A multi-style image migration and rendering method and system in indoor scene design

Abstract

The application discloses a kind of multi-style image migration and rendering method and system in indoor scene design.The surface domain of target indoor three-dimensional scene, light source group and diagnostic view angle are obtained, surface identification information and light source response data are generated;Multiple reference style images are decomposed into material reflection carrier and lighting carrier, and the material reflection carrier is bound to the corresponding surface domain;Surface-light source influence relationship is established based on light source response data, and the light parameter shared by the same light source group is calculated;According to the directional adjustment requirement of different surface domains to the same light source group, conflict information is determined to constrain light update and lighting carrier weight;The preliminary rendering residual is attributed to double domains using material response and illumination response, the surface material parameters and shared light parameters are updated respectively, and the target view rendering is completed based on the updated scene parameters.The application reduces fixed shadows and repeated highlights, maintains multi-view material consistency and light editability.

Description

Technical Field

[0001] This invention belongs to the fields of computer graphics, 3D image rendering and computer-aided interior design. Specifically, it relates to a method and system for multi-style image transfer and rendering in interior scene design. Background Technology

[0002] Interior design renderings are typically created by rendering a 3D scene that includes objects such as walls, floors, ceilings, cabinets, soft furnishings, and lighting fixtures. To combine these elements into a design scheme, designers often select wall textures, wood colors, fabric textures, and lighting atmospheres from multiple reference images, and then apply the styles from different reference images to the same target interior 3D scene.

[0003] Existing techniques can perform 2D style transfer on the final rendered image and can also write the stylization results into the UV map, 3D Gaussian field, or radiative field of a 3D model. However, the pixel appearance in the reference image is usually the result of the combined effects of material reflectivity, incident lighting, occlusion shadows, specular highlights, camera exposure, and tone mapping. If this pixel appearance is directly transferred to the material appearance of the target scene, the shadows and highlights in the reference image will be fixed in the material, and the target scene's own light source will generate new shadows and highlights when rendered again, resulting in fixed shadows, repeated highlights, and exposure overlay.

[0004] Existing technology CN118967915B discloses a 3D Gaussian scene style transfer method based on 2D priors. It improves appearance consistency across different viewpoints by supervising the appearance parameters of a 3D Gaussian field through stylized images. However, the stylized pixels, as the supervision target, are still coupled with material and lighting. CN113052972B discloses a lighting effect enhancement method based on neural networks. It uses the linear superposition of single-source unit light intensity rendering maps to obtain the light source intensity and color. However, it focuses on overall lighting effect enhancement and does not address the problem of multiple reference style images applying opposite adjustment requirements to the same physical light source when applied to different surface domains.

[0005] In multi-style combination scenes, wood grain from a warm-toned lighting environment, wall surfaces from a cool-toned sunlight environment, and stone from a high-exposure showroom environment may all be located in the same living / dining room and actually affected by the same chandelier and the same window light. If the original color temperature, brightness gradient, and highlights of each reference image are preserved in the local material, the same beam of light will appear with inconsistent colors and brightness on adjacent surfaces; when the designer adjusts the chandelier's color temperature, intensity, or viewing angle, the original bright spots and shadows in the texture will not change synchronously. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the basic concept of the technical solution adopted by the present invention is as follows:

[0007] A method for multi-style image transfer and rendering in interior scene design includes the following steps:

[0008] Step S1: Divide the target indoor 3D scene, which includes geometric model, surface material parameters, camera parameters and adjustable light source parameters, into multiple surface domains and multiple light source groups, select multiple diagnostic viewpoints, and obtain surface identification information and light source response data of each light source group when participating in rendering according to preset benchmark parameters under the diagnostic viewpoints;

[0009] Step S2: Perform exposure normalization and intrinsic component decomposition on at least two reference style images to obtain the material reflection carrier and illumination carrier of each reference style image. Map the material reflection carrier to the corresponding surface domain according to the surface correspondence and style weight to obtain the surface material parameters shared by multiple diagnostic perspectives. Based on the surface identification information, perform surface domain aggregation on the light source response data to establish the surface-light source influence relationship. Solve the shared lighting parameters of each light source group according to the illumination carrier corresponding to each surface domain and the surface-light source influence relationship, so that the same light source group corresponds to the same set of shared lighting parameters in different surface domains.

[0010] Step S3: Determine the directional adjustment requirements of multiple surface domains for the same light source group, obtain light source conflict information based on the directional consistency among the directional adjustment requirements, and generate update constraints on the corresponding shared light parameters and weight constraints on the corresponding lighting carriers based on the light source conflict information; perform preliminary rendering of the target indoor 3D scene using the surface material parameters and the shared light parameters, and obtain rendering residuals based on the preliminary rendering results and the style target composed of the material reflection carriers and lighting carriers;

[0011] Step S4: Determine the material response relationship corresponding to the adjustable surface material parameters and the lighting response relationship corresponding to each light source group. Under the update constraint and weight constraint, perform material domain-lighting domain attribution on the rendering residual. Write the attributed material increment into the corresponding surface domain and the attributed lighting increment into the shared light parameters of the corresponding light source group. Render an indoor scene image from at least one target viewpoint based on the updated surface material parameters and shared light parameters.

[0012] As a preferred embodiment of the present invention, the acquisition of surface marking information and light source response data from a diagnostic perspective includes: outputting a surface marking map, UV coordinate map, normal map and depth map for each diagnostic perspective; keeping the camera parameters, geometric model and surface material parameters of the target indoor three-dimensional scene unchanged, setting each light source group to a unit intensity or preset response level in sequence, and turning off the remaining adjustable light source groups to obtain the corresponding unit light source response map or segmented light source response map;

[0013] The set of pixels for each surface domain in the unit light source response map or the segmented light source response map is determined according to the surface identification map. The set of pixels is then weighted and aggregated according to pixel visibility, viewing angle importance and response brightness to form a surface-light source influence matrix that characterizes the actual contribution of each light source group to each surface domain.

[0014] As a preferred embodiment of the present invention, the exposure normalization and intrinsic component decomposition of the reference style image includes: converting the reference style image to a linear color space, and forming a specular mask based on at least two of the following: high brightness, saturation, local contrast and normal-viewing line relationship;

[0015] Within a smooth range defined by the material region boundary and surface normal, the reflection term and illumination term are jointly estimated to obtain the material reflection carrier, diffuse illumination carrier, and viewpoint-dependent specular carrier. At least one of the specular width, peak value, directionality, and color is extracted from the viewpoint-dependent specular carrier as a constraint on the target surface roughness, metallicity, or specular reflection parameters, and the viewpoint-dependent specular carrier and local projected shadows are prevented from being written into the base color texture of the target surface.

[0016] As a preferred embodiment of the present invention, the step of mapping the material reflection carrier to the corresponding surface domain includes: setting multiple non-negative style weights corresponding to multiple reference style images for each surface domain, and making the sum of the non-negative style weights equal to 1; establishing a mapping to the target surface domain according to the texture principal direction, world scale, repetition period, and boundary clipping rules of the reference material block; fusing the mapped material reflection carrier in the UV coordinate domain or world coordinate material domain, setting splicing continuity constraints for multiple UV islands on the same continuous surface, and setting texture scale constraints for different objects sharing the same material group.

[0017] As a preferred embodiment of the present invention, the step of solving the shared lighting parameters of each light source group includes: extracting a lighting descriptor from each lighting carrier that includes at least one of normalized luminance, illuminance chromaticity, primary light-secondary light ratio, and spatial luminance gradient; and obtaining the desired lighting descriptor for each surface domain based on surface correspondence and style weights.

[0018] Within the physical range of the shared lighting parameters, the weighted difference between the predicted lighting descriptor determined by the surface-light source influence relationship and the desired lighting descriptor, as well as the offset of the shared lighting parameters relative to the original lighting parameters of the target indoor 3D scene, is minimized to obtain shared lighting parameters that include at least one of the light source intensity coefficient and color coefficient.

[0019] As a preferred embodiment of the present invention, let the directional adjustment requirement of the m-th surface domain for the g-th light source group be: The importance of the m-th surface domain is The light source conflict information includes information according to...

[0020] The light source conflict coefficient Cg is determined, where ε is a positive number; when the light source conflict coefficient is greater than the conflict threshold, the allowable update range of the shared lighting parameters of the corresponding light source group is reduced, the participation weight of the corresponding lighting carrier is reduced, and the response quantity of the corresponding light source group in the illumination response relationship is multiplied by 1-Cg or the response quantity is frozen.

[0021] As a preferred embodiment of the present invention, the material response relationship is obtained by performing perturbation rendering or calculating the shader analytical derivative on at least one adjustable surface material parameter among base color, roughness, normal intensity, and metallicity. The lighting response relationship is obtained based on the light source response data and its color channels. The material domain-lighting domain attribution is also subject to the consistency constraints of de-illumination reflection at the same surface sampling point under multiple diagnostic viewpoints and the material continuity constraints across UV island boundaries. It is iterated in the following order: first, fixing the shared light parameters and updating the surface material parameters; then, updating the light source group with conflicts below the threshold; and finally, jointly fine-tuning, until the rendering residual decrease, parameter update amount, or number of iterations meets the preset stopping condition.

[0022] As a preferred embodiment of the present invention, the invention also discloses a multi-style image transfer and rendering system for indoor scene design, comprising:

[0023] The scene response acquisition module is used to divide the target indoor 3D scene into multiple surface domains and multiple light source groups, select multiple diagnostic viewpoints, and acquire surface identification information and light source response data of each light source group when participating in rendering according to preset benchmark parameters.

[0024] The style carrier separation module is used to perform exposure normalization and intrinsic component decomposition on at least two reference style images to obtain material reflection carrier and illumination carrier;

[0025] The surface style binding module is used to map the material reflection carrier to the corresponding surface domain according to the surface correspondence and style weight, so as to obtain the surface material parameters shared by multiple diagnostic perspectives;

[0026] The shared illumination solution module is used to aggregate the light source response data into surface domains based on surface identification information to establish the surface-light source influence relationship. It solves the shared lighting parameters of each light source group based on the lighting carrier corresponding to each surface domain, and obtains the light source conflict information based on the directional consistency between the directional adjustment requirements of multiple surface domains for the same light source group, so as to generate update constraints for shared lighting parameters and weight constraints for lighting carriers.

[0027] The dual-domain iteration module is used to obtain the rendering residual based on the initial rendering results. Based on the material response relationship and the lighting response relationship, the rendering residual is attributed to the material domain and the lighting domain under the update constraint and weight constraint. The material increment is written to the corresponding surface domain, and the lighting increment is written to the shared light parameters of the corresponding light source group.

[0028] And a unified rendering module for rendering indoor scene images from at least one target viewpoint based on updated surface material parameters and shared lighting parameters.

[0029] Compared with the prior art, the present invention has the following advantages:

[0030] Compared to directly transferring the final pixel appearance, this invention writes the material reflection carrier obtained from the reference style image into the surface material parameter domain and uses the illumination tendency to solve the shared light parameters, which can reduce the risk of the reference illumination being repeatedly superimposed on the target light source after being solidified to the target material.

[0031] This invention establishes a surface-light source influence relationship based on the light source response data of the target 3D scene. This relationship includes the influence of geometric occlusion, surface normals, indirect reflections, and the actual rendering pipeline on the light source. The same light source group is solved by multiple surface domains across the entire field and has only one set of shared lighting parameters, which can improve the illumination continuity of adjacent surfaces.

[0032] This invention determines light source conflict information based on the directional adjustment requirements of multiple surface domains for the same light source group. When the conflict is large, it simultaneously restricts the updating of shared light parameters and the weight of the corresponding lighting carrier, avoiding the destruction of the whole field illumination in order to match the local reference image, and preventing the local shadow contour from re-entering the base color texture.

[0033] This invention utilizes material response and lighting response relationships to perform dual-domain attribution on rendering residuals, with material increments and lighting increments having clearly defined write locations; all target viewpoints are uniformly generated from the updated 3D scene parameters, which can maintain cross-viewpoint material consistency on the same physical surface and recalculate shadows and specular highlights as lighting or viewpoint changes.

[0034] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0035] In the attached diagram:

[0036] Figure 1 This is a flowchart of a multi-style image migration and rendering method in indoor scene design according to the present invention;

[0037] Figure 2 This is a schematic diagram of the structure of a multi-style image transfer and rendering system for indoor scene design according to the present invention;

[0038] Figure 3 This is a schematic diagram of the material domain-illumination domain residual attribution process in this invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments will be clearly and completely described below with reference to the accompanying drawings. The following embodiments are used to illustrate the present invention.

[0040] Example 1:

[0041] like Figure 1 As shown in the figure, the method of multi-style image migration and rendering in indoor scene design in this embodiment includes steps S101 to S106.

[0042] Step S101: Obtain surface identification information and light source response data of the target indoor 3D scene. Divide the target indoor 3D scene into surface domains and light source groups, and select a diagnostic viewpoint covering the main design area. Under each diagnostic viewpoint v, the renderer outputs a surface identification map. UV coordinate diagram Normal diagram Depth map and basic material diagrams Surface marking diagram The pixel value in the UV coordinate map indicates the surface region to which the corresponding pixel belongs. Give the UV coordinates of the corresponding pixel in the surface domain.

[0043] Keeping the camera parameters, geometric model, and surface material parameters unchanged, set each light source group to unit intensity or preset response level in sequence, and turn off the other adjustable light source groups. Render each light source group g at the diagnostic viewpoint v to obtain its light response map. For ambient ambient light or uncontrollable indirect light that needs to be preserved, a separate baseline response map can be generated. The preset response level can be one or more sampled values ​​of light source intensity, RGB color coefficients, or color temperature. For light source groups with piecewise linear responses, light source response maps can be generated separately at multiple response levels and interpolated between adjacent levels.

[0044] The light source response maps are aggregated to each surface domain according to the surface identification map. The average degree of influence of the g-th light source group on the m-th surface domain is calculated. It can be determined according to the following formula:

[0045]

[0046] In the formula, For brightness or a specified color channel response value, It is the product of pixel visibility, viewpoint importance, and pixel confidence. This represents the sum of the corresponding weights. (All) Forming the surface-light source influence matrix A. Due to... It is derived from the actual light source response data generated by the target renderer, and therefore can simultaneously reflect the effects of geometric occlusion, surface normal orientation, indirect reflection, and the contribution of the rendering pipeline to the light source, rather than being limited to estimating the light source effect based on spatial distance.

[0047] Step S102: Perform exposure normalization and intrinsic component decomposition on the reference style image. First, perform material region segmentation on the reference style image and obtain estimated normals. The material region includes at least one of wall, wood, stone, fabric, metal, and glass. The material region segmentation and estimated normals can be obtained through image segmentation models, depth estimation models, manual selection, or a combination thereof, and are used to define the smoothing range of lighting items and the range of material texture preservation.

[0048] A specular mask is formed based on at least two of the following: high brightness, low saturation, local contrast, and normal-viewing distance relationship. For pixels outside the specular mask, the reflectance term can be jointly estimated in the logarithmic domain. and lighting items For example, an objective function can be used that includes data terms, lighting smoothing terms, normal lighting fitting terms, and reflection texture preservation terms:

[0049]

[0050] In the formula, , It is a low-order spherical harmonic basis of the surface normal. The illumination coefficients are used as a reference image; when pixels p and q belong to the same material region and their normals are close, Take the larger value at the object boundary. Take the smaller value; at the edges of the color texture Take a smaller value to avoid misclassifying wood grain, stone grain, or fabric texture as lighting variations. After solving, let... .

[0051] For areas covered by the specular mask, instead of directly incorporating their color patterns into R_k, at least one of the specular width, peak value, directionality, and color is extracted from H_k and used as candidate constraints for target surface roughness, metallicity, or specular reflection parameters. For exposure differences in the reference style image, normalization can be performed using the median brightness of the same material region, estimated grayscale values, or color constancy results, ensuring that subsequent processing retains the relative illumination tendency rather than the absolute exposure values ​​of the reference style image.

[0052] From each lighting carrier Extracting lighting descriptive quantities Illumination descriptors may include at least one of the following: normalized luminance, illumination chromaticity or color temperature, principal-to-complement ratio, top-to-bottom luminance gradient, principal light direction, and soft shadow scale. Illumination descriptors are technical data used to constrain the target light source group and are not directly written into the base color texture of the target surface.

[0053] Step S103: Bind multiple material reflective carriers to the target surface domain. For the m-th surface domain, set the style weight corresponding to the k-th reference style image. ,in ≥0 and =1. Style weights can be specified by the designer, or an initial value can be generated based on the matching degree between the target surface semantics and the reference material semantics, and then confirmed by the designer.

[0054] Through mapping operators Map the reference material block to the target surface region. Mapping operator. This can include the main texture direction, world scale, repetition period, boundary clipping rules, and coordinate transformation. For wood grain or brushed metal, the main texture direction of the reference material block is aligned with the preset tangent direction of the target surface; for cases where the same continuous surface is split into multiple UV islands, splicing continuity constraints are established based on the corresponding points of the UV boundaries; for different objects sharing the same material group, texture scale constraints are established.

[0055] In one implementation, the style material base color of the m-th surface domain In the logarithmic reflectance domain, it is formed according to the following formula:

[0056]

[0057] In the formula, u is the surface coordinate. The base color of the original material of the target surface. For reference material area, the exposure-independent center value, The style center color is set to the target surface. Roughness, normal intensity, metallicity, and specular reflection parameters can be fused separately for each surface domain and limited to their respective physical value ranges. Style weights can also be spatial variation weight maps defined in the UV coordinate domain or the world coordinate material domain, ensuring that all output views read the same set of surface material parameters.

[0058] Step S104: Solve for shared lighting parameters based on the surface-light source influence relationship and determine light source conflict information. First, based on the style weights of each surface domain, calculate the lighting descriptor... Transform into the desired illumination descriptor of the m-th surface domain For multiple surface regions simultaneously affected by the same window light or luminaire, instead of setting independent lights for each surface region, a unique shared lighting parameter λ is obtained using the surface-light source influence matrix A. The shared lighting parameter may include at least one of the following: intensity coefficient, RGB color coefficient, color temperature, and luminous size of each light source group.

[0059] The shared light parameter λ can be obtained by the following bounded regularization:

[0060]

[0061] In the formula, For the importance of the surface domain, The original lighting parameters for the target indoor 3D scene. To limit excessive offset of lighting parameters, a regularization coefficient is used. and These are the lower and upper bounds of the parameters, respectively, set according to the achievable power, color temperature range, or design specifications of the luminaire. Since the same light source group corresponds to only one set of parameters in the vector λ, it will not be assigned independent colors or intensities when applied to walls, floors, and furniture.

[0062] To identify illumination discrepancies between images with different reference styles, the directed adjustment requirements of the m-th surface domain to the g-th light source group are determined. Targeted adjustment requirements It can be obtained by projecting the difference between the desired illumination descriptor and the currently predicted illumination descriptor of the surface domain along the light source response direction. Light source conflict coefficient. It can be determined according to the following formula:

[0063]

[0064] When all surface regions require enhancement or reduction of the same light source group The directions are basically the same. Approaching 0; when some surface regions require enhancement while others require reduction, the directional adjustment requirements cancel each other out. Increase. The conflict threshold can be set from 0.45 to 0.70. When When the conflict threshold is exceeded, the allowed update range of the shared lighting parameters for the corresponding light source group is narrowed, while the participation weight of the corresponding lighting carrier is reduced. For the remaining differences that cannot be attributed to the shared lighting parameters due to the conflict, compensation is only allowed through material chromaticity, roughness, or specular reflection parameters that do not contain local light and dark contours.

[0065] The light source conflict information may also include the angle between adjustment vectors, the ratio of positive and negative adjustment requirements, or the weighted variance; as long as the information can characterize whether the adjustment directions of multiple surface domains to the same full-field light source group are consistent, and generate update constraints for shared lighting parameters and weight constraints for the lighting carrier accordingly, it can be used in this embodiment.

[0066] Step S105: Perform material domain-lighting domain attribution on the initial rendering residuals. For example... Figure 3 As shown, using the surface material parameters formed in step S103 and the shared lighting parameters obtained in step S104, preliminary rendering is performed on the diagnostic viewpoint. For pixel p in the diagnostic viewpoint v, the rendering result can be expressed as:

[0067]

[0068] In the formula, T represents the uniform tone map. This is a specular reflection term recalculated based on the target surface's roughness, metallicity, normals, viewing direction, and light source parameters. Material reflectivity and light source contribution are combined during 3D rendering, and viewpoint-dependent specular highlights are not copied from the reference style image to the fixed texture.

[0069] The initial rendering result is compared with the style target constructed based on the material reflection carrier and lighting carrier to obtain the rendering residual e. A small perturbation rendering is performed on at least one adjustable surface material parameter among base color, roughness, normal intensity, and metallicity, or the shader analytical derivative is calculated to form the material response matrix. A light response matrix is ​​formed based on the light source response data and its color channels. .

[0070] Material domain-lighting domain attribution can be achieved by solving the following equation:

[0071]

[0072] In the formula, For material increment, For the illumination increment, W includes pixel confidence, surface domain importance, and weight constraints of the illumination carrier; E_view is a material continuity constraint across the UV island boundary; E_view is a consistency constraint for the removal of illumination and reflection at the same surface sampling point under multiple diagnostic views. and For the corresponding constraint weights. A larger light source group will The corresponding response quantity multiplied by Alternatively, the response could be frozen, thus excluding conflicting lighting requirements from updates to shared lighting parameters.

[0073] Attribution Only the base color, roughness, normal intensity, metallicity, or specular reflection parameters of the corresponding surface domain are updated, and the UV material map, world coordinate material field, or material parameter table is written back; the attributed Δλ only updates the unique shared light parameters of the corresponding light source group in the target indoor 3D scene. After completing a parameter update, the scene is re-rendered and the rendering residual is calculated to form a closed-loop iteration.

[0074] The iteration can begin by fixing the shared lighting parameters and updating the surface material parameters for two to three rounds. Then, it unlocks light source groups with conflict coefficients below the conflict threshold and updates them once. Finally, it performs joint fine-tuning for one to two rounds. Iteration stops when any of the following conditions are met: the decrease in rendering residuals is less than the first threshold for two consecutive rendering iterations; the maximum update amount of both surface material parameters and shared lighting parameters is less than the second threshold; or the preset number of iterations is reached. The specific number of rounds and thresholds can be adjusted according to the scene complexity.

[0075] Step S106: Perform unified rendering based on the updated surface material parameters and shared lighting parameters. After the iteration, save the style material parameter set M for each surface domain. Shared lighting parameter set L for each light source group and tone mapping parameters. All renderings, panoramas, or walkthrough frames are derived from the same target indoor 3D scene G(M). ,L The same set of surface material parameters is read for the same surface when the camera view is changed; when the light intensity, color, or position is changed, shadows and highlights are recalculated by the renderer.

[0076] The output can include a target viewpoint rendered image, an editable 3D scene file, style textures, a shared lighting parameter table, and conflict warnings. Conflict warnings can identify conflicting light groups, surface regions requesting opposite adjustments, and their corresponding style weights, allowing designers to adjust style weights or retain the original lighting settings.

[0077] Example 2:

[0078] Taking a living and dining room measuring 6.2 meters long and 4.8 meters wide as an example, the target interior 3D scene was divided into 27 surface domains, including walls, floors, ceilings, dining tables, cabinets, sofas, and curtains. Window lighting, chandeliers, LED strips, and floor lamps were grouped into four light source groups, and 10 diagnostic viewpoints were set. Reference style image 1 was used for the mineral paint style of the walls, reference style image 2 for the walnut wood style of the dining table and cabinets, and reference style image 3 for the sofa fabric style; the three reference style images were respectively derived from cool-toned daylight, warm-toned chandeliers, and a high-exposure showroom environment.

[0079] Exposure normalization and intrinsic component decomposition are performed on the three reference style images. In reference style image 1, the window frame projection on the wall enters the lighting carrier but not the material reflection carrier; in reference style image 2, the striped highlight on the dining table enters the viewpoint-dependent highlight carrier, and its width and peak value are used to constrain the roughness of the target wood; the fine texture of the fabric in reference style image 3 is preserved in the material reflection carrier. Therefore, the generated wall base color does not include the window frame projection, and the wood grain base color does not include the fixed striped highlight.

[0080] The unit light source response maps of the four light sources are rendered sequentially, forming a 27×4 surface-light source influence matrix. The chandelier contributes significantly to the color temperature and intensity of the dining table, adjacent walls, and a local area of ​​the floor. If reference style image 1 requires the adjacent walls to remain cool and have low brightness, while reference style image 2 requires the dining table to remain warm and have high brightness, then the two surface domains have opposite directional adjustment requirements for the chandelier's color temperature and intensity.

[0081] When the light source conflict coefficient of the chandelier group is 0.68 and exceeds the conflict threshold of 0.55, the color temperature of the chandelier is limited to the original setting range of the target indoor 3D scene, and only slight adjustments to the intensity are allowed. At the same time, the participation weight of the lighting carriers in the two conflicting surface areas is reduced. The remaining tonal differences between the wall and the dining table are compensated by the base color chromaticity without local shadow shapes.

[0082] In the material domain-lighting domain attribution iteration, wood grain direction, wall particle scale, and fabric texture are written back to their respective surface domains, while the chandelier group always retains only one set of shared lighting parameters. After the iteration is complete, rendering is performed from new perspectives of the entrance, sofa area, and dining area, with all perspectives reading the same set of material and lighting parameters. When the chandelier intensity is adjusted from 100% to 70%, the highlights on the table and the lighting on the wall change together with the contribution of the chandelier, while the original highlights and window shadows in the reference style image do not remain as fixed patterns.

[0083] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for multi-style image transfer and rendering in interior scene design, characterized in that, Includes the following steps: Step S1: Divide the target indoor 3D scene, which includes geometric model, surface material parameters, camera parameters and adjustable light source parameters, into multiple surface domains and multiple light source groups, select multiple diagnostic viewpoints, and obtain surface identification information and light source response data of each light source group when participating in rendering according to preset benchmark parameters under the diagnostic viewpoints; Step S2: Perform exposure normalization and intrinsic component decomposition on at least two reference style images to obtain the material reflection carrier and illumination carrier of each reference style image. Map the material reflection carrier to the corresponding surface domain according to the surface correspondence and style weight to obtain the surface material parameters shared by multiple diagnostic viewpoints. Based on the surface identification information, the light source response data is aggregated into surface domains to establish the surface-light source influence relationship. The shared lighting parameters of each light source group are solved according to the lighting carrier corresponding to each surface domain and the surface-light source influence relationship, so that the same light source group corresponds to the same set of shared lighting parameters in different surface domains. Step S3: Determine the directional adjustment requirements of multiple surface domains for the same light source group, obtain light source conflict information based on the directional consistency among the directional adjustment requirements, and generate update constraints on the corresponding shared light parameters and weight constraints on the corresponding lighting carriers based on the light source conflict information; perform preliminary rendering of the target indoor 3D scene using the surface material parameters and the shared light parameters, and obtain rendering residuals based on the preliminary rendering results and the style target composed of the material reflection carriers and lighting carriers; Step S4: Determine the material response relationship corresponding to the adjustable surface material parameters and the lighting response relationship corresponding to each light source group. Under the update constraint and weight constraint, perform material domain-lighting domain attribution on the rendering residual. Write the attributed material increment into the corresponding surface domain and the attributed lighting increment into the shared light parameters of the corresponding light source group. Render an indoor scene image from at least one target viewpoint based on the updated surface material parameters and shared light parameters.

2. The method for multi-style image transfer and rendering in indoor scene design according to claim 1, characterized in that, The acquisition of surface identification information and light source response data from the diagnostic perspective includes: outputting a surface identification map, UV coordinate map, normal map and depth map for each diagnostic perspective; keeping the camera parameters, geometric model and surface material parameters of the target indoor 3D scene unchanged, setting each light source group to unit intensity or preset response level in sequence, and turning off the other adjustable light source groups to obtain the corresponding unit light source response map or segmented light source response map. The set of pixels for each surface domain in the unit light source response map or the segmented light source response map is determined according to the surface identification map. The set of pixels is then weighted and aggregated according to pixel visibility, viewing angle importance and response brightness to form a surface-light source influence matrix that characterizes the actual contribution of each light source group to each surface domain.

3. The method for multi-style image transfer and rendering in indoor scene design according to claim 1, characterized in that, The exposure normalization and intrinsic component decomposition of the reference style image includes: converting the reference style image to a linear color space and forming a specular mask based on at least two of the following: high brightness, saturation, local contrast and normal-viewing line relationship; Within a smooth range defined by the material region boundary and surface normal, the reflection term and illumination term are jointly estimated to obtain the material reflection carrier, diffuse illumination carrier, and viewpoint-dependent specular carrier. At least one of the specular width, peak value, directionality, and color is extracted from the viewpoint-dependent specular carrier as a constraint on the target surface roughness, metallicity, or specular reflection parameters, and the viewpoint-dependent specular carrier and local projected shadows are prevented from being written into the base color texture of the target surface.

4. The method for multi-style image transfer and rendering in indoor scene design according to claim 1, characterized in that, The process of mapping the material reflection carrier to the corresponding surface domain includes: setting multiple non-negative style weights corresponding to reference style images for each surface domain, and making the sum of the non-negative style weights equal to 1; establishing a mapping to the target surface domain according to the texture principal direction, world scale, repetition period, and boundary clipping rules of the reference material block; fusing the mapped material reflection carrier in the UV coordinate domain or world coordinate material domain, setting splicing continuity constraints for multiple UV islands on the same continuous surface, and setting texture scale constraints for different objects sharing the same material group.

5. The method for multi-style image transfer and rendering in indoor scene design according to claim 1, characterized in that, The process of solving for the shared lighting parameters of each light source group includes: extracting lighting descriptors from each lighting carrier that contain at least one of normalized luminance, chromaticity, primary light-secondary light ratio, and spatial luminance gradient; and obtaining the desired lighting descriptors for each surface domain based on surface correspondence and style weights. Within the physical range of the shared lighting parameters, the weighted difference between the predicted lighting descriptor determined by the surface-light source influence relationship and the desired lighting descriptor, as well as the offset of the shared lighting parameters relative to the original lighting parameters of the target indoor 3D scene, is minimized to obtain shared lighting parameters that include at least one of the light source intensity coefficient and color coefficient.

6. The method for multi-style image transfer and rendering in indoor scene design according to claim 1, characterized in that, Let the directional adjustment requirement of the m-th surface region for the g-th light source group be: The importance of the m-th surface domain is The light source conflict information includes information according to... Determined light source conflict coefficient Where ε is a positive number; when the light source conflict coefficient is greater than the conflict threshold, the allowable update interval of the shared lighting parameters of the corresponding light source group is reduced, the participation weight of the corresponding lighting carrier is decreased, and the response quantity corresponding to the corresponding light source group in the illumination response relationship is multiplied by ε. Or freeze the response volume.

7. The method for multi-style image transfer and rendering in interior scene design according to claim 1, characterized in that, The material response relationship is obtained by performing perturbation rendering or calculating the shader analytical derivative on at least one adjustable surface material parameter among base color, roughness, normal intensity, and metallicity. The lighting response relationship is obtained based on the light source response data and its color channels. The material domain-lighting domain attribution is also subject to the consistency constraints of de-illumination reflection at the same surface sampling point under multiple diagnostic viewpoints and the material continuity constraints across UV island boundaries. It iterates in the following order: first, fix the shared light parameters and update the surface material parameters; then, update the light source groups with conflicts below the threshold; and finally, jointly fine-tune until the rendering residual decrease, parameter update amount, or number of iterations meets the preset stopping conditions.

8. A multi-style image transfer and rendering system for interior scene design, applied to the multi-style image transfer and rendering method for interior scene design as described in any one of claims 1-7, characterized in that, include: The scene response acquisition module is used to divide the target indoor 3D scene into multiple surface domains and multiple light source groups, select multiple diagnostic viewpoints, and acquire surface identification information and light source response data of each light source group when participating in rendering according to preset benchmark parameters. The style carrier separation module is used to perform exposure normalization and intrinsic component decomposition on at least two reference style images to obtain material reflection carrier and illumination carrier; The surface style binding module is used to map the material reflection carrier to the corresponding surface domain according to the surface correspondence and style weight, so as to obtain the surface material parameters shared by multiple diagnostic perspectives; The shared illumination solution module is used to aggregate the light source response data into surface domains based on surface identification information to establish the surface-light source influence relationship. It solves the shared lighting parameters of each light source group based on the lighting carrier corresponding to each surface domain, and obtains the light source conflict information based on the directional consistency between the directional adjustment requirements of multiple surface domains for the same light source group, so as to generate update constraints for shared lighting parameters and weight constraints for lighting carriers. The dual-domain iteration module is used to obtain the rendering residual based on the initial rendering results. Based on the material response relationship and the lighting response relationship, the rendering residual is attributed to the material domain and the lighting domain under the update constraint and weight constraint. The material increment is written to the corresponding surface domain, and the lighting increment is written to the shared light parameters of the corresponding light source group. And a unified rendering module for rendering indoor scene images from at least one target viewpoint based on updated surface material parameters and shared lighting parameters.

Images