Three-dimensional face reconstruction method based on micro-ray tracing

By employing a three-dimensional face reconstruction method based on micronizable ray tracing, and utilizing 3DMM and bidirectional reflectance distribution function to simulate the reflective properties of facial skin, combined with parametric virtual stage lighting and micronizable ray tracing technology, the reconstruction process is optimized. This solves the problems of high data acquisition cost and low environmental adaptability in existing three-dimensional face reconstruction technologies, achieving efficient and accurate three-dimensional face reconstruction.

CN118941723AInactive Publication Date: 2024-11-12NANCHANG UNIV

Patent Information

Application Number
CN202411421781.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-12
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing 3D face reconstruction technology is costly and has low adaptability in terms of data acquisition and model accuracy. Two-dimensional image data is easily affected by environmental factors, leading to distortion and inconsistency in the reconstruction model.

Method used

A method based on differentiable ray tracing is adopted to generate an initial 3D face model through 3DMM. The real environment is simulated by combining bidirectional reflection distribution function and parametric virtual stage lighting. The image is rendered using differentiable ray tracing technology, and the reconstruction process is optimized by landmark loss function and regularization method to accurately simulate the facial skin reflection characteristics and lighting conditions.

Benefits of technology

It improves the realism, robustness, and adaptability of 3D face models, enabling efficient and accurate reconstruction of 3D face models under ordinary equipment and complex lighting conditions. It reduces the mixed effects of illumination and reflectivity, and improves the accuracy and personalization of reconstruction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941723A_ABST
    Figure CN118941723A_ABST
Patent Text Reader

Abstract

The invention provides a three-dimensional face reconstruction method based on micro-ray tracing, and relates to the field of computer vision and graphics. The method comprises the following steps: generating an initial three-dimensional face model matched with an input image based on 3DMM; obtaining reflection characteristics of the initial three-dimensional face model based on statistical prior and a bidirectional reflection distribution function; simulating interaction between light and the initial three-dimensional face model based on parameterized virtual stage lighting and a micro light ray tracing technology to generate a lighting model; rendering the initial three-dimensional face model based on a micro ray tracing technology to obtain a composite image; and optimizing parameters of the composite image based on a landmark loss function, and optimizing the reconstruction process of the composite image based on a regularization method to obtain a three-dimensional face model. According to the three-dimensional face reconstruction method based on micro-ray tracing, the reconstructed three-dimensional face model is more real and natural, the accuracy and individuation degree of reconstruction are improved, and the robustness and adaptability of reconstruction are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of computer vision and graphics, and in particular to a three-dimensional face reconstruction method based on differentiable ray tracing. Background Art

[0002] 3D face reconstruction technology is a technology that restores 3D face structure from 2D images. With the rapid development of deep learning, computer graphics and physical simulation, 3D face reconstruction technology has gradually moved from theory to practical application, and has shown great potential in many fields such as medicine, film and television entertainment, and face recognition.

[0003] However, in terms of data acquisition and model accuracy, high-quality 3D face data usually relies on expensive 3D scanners for collection, which not only increases the cost of data acquisition, but also limits the size of the data set; small-scale data sets make it difficult to train a 3D face reconstruction model with strong generalization ability and high accuracy. In terms of environmental adaptability, although 2D image data is easy to obtain, it is easily affected by a variety of environmental factors such as lighting, angle, expression, and shadow; these factors are often difficult to accurately handle during the conversion process from 2D to 3D, resulting in distortion and inconsistency in the reconstructed 3D face model. Therefore, there is an urgent need to provide a solution to improve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a three-dimensional face reconstruction method based on differentiable ray tracing, which improves the problems of insufficient details, low robustness and low adaptability of three-dimensional face reconstruction in the prior art.

[0005] The present invention provides a 3D face reconstruction method based on differentiable ray tracing, which adopts the following technical solution: Generate an initial three-dimensional face model matching the input image based on the 3DMM, and obtain basic facial geometry based on the initial three-dimensional face model; Based on statistical priors and a bidirectional reflectance distribution function, a parameterized reflectance model is established by performing personalized modeling on the reflectance of the initial three-dimensional face model, and a reflectance characteristic of the facial skin is obtained based on the parameterized reflectance model; Based on parameterized virtual stage lighting, the lighting conditions in the real environment are simulated, and based on differentiable ray tracing technology, the interaction between light and the basic geometric shapes of the face is simulated to generate a lighting model; Rendering the initial three-dimensional face model based on differentiable ray tracing technology, the reflection characteristics and the illumination model to obtain a composite image with shadows and reflections; The parameters of the synthetic image are optimized based on a landmark loss function, and the reconstruction process of the synthetic image is optimized based on a regularization method to obtain a three-dimensional face model.

[0006] The above-mentioned three-dimensional face reconstruction method based on differentiable ray tracing, wherein the process of obtaining the basic geometric shape of the face based on the initial three-dimensional face model includes: A face model is constructed based on 3DMM and light consistency loss. The light consistency loss is to obtain sparse facial landmarks based on 2D landmark estimation technology and perform regularization. and translation vectors The pinhole camera model is used as the camera model, is a three-dimensional special orthogonal group; based on the rotation matrix and translation vector The initial three-dimensional face model is transformed into a camera coordinate system.

[0007] The three-dimensional face reconstruction method based on differentiable ray tracing as described above, wherein the process of performing personalized modeling on the reflectivity of the initial three-dimensional face model based on statistical priors and a bidirectional reflectance distribution function includes: The facial skin reflectivity is modeled based on the bidirectional reflectance distribution function, where the reflectivity includes diffuse reflection and specular reflection. Based on statistical priors, the data of real facial skin diffuse reflection and specular reflection are collected and analyzed, and the distribution law of reflective properties of surfaces of different materials is estimated to establish a parameterized reflection model.

[0008] In the above-mentioned three-dimensional face reconstruction method based on differentiable ray tracing, the process of obtaining the reflection characteristics of the facial skin includes: Based on the parameterized reflection model, diffuse reflection and specular reflection are adjusted to control the reflection characteristics of different facial skin under different lighting conditions, and obtain specific details of facial skin; the specific details include facial hair, moles, color and oiliness.

[0009] The three-dimensional face reconstruction method based on differentiable ray tracing as described above, wherein the lighting conditions in a real environment are simulated based on parameterized virtual stage lighting, comprises: The basic geometric shape of the face is configured as an icosahedron geometric structure, a virtual stage light is defined based on the icosahedron, and an area light is placed at the center of each triangle of the light table icosahedron. The virtual stage light is based on a control mechanism of soft frame constraints, so that the area light is always kept within a defined homogeneous area. The area light controls the position and size of the incident light beam and determines the position and appearance of the self-shadow.

[0010] The three-dimensional face reconstruction method based on differentiable ray tracing as described above, wherein the interaction between light and the basic geometric shape of the face is simulated based on differentiable ray tracing technology to generate an illumination model, including: controlling the intensity, position and surface area of ​​the light based on turning on or off the regional light, thereby realizing modeling of self-shadowing scenes and illumination models.

[0011] As described above, the three-dimensional face reconstruction method based on differentiable ray tracing, wherein the process of obtaining a synthetic image with shadows and reflections includes: selecting a subset of the reflection characteristics, synthesizing the image based on the illumination model and differentiable ray tracing technology, and the differentiable ray tracing technology uses a Gaussian smoothing method.

[0012] The three-dimensional face reconstruction method based on differentiable ray tracing as described above, wherein, before optimizing the parameters of the synthetic image based on the landmark loss function, it includes: measuring the distance between the projection of the facial landmark and its corresponding pixel projection on the input image based on the sparse facial landmark loss.

[0013] The three-dimensional face reconstruction method based on differentiable ray tracing as described above, wherein the parameters of the synthetic image are optimized based on the landmark loss function, including: optimizing camera parameters and BlendShape coefficients based on the landmark loss function, introducing statistical albedo, facial geometry and expression, lighting, and camera optimization after initializing the posture and expression, while keeping the specular reflection, diffuse reflection, and roughness unchanged, and independently modeling the statistical reflectivity and lighting.

[0014] The three-dimensional face reconstruction method based on differentiable ray tracing as described above, wherein the three-dimensional face model is obtained by optimizing the synthetic image reconstruction process based on a regularization method, comprising: A symmetry constraint is introduced based on a symmetry regularizer to penalize asymmetric changes produced in the process of reconstructing the synthetic image, thereby maintaining the balance and consistency of image information between the two sides of the face. Based on consistency regularization, the chromaticity deviation caused by self-shadowing in the process of reconstructing the synthetic image is reduced to obtain a reconstructed three-dimensional face model.

[0015] The beneficial effects of the 3D face reconstruction method based on differentiable ray tracing proposed in the present invention are: 1. The present invention performs parameterized modeling of the reflectivity of facial skin through a bidirectional reflectance distribution function, which can accurately simulate the reflective characteristics of facial skin, making the reconstructed three-dimensional face model more realistic and natural.

[0016] 2. The present invention proposes a virtual lighting stage lighting model, which sets the lighting environment by parameterizing virtual stage lights. It can flexibly adjust parameters such as light source position, color, and intensity to adapt to different lighting conditions, thereby improving the robustness and adaptability of reconstruction.

[0017] 3. The present invention introduces differentiable ray tracing technology to achieve efficient and accurate reconstruction of three-dimensional face models from two-dimensional images under common equipment and complex lighting environments.

[0018] 4. The present invention optimizes the synthetic image through a regularization method, effectively avoiding the mixed influence of reflectivity and illumination, and improving the accuracy and personalization of reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart of a 3D face reconstruction method based on differentiable ray tracing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0021] The embodiment of the present invention provides a 3D face reconstruction method based on differentiable ray tracing, see Figure 1 ,include: S1. Generate an initial three-dimensional face model matching the input image based on the 3DMM, and obtain basic facial geometric shapes based on the initial three-dimensional face model; S2, based on statistical priors and bidirectional reflectance distribution function, the reflectance of the initial three-dimensional face model is personalized and modeled to establish a parameterized reflectance model, and the reflectance characteristics of the facial skin are obtained based on the parameterized reflectance model; S3, simulating lighting conditions in a real environment based on parameterized virtual stage lighting, and simulating the interaction between light and the basic geometric shape of the face based on differentiable ray tracing technology to generate a lighting model; S4, rendering the initial three-dimensional face model based on the differentiable ray tracing technology, the reflection characteristics and the illumination model to obtain a composite image with shadows and reflections; S5. Optimizing the parameters of the synthetic image based on a landmark loss function, and optimizing the synthetic image reconstruction process based on a regularization method to obtain a three-dimensional face model.

[0022] In some embodiments, the process of executing step S1 includes: S1.1, constructing a face model; S1.2. Build a camera model.

[0023] Specifically, when executing step S1.1, constructing the face model, the face model is constructed based on 3DMM and light consistency loss, and the light consistency loss is to obtain sparse facial landmarks based on 2D landmark estimation technology and perform regularization; Furthermore, the sparse facial landmarks usually refer to points on the face that have significant features and relatively fixed positions, such as the corners of the eyes, the corners of the mouth, the tip of the nose, etc. The facial landmarks can help the model more accurately locate and understand the structure and features of the face, and help to accurately detect and locate landmark points in facial images when light consistency loss is limited, especially under low light, heavy mirror or self-shadow conditions.

[0024] Specifically, when executing step S1.2 and constructing the camera model, based on the rotation matrix and translation vector The pinhole camera model is used as the camera model, is a three-dimensional special orthogonal group; based on the rotation matrix and translation vector The initial three-dimensional face model is transformed into a camera coordinate system.

[0025] Further, the face model vertices The conversion formula mapped to the camera coordinate system is as follows: , in, are the camera parameters, R is the rotation matrix, and T is the translation vector.

[0026] In some embodiments, the process of executing step S2 includes: S2.1. Based on statistical priors and a bidirectional reflectance distribution function, a parameterized reflectance model is established by performing personalized modeling on the reflectance of the initial three-dimensional face model; S2.2. Obtaining the reflection characteristics of the facial skin based on the parameterized reflection model.

[0027] Specifically, when executing step S2.1, it includes: The facial skin reflectivity is modeled based on the bidirectional reflectance distribution function, where the reflectivity includes diffuse reflection and specular reflection. Based on statistical priors, the data of real facial skin diffuse reflection and specular reflection are collected and analyzed, and the distribution law of reflective properties of surfaces of different materials is estimated to establish a parameterized reflection model.

[0028] In fact, when using the bidirectional reflectance distribution function for reflection modeling, it defines the reflectivity of diffuse, specular, and roughness at each geometric vertex as follows: , in, represents the specular reflectivity, Indicates the roughness, represents diffuse reflectance, Represents a vertex The normal line at represents the light direction of the incident area, The viewing direction pointing to the pinhole camera origin, is the material term for diffusely reflected light in all directions, is the specular term for light reflected from a given viewing direction.

[0029] In fact, roughness is used to describe the microscopic geometric structure, particle distribution and other characteristics of the object surface, which can usually be directly obtained through experimental measurement or image processing. Diffuse reflection and specular reflection are not only related to the roughness of the object surface, but also affected by many other factors, such as the position of the light source, the angle of the observer, the material of the surface, etc. In this case, statistical priors can help us better understand and predict reflection behavior.

[0030] Furthermore, in the UV image texture space, Defined as the statistical diffuse reflectance, Defined as specular reflectance, Defined as the roughness with no given statistical prior, with a texture resolution of M×M; for each projected vertex onto the texture , , , Described separately Interpolated color, sharpness, and roughness factors for .

[0031] Specifically, when executing step S2.2, it includes: adjusting diffuse reflection and specular reflection based on the parameterized reflection model, controlling the reflection characteristics of different facial skins under different lighting conditions, and obtaining specific details of the facial skin; the specific details include facial hair, moles, color and oiliness.

[0032] In some embodiments, the process of executing step S3 includes: S3.1, simulate the lighting conditions in real environment based on parameterized virtual stage lighting; S3.2. Generate a lighting model by simulating the interaction between light and the basic geometric shapes of the face based on differentiable ray tracing technology.

[0033] Specifically, when executing step S3.1, simulating the lighting conditions in a real environment based on the parameterized virtual stage lighting includes: The basic geometric shape of the face is configured as an icosahedron geometric structure, a virtual stage light is defined based on the icosahedron, and an area light is placed at the center of each triangle of the light table icosahedron. The virtual stage light is based on a control mechanism of soft frame constraints, so that the area light is always kept within a defined homogeneous area. The area light controls the position and size of the incident light beam and determines the position and appearance of the self-shadow.

[0034] Furthermore, the regional light parameter set of the virtual light stage is as follows: , in, represents the distance from the origin to the face geometry, represents the relative surface area, represents the local position of the light center in the face triangle in the centroid coordinates, Indicates the perceived intensity. The corresponding light can be turned off.

[0035] In fact, there is no direct dependency between the parameters. When the light source intensity needs to be adjusted, in the above formula, only the parameter To modify the intensity, it simplifies the optimization process and makes adjusting lighting effects more intuitive and controllable.

[0036] Specifically, when executing step S3.2, simulating the interaction between light and the basic geometric shape of the face based on differentiable ray tracing technology to generate a lighting model, it includes: controlling the intensity, position and surface area of ​​the light based on turning on or off the regional light, so as to achieve modeling of the self-shadowing scene and the lighting model.

[0037] In some embodiments, the process of executing step S4 includes: A subset of the reflective properties is selected and an image is synthesized based on an illumination model and a differentiable ray tracing technique employing a Gaussian smoothing method.

[0038] Specifically, use the parameter Synthesize images with differentiable ray tracing technology to reduce photo consistency loss between synthesized and real images on a per-pixel basis; , is the distance from the origin to the face geometry, is the relative surface area, is the local position of the light center in the face triangle in the centroid coordinates, For perceived intensity, is the BlendShape coefficient, is the facial reflectivity, is the deformation coefficient of the facial geometry, It is a set of regional light parameters for the virtual lighting stage. is the rotation matrix, and T is the translation vector of the camera.

[0039] The loss formula is as follows: , in, is the consistency loss, , The ray tracing pixel color and the real image pixel color are respectively. The rendering pixel color is Given, F is the Monte Carlo estimator of the rendering equation.

[0040] In some embodiments, the process of executing step S5 includes: S5.1. Optimizing parameters of the synthetic image based on a landmark loss function; S5.2. Optimize the synthetic image reconstruction process based on a regularization method to obtain a three-dimensional face model.

[0041] Specifically, when executing step S5.1, optimizing the parameters of the synthetic image based on the landmark loss function includes: The distance between the projection of facial landmarks and their corresponding pixel projections on the input image is measured based on the sparse facial landmark loss; The camera parameters and BlendShape coefficients are optimized based on the landmark loss function. After initializing the posture and expression, the statistical albedo, facial geometry and expression, lighting, and camera optimization are introduced, while the specular reflection, diffuse reflection, and roughness are kept unchanged, and the statistical reflectivity and lighting are independently modeled.

[0042] The loss formula when the statistical reflectance and illumination are modeled independently is: , , in, is a priori that guarantees optimization traceability, is a constraint box, is the loss function, is the consistency loss, is the projection of the facial landmark and its corresponding pixel projection on the input image The distance between is the BlendShape coefficient, is the facial reflectivity, is the deformation coefficient of the facial geometry, It is a set of regional light parameters for the virtual lighting stage. and Make sure the area light stays inside the uniform icosahedron, is the optimized BlendShape coefficient.

[0043] Specifically, when executing step S5.2, optimizing the synthetic image reconstruction process based on the regularization method to obtain the three-dimensional face model, it includes: A symmetry constraint is introduced based on a symmetry regularizer to penalize asymmetric changes produced in the process of reconstructing the synthetic image, thereby maintaining the balance and consistency of image information between the two sides of the face. Based on consistency regularization, the chromaticity deviation caused by self-shadowing in the process of reconstructing the synthetic image is reduced to obtain a reconstructed three-dimensional face model.

[0044] In fact, the following information is extracted from the image during the optimization process: , in, is the new parameter set, is the light stage parameter, is the BlendShape coefficient, is the facial reflectivity, , , For lossy personalized diffuse, specular, and roughness, is the rotation matrix, T is the camera's translation vector, , , is the weighted coefficient of the loss function, , is a symmetric regularizer, , is the consistency regularization, , , It is a constraint that ensures that each vertex is locally smooth relative to a ring of vertices in the UV space. The ring of vertices usually represents the direct neighboring points of a vertex. , are soft and constraints, limiting the specular reflection and roughness to remain within acceptable ranges respectively.

[0045] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A 3D face reconstruction method based on differentiable ray tracing, characterized in that: The following steps are involved: Generate an initial three-dimensional face model matching the input image based on the 3DMM, and obtain basic facial geometry based on the initial three-dimensional face model; Based on statistical priors and a bidirectional reflectance distribution function, a parameterized reflectance model is established by performing personalized modeling on the reflectance of the initial three-dimensional face model, and a reflectance characteristic of the facial skin is obtained based on the parameterized reflectance model; Based on parameterized virtual stage lighting, the lighting conditions in the real environment are simulated, and based on differentiable ray tracing technology, the interaction between light and the basic geometric shapes of the face is simulated to generate a lighting model; Rendering the initial three-dimensional face model based on the reflection characteristics of the differentiable ray tracing technology and the illumination model to obtain a composite image with shadows and reflections; The parameters of the synthetic image are optimized based on a landmark loss function, and the reconstruction process of the synthetic image is optimized based on a regularization method to obtain a three-dimensional face model.

2. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: The process of acquiring the basic geometric shape of the face based on the initial three-dimensional face model includes: A face model is constructed based on 3DMM and light consistency loss, wherein the light consistency loss is to obtain sparse facial landmarks based on 2D landmark estimation technology and perform regularization; Based on the rotation matrix and translation vectors The pinhole camera model is used as the camera model, is a three-dimensional special orthogonal group; Based on the rotation matrix and translation vector The initial three-dimensional face model is transformed into a camera coordinate system.

3. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: The process of performing personalized modeling on the reflectivity of the initial three-dimensional face model based on statistical priors and a bidirectional reflectance distribution function includes: The facial skin reflectivity is modeled based on the bidirectional reflectance distribution function, where the reflectivity includes diffuse reflection and specular reflection. Based on statistical priors, the data of real facial skin diffuse reflection and specular reflection are collected and analyzed, and the distribution law of reflective properties of surfaces of different materials is estimated to establish a parameterized reflection model.

4. The 3D face reconstruction method based on differentiable ray tracing according to claim 3, characterized in that: The process of obtaining the reflective properties of facial skin includes: Based on the parameterized reflection model, diffuse reflection and specular reflection are adjusted to control the reflection characteristics of different facial skin under different lighting conditions, and obtain specific details of facial skin; the specific details include facial hair, moles, color and oiliness.

5. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: Based on parameterized virtual stage lighting, it simulates the lighting conditions in real environments, including: The basic geometric shape of the face is configured as an icosahedron geometric structure, a virtual stage light is defined based on the icosahedron, and an area light is placed at the center of each triangle of the light table icosahedron. The virtual stage light is based on a control mechanism of soft frame constraints, so that the area light is always kept within a defined homogeneous area. The area light controls the position and size of the incident light beam and determines the position and appearance of the self-shadow.

6. The 3D face reconstruction method based on differentiable ray tracing according to claim 5, characterized in that: The interaction between light and the basic geometric shape of the face is simulated based on differentiable ray tracing technology to generate a lighting model, including: controlling the intensity, position and surface area of ​​the light based on turning on or off the regional light, so as to achieve modeling of self-shadowing scenes and lighting models.

7. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: The process of obtaining a composite image with shadows and reflections includes: selecting a subset of the reflection characteristics, synthesizing the image based on an illumination model and a differentiable ray tracing technique, wherein the differentiable ray tracing technique uses a Gaussian smoothing method.

8. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: Before optimizing the parameters of the synthesized image based on the landmark loss function, it includes: measuring the distance between the projection of the facial landmark and the corresponding pixel projection on the input image based on the sparse facial landmark loss.

9. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: Optimizing the parameters of the synthetic image based on the landmark loss function includes: The camera parameters and BlendShape coefficients are optimized based on the landmark loss function. After initializing the posture and expression, the statistical albedo, facial geometry and expression, lighting, and camera optimization are introduced, while the specular reflection, diffuse reflection, and roughness are kept unchanged, and the statistical reflectivity and lighting are independently modeled.

10. The 3D face reconstruction method based on differentiable ray tracing according to claim 1, characterized in that: The three-dimensional face model is obtained by optimizing the synthetic image reconstruction process based on a regularization method, including: A symmetry constraint is introduced based on a symmetry regularizer to penalize asymmetric changes produced in the process of reconstructing the synthetic image, thereby maintaining the balance and consistency of image information between the two sides of the face. Based on consistency regularization, the chromaticity deviation caused by self-shadowing in the process of reconstructing the synthetic image is reduced to obtain a reconstructed three-dimensional face model.

Citation Information

Patent Citations

  • Facial reconstruction method based on non-supervision automatic encoder

    CN107292953A

  • Three-dimensional face reconstruction method and device based on single-frame image, equipment and medium

    CN110428491A

  • Single-image face three-dimensional reconstruction method based on illumination prior

    CN111445582A

  • Single-image three-dimensional face reconstruction method based on video supervision

    CN111950477A

  • Transparent object reconstruction method and system capable of micro rendering based on an SDF field

    CN113888694A

Cited By

  • Three-dimensional object material reconstruction method and system based on area light source

    CN120411337A