Picture rendering method and device of building module, equipment and storage medium
By extracting style feature vectors from building modules and training multi-view models, the problem of poor rendering of building modules was solved, achieving efficient and unified style transfer rendering effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-08
- Publication Date
- 2026-03-17
AI Technical Summary
In existing technologies, the rendering effect is poor when performing style transfer rendering for building modules.
By acquiring the images to be rendered and the target rendering style information, a pre-defined feature extraction network is used to extract style feature vectors from the set of effect images. A multi-view rendering model is constructed and a multi-stage consistency model is trained. The trained rendering model is then used for rendering, and the consistency of multiple views under the same rendering style is maintained during rendering.
It improves rendering quality, ensures consistency in style and lighting across multiple images, reduces redundant reasoning, increases rendering speed, and achieves consistency across multiple images in batch rendering.
Smart Images
Figure CN121685802A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of architectural rendering, and more particularly to a method, apparatus, device, and storage medium for rendering images of architectural modules. Background Technology
[0002] Currently, there are two methods for rendering architectural modules: manual rendering and rendering using artificial intelligence (AI) technology. Manual rendering is inefficient and relies heavily on the experience of technical personnel. While AI-powered rendering is highly efficient and automated, automatically generating architectural scenes, the rendering quality is poor, often deviating significantly from the actual scene, especially when style transfer is required. Therefore, existing technologies suffer from poor rendering results when performing style transfer on architectural modules.
[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention
[0004] The main objective of this invention is to provide an image rendering method for building modules, aiming to solve the technical problem of poor rendering effect when rendering building modules for style transfer in the prior art.
[0005] To achieve the above objectives, this application provides a method for rendering images of a building module, the method comprising: Obtain the image to be rendered, as well as the target rendering style information; Based on the preset feature extraction network and the target rendering style information, style feature vectors under the target rendering style are extracted from the preset set of effect images. Based on the style feature vector, multiple rendering models with different perspectives under the target rendering style are constructed, and multi-stage consistency model training is performed to obtain multiple trained rendering models. The perspectives include perspective, bird's-eye view and elevation view. During multi-stage consistency model training, the style consistency of multiple perspectives under the same rendering style is maintained. After selecting a viewpoint as the target viewpoint, the image to be rendered is rendered based on the rendering control parameters randomly generated from the target viewpoint and the trained rendering model to obtain the rendered image. The same rendering control parameters are used for multiple images during rendering.
[0006] In one possible implementation of this application, after selecting a viewpoint as the target viewpoint, rendering the image to be rendered based on rendering control parameters randomly generated from the target viewpoint and the trained rendering model, and obtaining the rendered image, the process includes: Extract key parameters from the image to be rendered to obtain the first key parameter; Extract key parameters from the rendered image to obtain the second key parameter; The consistency between the first key parameter and the second key parameter in room size, component position and modular component information is compared to obtain a matching score. The room size is verified by size ratio, the component position is detected by offset, and the edge contour similarity of the modular components is checked. If the matching score is lower than a preset threshold, the process of selecting a viewpoint as the target viewpoint, rendering the image to be rendered based on the rendering control parameters randomly generated by the target viewpoint and the trained rendering model is returned until the matching score is not lower than the preset threshold.
[0007] In one possible implementation of this application, the step of extracting style feature vectors under the target rendering style from a preset set of effect images based on a preset feature extraction network and the target rendering style information includes: Based on a pre-defined convolutional neural network and target rendering style information, content information and style information are determined from effect images in a pre-defined set of effect images. Style feature maps are selected from the content information and style information, and the style feature maps include channel dimension features and size features; The size feature of the style feature map is converted into a feature map vector, resulting in multiple feature map vectors with the same number of channels as the channel dimension feature; Calculate the inner product among multiple feature map vectors, and determine the style feature vector under the target rendering style based on the inner product.
[0008] In one possible implementation of this application, the step of constructing multiple rendering models from different perspectives under the target rendering style based on the style feature vector, and performing multi-stage consistency model training to obtain multiple trained rendering models includes: Based on the style feature vectors, construct multiple rendering models from different perspectives under the target rendering style; A noisy image is randomly generated based on the rendering model; The mean square error between the style feature vector and the style feature vector of the noisy image with respect to spatial geometry is calculated to obtain the first style loss value; The rendering model is trained based on the first style loss value to obtain a first rendering model that maintains the spatial geometry unchanged. The mean square error of texture between the style feature vector and the style feature vector of the noisy image is calculated to obtain the second style loss value; The first rendering model is trained based on the second style loss value to obtain a second rendering model that remains unchanged during texture style transfer. The texture style includes material, lighting and hue. The mean square error between the style feature vector and the style feature vector of the noisy image with respect to the sharpness of the constructed nodes is calculated to obtain the third style loss value, wherein the constructed nodes include window frames, baseboards and smart switch panels. The second rendering model is trained based on the third style loss value to obtain the rendering model.
[0009] In one possible implementation of this application, after selecting a viewpoint as the target viewpoint, rendering the image to be rendered based on rendering control parameters randomly generated from the target viewpoint and a trained rendering model to obtain the rendered image includes: After selecting a viewpoint as the target viewpoint, the image to be rendered is rendered based on the rendering control parameters randomly generated from the target viewpoint and the trained rendering model. During rendering, if multiple images to be rendered contain common image content, the common image content shares content features to ensure the consistency of the rendering effect of multiple images.
[0010] In one possible implementation of this application, during the training of a multi-stage consistency model, the perspective, bird's-eye view, and facade view are kept consistent in terms of lighting direction and material texture under the same rendering style for the same apartment type or facade.
[0011] In one possible implementation of this application, the set of effect images includes either architectural renderings or actual architectural images, and both architectural renderings and actual architectural images correspond to multiple styles.
[0012] Furthermore, to achieve the above objectives, this application also provides an image rendering apparatus for a building module, the image rendering apparatus for a building module comprising: The acquisition module is used to acquire the image to be rendered, as well as the target rendering style information; The extraction module is used to extract style feature vectors under the target rendering style from a preset set of effect images based on a preset feature extraction network and the target rendering style information. The training module is used to construct multiple rendering models with different perspectives under the target rendering style based on the style feature vector, and to perform multi-stage consistency model training to obtain multiple trained rendering models. The perspectives include perspective, bird's-eye view and elevation view. During multi-stage consistency model training, the style of multiple perspectives under the same rendering style is kept consistent. The rendering module is used to render the image to be rendered based on the randomly generated rendering control parameters of the target viewpoint and the trained rendering model after selecting a viewpoint as the target viewpoint, so as to obtain the rendered image. The same rendering control parameters are used for multiple images during rendering.
[0013] Furthermore, to achieve the above objectives, this application also provides an image rendering device for a building module. The image rendering device for the building module is a physical node device. The image rendering device for the building module includes: a memory, a processor, and an image rendering program for the building module stored in the memory and executable on the processor. The processor executes the image rendering program for the building module to implement the steps of the image rendering method for the building module.
[0014] In addition, to achieve the above objectives, this application also provides a storage medium storing a program that implements an image rendering method for a building module. When the image rendering program for the building module is executed by a processor, it implements the steps of the image rendering method for the building module described above.
[0015] This application provides a method, apparatus, device, and storage medium for rendering images of building modules. Compared with the existing technology that suffers from poor rendering effects when performing style transfer rendering for building modules, this application obtains the image to be rendered and target rendering style information; based on a preset feature extraction network and the target rendering style information, it extracts style feature vectors under the target rendering style from a preset set of effect images; based on the style feature vectors, it constructs multiple rendering models from different perspectives under the target rendering style and performs multi-stage consistency model training to obtain multiple trained rendering models. The perspectives include perspective, bird's-eye view, and elevation view. During multi-stage consistency model training, the style uniformity of multiple perspectives under the same rendering style is maintained; after selecting a perspective as the target perspective, it renders the image to be rendered based on rendering control parameters randomly generated from the target perspective and the trained rendering model to obtain the rendered image. The same rendering control parameters are used for multiple images during rendering. In this application, multi-perspective rendering models are automatically rendered based on style feature vectors extracted from the effect image set. Multi-stage consistency training across various perspectives maintains style uniformity and improves the rendering effect. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating an embodiment of the image rendering method for the building module of this application; Figure 2 This is a schematic diagram of the image rendering device for the building module in an embodiment of the image rendering method for the building module of this application; Figure 3This is a schematic diagram of the hardware operating environment involved in the image rendering method embodiment of the building module of this application. Detailed Implementation
[0017] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0018] Example 1 This application provides an image rendering method for a building module, which is applied to an image rendering device for a building module.
[0019] In modular architectural design, interior renderings, facade style drawings, and bird's-eye view drawings often require high-quality rendering to aid in presentation. However, traditional rendering methods rely on manual modeling, material processing, lighting and shadow settings, and rendering, which is time-consuming and costly.
[0020] While current AI rendering technology can automatically generate architectural scenes, it suffers from the following problems: the style transfer effect is unstable and it is difficult to guarantee an accurate correspondence with the key components and dimensions of the original scene.
[0021] When rendering architectural modules, two methods are used: manual rendering and rendering using artificial intelligence (AI) technology. Manual rendering is inefficient and relies heavily on the experience of technical personnel. While AI-powered rendering is highly efficient and automated, automatically generating architectural scenes, the rendering quality is poor, often deviating significantly from the actual scene, especially when style transfer is required. Therefore, existing technologies suffer from poor rendering results when performing style transfer rendering on architectural modules.
[0022] like Figure 1 The image rendering method for the building module includes steps S110 to S140: Step S110: Obtain the image to be rendered and the target rendering style information; The image rendering device of the building module acquires images to be rendered from the outside. These images can be original images from Building Information Modeling (BIM). Building module design can be based on the BIM model, transforming the building module from scattered two-dimensional lines into a centralized, information-rich three-dimensional model.
[0023] The target rendering style information includes the target rendering style. There are various target rendering styles, such as line art style, white model style, comic style, watercolor style, low-fidelity technical style, and collage style. For example, the white model style emphasizes a pure white model, highlighting the shape, volume, spatial relationships, and lighting of the building modules. It primarily aims to emphasize the building's volume, spatial composition, sectional relationships, and aesthetic features. Another example is the watercolor style, which mainly aims to emphasize artistic beauty or design inspiration, simulating the characteristics of watercolor painting.
[0024] Step S120: Based on the preset feature extraction network and the target rendering style information, extract the style feature vector under the target rendering style from the preset effect image set; Feature extraction networks are neural network structures that learn from raw images and extract the most useful information for subsequent tasks. They achieve feature extraction from shallow to medium and then to deep networks through operations such as convolution, activation, and pooling.
[0025] The set of effect images includes either architectural renderings or actual architectural images, and each of the architectural renderings or actual architectural images corresponds to multiple styles.
[0026] As an example, collect images of different styles, with no fewer than 500 real architectural renderings or real-life photos for each style. These images can be interior perspective, facade perspective, or bird's-eye view.
[0027] Based on a preset feature extraction network, such as a convolutional neural network, and target rendering style information, style feature vectors under the target rendering style are extracted from a preset set of effect images.
[0028] Step S120 includes steps S1201 to S1204: Step S1201: Based on the preset convolutional neural network and target rendering style information, determine the content information and style information from the effect images in the preset effect image set; The rendered image includes both content and style components. Based on the target rendering style information, a convolutional neural network is used to determine the content and style information from the rendered image.
[0029] Step S1202: Select a style feature map from the content information and style information. The style feature map includes channel dimension features and size features. Based on the convolutional neural network model, the content information and style information are classified, the style information is filtered out, and the style feature map is determined from the style information. The style feature map includes the length and width dimensions, as well as the channel dimension features of multiple channels.
[0030] Step S1203: Convert the size feature of the style feature map into a feature map vector to obtain multiple feature map vectors with the same number of channels as the channel dimension feature. As an example, stretching the size feature yields a feature map vector, and based on this, multiple feature map vectors corresponding to the number of channels are generated.
[0031] Step S1204: Calculate the inner product among the multiple feature map vectors, and determine the style feature vector under the target rendering style based on the inner product.
[0032] Calculate the inner product between feature map vectors, and then convert the inner product into a style feature vector under the target rendering style.
[0033] Step S130: Construct multiple rendering models with different perspectives under the target rendering style based on the style feature vector, and perform multi-stage consistency model training to obtain multiple trained rendering models. The perspectives include perspective, bird's-eye view and elevation view. During multi-stage consistency model training, the style of multiple perspectives under the same rendering style is kept consistent. During multi-stage consistency model training, the perspective, bird's-eye view, and facade view are kept consistent in terms of lighting direction and material texture under the same rendering style for the same apartment type or facade.
[0034] The training process employs a multi-stage approach. The first stage is structure preservation, maintaining the spatial geometry unchanged. The second stage is texture transfer, injecting materials, lighting, and color tones. The third stage is detail enhancement, optimizing the clarity of structural nodes such as window frames, baseboards, and intelligent control switch panels. During multi-stage training, each viewpoint is trained to maintain stylistic consistency across different perspectives while adhering to the same rendering style.
[0035] As an example, the trained rendering model is used to quickly convert screenshots exported from the original BIM into renderings of the target architectural style.
[0036] Step S130 includes steps S1301 to S130: Step S1301: Construct multiple rendering models from different perspectives under the target rendering style based on the style feature vector; The model is constructed based on the target rendering style and multiple perspectives to obtain multiple rendering models under different perspectives.
[0037] Step S1302: Randomly generate a noisy image based on the rendering model; A rendering model from different perspectives randomly generates a noisy image after initial rendering.
[0038] Step S1303: Calculate the mean square error between the style feature vector and the style feature vector of the noisy image with respect to spatial geometry to obtain the first style loss value; The style feature vector of the noisy image is calculated. The style feature vector corresponding to the noisy image is used as the first style feature vector. The style feature vector generated from the effect images in the preset effect image set in this embodiment is used as the second style feature vector. The mean square error in spatial geometry between the first and second style feature vectors is calculated. This mean square error is used as the first style loss value.
[0039] Step S1304: Train the rendering model based on the first style loss value to obtain a first rendering model that keeps the spatial geometry unchanged; The rendering model is iteratively optimized and trained with the goal of minimizing the first style loss value to obtain the first rendering model. Minimizing the first style loss value means that the trained first rendering model will maintain the spatial geometry of the image when rendering it, thus ensuring that the rendered image does not change the spatial structure.
[0040] Step S1305: Calculate the mean square error of texture between the style feature vector and the style feature vector of the noisy image to obtain the second style loss value; Calculate the mean square error in texture between the first style feature vector and the second style feature vector, and use this mean square error as the second style loss value.
[0041] Step S1306: Train the first rendering model based on the second style loss value to obtain a second rendering model to keep the texture style unchanged during texture style transfer, wherein the texture style includes material, lighting and hue; The first rendering model is iteratively optimized and trained to minimize the second style loss value, resulting in the second rendering model. Minimizing the second style loss value means that the trained second rendering model will maintain the texture style when rendering images, ensuring that the rendered images do not change the texture style. Under the same rendering style, the material, lighting, and hue of the original image are not changed.
[0042] Step S1307: Calculate the mean square error between the style feature vector and the style feature vector of the noise image with respect to the sharpness of the constructed nodes to obtain the third style loss value, wherein the constructed nodes include window frames, baseboards and smart switch panels. Calculate the mean squared error between the first style feature vector and the second style feature vector in terms of the sharpness of the constructed nodes, and use this mean squared error as the third style loss value.
[0043] Step S1308: Train the second rendering model based on the third style loss value to obtain the rendering model.
[0044] The second rendering model is iteratively optimized and trained to minimize the third style loss value, resulting in the third rendering model. Minimizing the third style loss value means that the trained third rendering model will maintain the clarity of the constructed nodes when rendering images, ensuring that the clarity of the constructed nodes is not changed when rendering images.
[0045] Step S140: After selecting a viewpoint as the target viewpoint, the image to be rendered is rendered based on the rendering control parameters randomly generated by the target viewpoint and the trained rendering model to obtain the rendered image. In this process, the same rendering control parameters are used for multiple images.
[0046] After selecting a viewpoint as the target viewpoint, the image to be rendered is rendered based on the rendering control parameters randomly generated from the target viewpoint and the trained rendering model. During rendering, if multiple images to be rendered share common image content, the common image content shares content features to ensure consistency in the rendering effect of multiple images. Rendered images share some spatial features, reducing batch-to-batch bias. Through scheduling optimization, the model generates multiple images at once, reducing redundant inference and improving rendering speed by approximately 40% or more. This embodiment ensures the synergy of style and lighting in multi-angle floor plan renderings and multi-scheme elevation renderings. Synchronous rendering enables batch rendering and multi-image rendering. Figure 1 Consistency optimization.
[0047] Following step S140, the sequence includes steps S150 through S180: Step S150: Extract key parameters from the image to be rendered to obtain the first key parameter; The key parameters extracted before rendering the image are used as the first key parameter, and the key parameters extracted after rendering the image are used as the second key parameter. Key parameters, such as the dimensions of a room (length, width, and height), and other key parameters, such as the location of doors and windows, and modular components, are used as examples.
[0048] Step S160: Extract key parameters from the rendered image to obtain the second key parameters; As an example, the image to be rendered can be an image of a building module obtained from the original BIM model.
[0049] Step S170: Compare the consistency of the first key parameter and the second key parameter in room size, component position and modular component information respectively to obtain a matching score. In this step, the room size is verified by size ratio, the component position is detected by offset, and the edge contour similarity of the modular components is checked. The matching between the first key parameter and the second key parameter is compared. The key parameters to be compared include room size, construction location and modular components. The matching score is obtained by comparing whether these parameters are consistent and the matching degree score is output. If the score is lower than the threshold, the rendering is automatically rolled back or a prompt is made to regenerate.
[0050] Specifically, when matching room dimensions, the room dimensions are verified for size ratio; when matching component positions, the component positions are detected for offset; and when matching modular component information, the similarity of the edge contours of the modular components is checked.
[0051] Step S180: If the matching score is lower than a preset threshold, return to the step of rendering the image to be rendered based on the rendering control parameters randomly generated by the target viewpoint and the trained rendering model after selecting a viewpoint as the target viewpoint, until the matching score is not lower than the preset threshold.
[0052] Output a matching score. If the matching score is lower than a preset threshold, return to re-render the image to be rendered. Select a new viewpoint from multiple perspectives as the target viewpoint, such as a bird's-eye view. Randomly generate rendering control parameters for the bird's-eye view. Based on the regenerated rendering control parameters and the trained rendering model, render the image to be rendered to obtain the rendered image and the matching score. If the matching score is still lower than the preset threshold, continue iterating until the matching score is not lower than the preset threshold.
[0053] This application provides a method, apparatus, device, and storage medium for rendering images of building modules. Compared with the existing technology that suffers from poor rendering effects when performing style transfer rendering for building modules, this application obtains the image to be rendered and target rendering style information; based on a preset feature extraction network and the target rendering style information, it extracts style feature vectors under the target rendering style from a preset set of effect images; based on the style feature vectors, it constructs multiple rendering models from different perspectives under the target rendering style and performs multi-stage consistency model training to obtain multiple trained rendering models. The perspectives include perspective, bird's-eye view, and elevation view. During multi-stage consistency model training, the style uniformity of multiple perspectives under the same rendering style is maintained; after selecting a perspective as the target perspective, it renders the image to be rendered based on rendering control parameters randomly generated from the target perspective and the trained rendering model to obtain the rendered image. The same rendering control parameters are used for multiple images during rendering. In this application, multi-perspective rendering models are automatically rendered based on style feature vectors extracted from the effect image set. Multi-stage consistency training across various perspectives maintains style uniformity and improves the rendering effect.
[0054] Example 2 Furthermore, based on all the above embodiments, another embodiment of this application is provided, in which the method is applied to an image rendering device for a building module, such as... Figure 2 A device for rendering images of building modules is provided, the device comprising: The acquisition module is used to acquire the image to be rendered, as well as the target rendering style information; The extraction module is used to extract style feature vectors under the target rendering style from a preset set of effect images based on a preset feature extraction network and the target rendering style information. The training module is used to construct multiple rendering models with different perspectives under the target rendering style based on the style feature vector, and to perform multi-stage consistency model training to obtain multiple trained rendering models. The perspectives include perspective, bird's-eye view and elevation view. During multi-stage consistency model training, the style of multiple perspectives under the same rendering style is kept consistent. The rendering module is used to render the image to be rendered based on the randomly generated rendering control parameters of the target viewpoint and the trained rendering model after selecting a viewpoint as the target viewpoint, so as to obtain the rendered image. The same rendering control parameters are used for multiple images during rendering.
[0055] In one possible embodiment of this application, the apparatus further includes a verification module. After selecting a viewpoint as the target viewpoint, and rendering the image to be rendered based on rendering control parameters randomly generated from the target viewpoint and the trained rendering model, to obtain the rendered image, the verification module is specifically used for: Extract key parameters from the image to be rendered to obtain the first key parameter; Extract key parameters from the rendered image to obtain the second key parameter; The consistency between the first key parameter and the second key parameter in room size, component position and modular component information is compared to obtain a matching score. The room size is verified by size ratio, the component position is detected by offset, and the edge contour similarity of the modular components is checked. If the matching score is lower than a preset threshold, the process of selecting a viewpoint as the target viewpoint, rendering the image to be rendered based on the rendering control parameters randomly generated by the target viewpoint and the trained rendering model is returned until the matching score is not lower than the preset threshold.
[0056] In one possible implementation of this application, the extraction module, in the step of extracting style feature vectors under the target rendering style from a preset set of effect images based on a preset feature extraction network and the target rendering style information, is specifically used for: Based on a pre-defined convolutional neural network and target rendering style information, content information and style information are determined from effect images in a pre-defined set of effect images. Style feature maps are selected from the content information and style information, and the style feature maps include channel dimension features and size features; The size feature of the style feature map is converted into a feature map vector, resulting in multiple feature map vectors with the same number of channels as the channel dimension feature; Calculate the inner product among multiple feature map vectors, and determine the style feature vector under the target rendering style based on the inner product.
[0057] In one possible implementation of this application, the training module constructs multiple rendering models from different perspectives under the target rendering style based on the style feature vectors, and performs multi-stage consistency model training to obtain multiple trained rendering models. Specifically, this step is used for: Based on the style feature vectors, construct multiple rendering models from different perspectives under the target rendering style; A noisy image is randomly generated based on the rendering model; The mean square error between the style feature vector and the style feature vector of the noisy image with respect to spatial geometry is calculated to obtain the first style loss value; The rendering model is trained based on the first style loss value to obtain a first rendering model that maintains the spatial geometry unchanged. The mean square error of texture between the style feature vector and the style feature vector of the noisy image is calculated to obtain the second style loss value; The first rendering model is trained based on the second style loss value to obtain a second rendering model that remains unchanged during texture style transfer. The texture style includes material, lighting and hue. The mean square error between the style feature vector and the style feature vector of the noisy image with respect to the sharpness of the constructed nodes is calculated to obtain the third style loss value, wherein the constructed nodes include window frames, baseboards and smart switch panels. The second rendering model is trained based on the third style loss value to obtain the rendering model.
[0058] In one possible implementation of this application, the step of rendering the image to be rendered based on rendering control parameters randomly generated from the target viewpoint and a trained rendering model after selecting a viewpoint as the target viewpoint, to obtain the rendered image, is specifically used for: After selecting a viewpoint as the target viewpoint, the image to be rendered is rendered based on the rendering control parameters randomly generated from the target viewpoint and the trained rendering model. During rendering, if multiple images to be rendered contain common image content, the common image content shares content features to ensure the consistency of the rendering effect of multiple images.
[0059] The specific implementation of the image rendering device for the building module in this application is basically the same as the various embodiments of the image rendering method for the building module described above, and will not be repeated here.
[0060] Example 3 Furthermore, based on all the above embodiments, another embodiment of this application is provided. In this embodiment, an image rendering device for a building module is provided. The image rendering device for the building module is a physical node device. The image rendering device for the building module includes: a memory, a processor, and a program stored in the memory for implementing the image rendering method of the building module. The memory is used to store the program for implementing the image rendering method of the building module; the processor is used to execute the program for implementing the image rendering method of the building module to implement the steps of the image rendering method of the building module in the above embodiments.
[0061] Reference Figure 3 , Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.
[0062] like Figure 3 As shown, the image rendering device for this building module may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to establish communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM or a stable, non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0063] In one possible implementation of this application, the image rendering device of the building module may further include a network interface, audio circuit, display, connecting cable, sensor, input module, etc. The network interface may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface or a Bluetooth interface), and the input module may optionally include a keyboard, a system soft keyboard, voice input, wireless receiver input, etc.
[0064] Those skilled in the art will understand that the structure of the image rendering device for the building module does not constitute a limitation on the image rendering device for the building module, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.
[0065] The memory, as a defined storage medium, may include an operating system, an information exchange module, and an image rendering program for the building module. The operating system is a program that manages and controls the hardware and software resources of the building module's image rendering device, supporting the operation of the building module's image rendering program and other software and / or programs. The information exchange module is used to enable communication between the various components within the memory, as well as communication with other hardware and software in the management system.
[0066] In the image rendering device for the building module, the processor is used to execute the image rendering program for the building module stored in the memory, thereby implementing the above-mentioned image rendering steps for the building module.
[0067] The specific implementation of the image rendering device for the building module in this application is basically the same as the various embodiments of the image rendering method for the building module described above, and will not be repeated here.
[0068] Example 4 This application provides a storage medium that stores one or more programs, which can be executed by one or more processors to implement the steps of the image rendering method for the building module in the above embodiments.
[0069] The specific implementation of the storage medium in this application is basically the same as the various embodiments of the image rendering method of the building module described above, and will not be repeated here.
[0070] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0071] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0072] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This device software product is stored in a storage medium (such as ROM or RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, a device, a server, an air conditioner, or a network device, etc.) to execute the methods described in the various embodiments of this application.
[0073] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A method of picture rendering of a building module, characterized in that, The picture rendering method of the building module comprises: acquiring a picture to be rendered and target rendering style information; extracting a style feature vector under a target rendering style from a preset effect picture set based on a preset feature extraction network and the target rendering style information; constructing a plurality of rendering models under different viewing angles of the target rendering style based on the style feature vector, and performing multi-stage consistency model training to obtain a plurality of trained rendering models, wherein the viewing angles include a perspective viewing angle, an aerial viewing angle, and an elevation viewing angle, and the style uniformity of multiple viewing angles under the same rendering style is maintained during the multi-stage consistency model training; after selecting one viewing angle as a target viewing angle, rendering the picture to be rendered based on a rendering control parameter randomly generated for the target viewing angle and the trained rendering model to obtain a rendered picture, wherein the same rendering control parameter is used for rendering multiple pictures.
2. The picture rendering method of an architectural module according to claim 1, characterized in that, After the step of selecting one viewing angle as a target viewing angle and rendering the picture to be rendered based on a rendering control parameter randomly generated for the target viewing angle and the trained rendering model to obtain a rendered picture, the method further comprises: extracting key parameters from the picture to be rendered to obtain first key parameters; extracting key parameters from the rendered picture to obtain second key parameters; comparing the consistency of the first key parameters and the second key parameters in room size, component position, and modular part information to obtain a matching score, wherein the room size is subjected to size ratio verification, the component position is subjected to offset detection, and the edge contour similarity of the modular part is verified; if the matching score is lower than a preset threshold, returning to the step of selecting one viewing angle as a target viewing angle and rendering the picture to be rendered based on a rendering control parameter randomly generated for the target viewing angle and the trained rendering model to obtain a rendered picture until the matching score is not lower than the preset threshold.
3. The picture rendering method of an architectural module according to claim 1, wherein, The step of extracting a style feature vector under a target rendering style from a preset effect picture set based on a preset feature extraction network and the target rendering style information comprises: determining content information and style information from an effect picture in the preset effect picture set based on a preset convolutional neural network and the target rendering style information; screening a style feature map from the content information and the style information, the style feature map comprising channel dimension features and size features; converting the size features of the style feature map into a feature map vector to obtain a plurality of feature map vectors with the same number of channels as the number of channel dimension features; calculating the inner product between the plurality of feature map vectors, and determining the style feature vector under the target rendering style based on the inner product.
4. The picture rendering method of an architectural module according to claim 1, wherein, The step of constructing a plurality of rendering models under different viewing angles of the target rendering style based on the style feature vector and performing multi-stage consistency model training to obtain a plurality of trained rendering models comprises: constructing a plurality of rendering models under different viewing angles of the target rendering style based on the style feature vector; randomly generating a noise image based on the rendering model; Calculate the mean square error between the style feature vector and the style feature vector of the noise image about the spatial geometry to obtain a first style loss value; Based on the first style loss value, the rendering model is trained to obtain a first rendering model to keep the spatial geometry unchanged; Calculate the mean square error between the style feature vector and the style feature vector of the noise image about the texture to obtain a second style loss value; Based on the second style loss value, the first rendering model is trained to obtain a second rendering model to keep the texture style unchanged during texture style migration, wherein the texture style includes material, light and shadow, and color tone; Calculate the mean square error between the style feature vector and the style feature vector of the noise image about the construction node clarity to obtain a third style loss value, wherein the construction node includes window frame, skirting line and intelligent switch panel; Based on the third style loss value, the second rendering model is trained to obtain a rendering model.
5. The picture rendering method of an architectural module according to claim 1, wherein, After selecting one view as a target view, the rendering control parameters randomly generated based on the target view and the trained rendering model are used to render the to-be-rendered picture to obtain a rendered picture, including: After selecting one view as a target view, the rendering control parameters randomly generated based on the target view and the trained rendering model are used to render the to-be-rendered picture, wherein during rendering, if multiple to-be-rendered pictures include common picture content, the content features of the common picture content are shared to ensure the consistency of the multi-picture rendering effect.
6. The picture rendering method of an architectural module according to claim 1, wherein, During multi-stage consistent model training, the perspective view, the bird's eye view and the facade view are consistent in the light direction and the material texture under the same rendering style of the same house type or facade.
7. The picture rendering method of an architectural module according to claim 1, wherein, The effect picture set includes any one of building rendering pictures or building real scene pictures, and the building rendering pictures or the building real scene pictures each correspond to multiple styles.
8. An apparatus for rendering pictures of a building module, characterized by The picture rendering device of the building module comprises: An acquisition module is configured to acquire a to-be-rendered picture and target rendering style information; An extraction module is configured to extract a style feature vector under a target rendering style from a preset effect picture set based on a preset feature extraction network and the target rendering style information; A training module is configured to construct multiple rendering models of different views under the target rendering style based on the style feature vector, and perform multi-stage consistent model training to obtain multiple trained rendering models, wherein the views include a perspective view, a bird's eye view and a facade view, and the style is uniform under the same rendering style during multi-stage consistent model training; A rendering module is configured to, after selecting one view as a target view, render the to-be-rendered picture based on rendering control parameters randomly generated based on the target view and the trained rendering model to obtain a rendered picture, wherein the same rendering control parameters are used for multiple pictures during rendering.
9. A picture rendering device of a building module, characterized by A computer program product comprising a memory, a processor and a picture rendering program of a building module stored on the memory and executable on the processor, the processor implementing the steps of the picture rendering method of a building module according to any one of claims 1 to 7.
10. A storage medium, characterized by A storage medium having stored thereon a program implementing a picture rendering method of a building module, the program implementing the picture rendering method of a building module being executable by a processor to implement the steps of the picture rendering method of a building module according to any one of claims 1 to 7.