when the method proposed in the embodiment of the present application and the methods in various prior arts are used to render the same scene view (e.g., a scene where a person stands), there is a significant difference in rendering quality, wherein the GT group is a real image, the Ours group is an image rendered using the dynamic scene reconstruction method in the embodiment of the present application, and the remaining groups are images rendered using prior art methods such as 4DGS, D-NeRF, KPlans, and Hexplane. As can be seen from the figure, the image rendered using the dynamic scene reconstruction method in the embodiment of the present application is closest to the real image, especially for the rendering of the details of the face and hands of the person, which has a significant improvement effect. The experimental results under monocular and multi-view settings show that the method in the embodiment of the present application achieves high-quality rendering quality while achieving real-time rendering.

[0119] This embodiment adopts the above scheme, specifically by obtaining the sampling probability of the 4D Gaussian basis element; calculating the definite integral of the 4D Gaussian basis element in the time domain according to the sampling probability to obtain the cumulative distribution function; dynamically adjusting the densification threshold and learning rate of the 4D Gaussian basis element according to the cumulative distribution function, and performing iterative training until a preset number of iterations is reached to obtain the 4D Gaussian representation. By introducing an adaptive optimization strategy, the ability of the model to reconstruct high-frequency time information in dynamic scenes is enhanced.

[0120] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the dynamic scene reconstruction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0121] This application also provides a dynamic scene reconstruction device, please refer to Figure 7 , the dynamic scene reconstruction device comprises:

[0122] A response module, configured to, in response to receiving a dynamic scene reconstruction instruction, determine a rendering time and a rendering viewing angle according to the dynamic scene reconstruction instruction;

[0123] A rendering module, configured to perform rendering based on a pre-trained 4D Gaussian representation according to the rendering time and the rendering viewing angle to obtain a target rendering view;

[0124] The 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in the 4D space based on the scale-aware residual field.

[0125] The dynamic scene reconstruction device provided by the present application adopts the dynamic scene reconstruction method in the above embodiment, which can solve the technical problem of dynamic scene reconstruction. Compared with the prior art, the beneficial effects of the dynamic scene reconstruction device provided by the present application are the same as the beneficial effects of the dynamic scene reconstruction method provided by the above embodiment, and other technical features in the dynamic scene reconstruction device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0126] The present application provides a dynamic scene reconstruction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the dynamic scene reconstruction method in the above-mentioned embodiment one.

[0127] Reference below Figure 8, which shows a schematic diagram of the structure of a dynamic scene reconstruction device suitable for implementing the embodiment of the present application. The dynamic scene reconstruction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The dynamic scene reconstruction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0128] like Figure 8 As shown, the dynamic scene reconstruction device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the dynamic scene reconstruction device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the dynamic scene reconstruction device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a dynamic scene reconstruction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided instead.

[0129] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0130] The dynamic scene reconstruction device provided by the present application adopts the dynamic scene reconstruction method in the above embodiment to solve the technical problem of dynamic scene reconstruction. Compared with the prior art, the beneficial effects of the dynamic scene reconstruction device provided by the present application are the same as the beneficial effects of the dynamic scene reconstruction method provided by the above embodiment, and the other technical features in the dynamic scene reconstruction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0131] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0132] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0133] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the dynamic scene reconstruction method in the above-mentioned embodiment.

[0134] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0135] The computer-readable storage medium may be included in the dynamic scene reconstruction device; or may exist independently without being assembled into the dynamic scene reconstruction device.

[0136] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by a dynamic scene reconstruction device, the dynamic scene reconstruction device: in response to receiving a dynamic scene reconstruction instruction, determines the rendering time and rendering perspective according to the dynamic scene reconstruction instruction; based on a pre-trained 4D Gaussian representation, renders according to the rendering time and rendering perspective to obtain a target rendering view; wherein the 4D Gaussian representation is obtained by adaptively optimizing the Gaussian primitives in the 4D space based on the scale-aware residual field.

[0137] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0138] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0139] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0140] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned dynamic scene reconstruction method, and can solve the technical problem of dynamic scene reconstruction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the dynamic scene reconstruction method provided in the above-mentioned embodiment, and will not be repeated here.

[0141] The present application also provides a computer program product, including a computer program, which implements the steps of the dynamic scene reconstruction method as described above when executed by a processor.

[0142] The computer program product provided by the present application can solve the technical problem of dynamic scene reconstruction. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the dynamic scene reconstruction method provided by the above embodiment, which will not be repeated here.

[0143] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A dynamic scene reconstruction method, characterized in that: The method comprises: In response to receiving a dynamic scene reconstruction instruction, determining a rendering time and a rendering viewing angle according to the dynamic scene reconstruction instruction; Based on the pre-trained 4D Gaussian representation, rendering is performed according to the rendering time and the rendering perspective to obtain a target rendering view; The 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in the 4D space based on the scale-aware residual field.

2. The method according to claim 1, characterized in that The step of rendering based on the pre-trained 4D Gaussian representation according to the rendering time and the rendering perspective to obtain the target rendering view also includes: Acquire a calibrated multi-view video sequence, wherein the calibrated multi-view video sequence includes a sampling time and a corresponding real image; Initializing a 4D Gaussian basis and a scale-aware residual field in a 4D space according to the calibrated multi-view video sequence; Given the sampling time, determining a residual feature of the 4D Gaussian basis element in the scale-aware residual field; Calculating the residual of the survival state and the attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature; combining the initial attributes of the 4D Gaussian primitives with the residuals of the attributes to obtain a 3D Gaussian representation in 3D space; Projecting the 3D Gaussian representation to obtain a rendered image; Gradient inversion is performed according to the loss between the rendered image and the real image, and adaptive optimization training is performed on the 4D Gaussian basis to obtain the 4D Gaussian representation.

3. The method according to claim 2, characterized in that The step of calculating the residual of the survival state and the attribute of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature comprises: Calculate the survival state of the 4D Gaussian primitive at the sampling time according to the sampling time and the residual feature; The residual feature of the 4D Gaussian primitive at the sampling time is decoded by a multi-layer perceptron, and combined with the survival state to obtain the residual of the attribute.

4. The method according to claim 2, characterized in that The scale-aware residual field is represented by six planes, including a spatial-only plane and a spatiotemporal plane.

5. The method according to claim 4, characterized in that The step of determining the residual features of the 4D Gaussian primitives in the scale-aware residual field comprises: A multi-scale feature representation framework is established through pyramid texture mapping stack technology; Stores and optimizes the lowest resolution level by dynamically calculating pyramid texture map levels; Determining a scaling level of the 4D Gaussian primitive in the multi-scale feature representation framework according to a projection size and a basic spatial scale of the 4D Gaussian primitive; Scale-aware feature extraction is performed according to the minimum resolution level and the zoom level to obtain the residual feature.

6. The method according to claim 2, characterized in that The step of performing adaptive optimization training on the 4D Gaussian primitives to obtain the 4D Gaussian representation comprises: Obtaining the sampling probability of the 4D Gaussian basis element; Calculate the definite integral of the 4D Gaussian basis element in the time domain according to the sampling probability to obtain a cumulative distribution function; The densification threshold and learning rate of the 4D Gaussian basis element are dynamically adjusted according to the cumulative distribution function, and iterative training is performed until a preset number of iterations is reached to obtain the 4D Gaussian representation.

7. A dynamic scene reconstruction device, characterized in that: The device comprises: A response module, configured to, in response to receiving a dynamic scene reconstruction instruction, determine a rendering time and a rendering viewing angle according to the dynamic scene reconstruction instruction; A rendering module, used for rendering based on the pre-trained 4D Gaussian representation according to the rendering time and the rendering perspective to obtain a target rendering view; The 4D Gaussian representation is obtained by adaptively optimizing and training Gaussian primitives in the 4D space based on the scale-aware residual field.

8. A dynamic scene reconstruction device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the dynamic scene reconstruction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the dynamic scene reconstruction method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the dynamic scene reconstruction method according to any one of claims 1 to 6 are implemented.

Images