A method and system for generating an exhibition hall layout

Abstract

A hall layout generation method and system, comprising S100, acquiring a multi-channel condition image of a hall; S200, inputting the multi-channel condition image into a hall generation condition diffusion model to perform feature coding and fusion with a noise image, and generating a hall layout image through denoising processing; S300, extracting spatial structure information based on the hall layout image and encoding the spatial structure information into a basic multi-channel mask; S400, inputting an enhanced multi-channel mask corresponding to a current exhibit into a pre-constructed exhibit position prediction network according to the importance sequence of the exhibits, to obtain a position probability distribution map of the current exhibit; S500, determining the arrangement coordinates of the current exhibit according to the position probability distribution map and updating the multi-channel mask; S600, based on the updated multi-channel mask, repeatedly performing S400 and S500 until the position prediction of all exhibits is completed, to generate a hall layout result; the automation degree and efficiency of the hall layout are improved.

Description

Technical Field

[0001] This invention relates to the field of image generation technology, and in particular to a method and system for generating exhibition hall layouts. Background Technology

[0002] Currently, exhibition space layout design is gradually evolving from traditional manual planning towards digitalization and automation. An exhibition hall typically consists of multiple exhibition units, wall structures, corridors, and art display areas. A well-designed exhibition space layout must not only meet the constraints of the overall external boundaries and functional zoning requirements, but also ensure the continuity and accessibility of the exhibition flow, while simultaneously considering the display logic of the artworks and the viewing experience.

[0003] Existing layout generation methods primarily focus on generating floor plans for residential buildings or general interior spaces. These methods typically take the building's outline, number of rooms, or room adjacency relationships as input, and generate the interior space layout through rule constraints, heuristic algorithms, or deep learning models. However, exhibition hall layouts differ significantly from residential layouts in terms of spatial form and functional requirements: on the one hand, exhibition halls often consist of a large number of relatively homogeneous exhibition units, and their spatial organization emphasizes overall connectivity and circular visitor flow; on the other hand, corridors play a central role in exhibition halls, serving as crucial structural units for organizing visitor flow and connecting different exhibition halls.

[0004] Existing residential layout generation methods often rely on room function differences and intermediate representations such as bubble diagrams to guide spatial division. When these methods are directly applied to exhibition hall layout generation, they often fail to effectively express the geometric relationships between exhibition halls and are difficult to automatically generate a well-structured and interconnected corridor system. Usually, the corridor structure needs to be manually supplemented or corrected through complex post-processing steps after generation, which reduces the automation level and generation efficiency of exhibition hall layout. Summary of the Invention

[0005] To improve the automation and generation efficiency of exhibition hall layouts, this invention provides an exhibition hall layout generation method, which includes the following steps:

[0006] S100: Acquire multi-channel conditional images of the exhibition hall, including external mask, wall mask, internal mask, entrance mask, and center mask; S200. Input the multi-channel conditional image into the pre-constructed exhibition hall generation conditional diffusion model, encode its features and fuse it with the noisy image, and generate an exhibition hall layout image through denoising processing; S300. Extract spatial structure information based on the exhibition hall layout image and encode it into a basic multi-channel mask; the spatial structure information includes at least wall information, exhibition hall opening location information, and exhibition hall internal space distribution information. S400. Based on the importance order of the exhibits, the enhanced multi-channel mask corresponding to the current exhibit is input into the pre-constructed exhibit position prediction network to obtain the position probability distribution map of the current exhibit; wherein, the enhanced multi-channel mask is generated based on the basic multi-channel mask and the attribute information of the current exhibit. S500: Determine the arrangement coordinates of the current exhibit based on the location probability distribution map, and update the enhanced multi-channel mask; S600: Based on the updated enhanced multi-channel mask, repeat S400 and S500 until the position prediction of all exhibits is completed, generating the exhibition hall layout result.

[0007] Optionally, S100 includes: S110. Construct an initial multi-channel conditional image, which includes the outer mask, the wall mask, the inner mask, the entrance mask, and the initial center mask; wherein, the geometric centroid is calculated based on the contour information of each sub-exhibition hall, and a marked area is generated in the initial center mask with the geometric centroid as the center; S120. Input the current initial multi-channel conditional image into the pre-constructed exhibition hall center prediction network to generate a prediction image of the center position of a single sub-exhibition hall and perform correction processing to update the current initial center mask; the exhibition hall center prediction network adopts an encoder-decoder structure; S130, repeat S120 until the iteration termination condition is met, then use the current initial center mask as the center mask; S140. Merge the central mask, outer mask, wall mask, inner mask, and entrance mask to obtain a multi-channel conditional image of the exhibition hall.

[0008] Optionally, S120 includes: S121. Input the current initial multi-channel conditional image into the pre-constructed exhibition hall center prediction network to generate a prediction image of the center position of a single sub-exhibition hall; S122. Traverse the predicted image through a sliding window of a preset size, and count the number of pixels with a preset value in each sliding window; S123. Select the sliding window corresponding to the maximum number of pixels as the target window; S124. Set the pixel value in the target window to the preset value, take the image area covered by the target window as the corrected center position of the sub-exhibition hall, and update the pixel value of the corresponding position in the current initial center mask.

[0009] Optionally, the exhibition hall generation conditional diffusion model includes a pre-trained exhibition hall attribute encoder and a U-Net structure; the exhibition hall attribute encoder is composed of a multi-layer convolutional neural network; S200 includes: S210. Input the multi-channel conditional image into the exhibition hall attribute encoder, extract its features and map them into a conditional vector; S220. Inject the conditional vector into the denoising network of the U-Net structure, perform progressive denoising processing on the noisy image, and generate the exhibition hall layout image including walls and passages.

[0010] Optionally, the wall information includes at least wall outline information, wall length information, and coordinates of wall inflection points; the exhibition hall opening information includes at least exhibition hall entrance location information, exhibition hall exit location information, and exhibition hall window location information; the exhibition hall interior space distribution information includes at least information on the already arranged exhibits. The basic multi-channel mask includes wall mask, door mask, window mask, interior mask, mask of already laid-out exhibits, and integrated mask of already laid-out exhibits.

[0011] Optional, the S400 includes: S410. Obtain attribute information for all exhibits, wherein the attribute information includes at least importance values, dimensions, classification values, and images of exhibits. S420. Based on the importance values ​​of all exhibits, determine the predicted order of the locations of all exhibits; S430. Based on the predicted position order, for the current exhibit, dynamically enhance the basic multi-channel mask based on its attribute information to generate an enhanced multi-channel mask. S440. Input the enhanced multi-channel mask into the pre-constructed exhibit location prediction network to obtain the current exhibit location probability distribution map.

[0012] Optionally, the S430 includes: S431. Generate a classification constraint mask to characterize classification similarity by combining the classification values ​​of the current exhibit and the classification values ​​of the already laid-out exhibits. S432. Based on the exhibit image of the current exhibit and the exhibit image of the already laid out exhibit, extract the corresponding image feature vectors of the two and calculate the corresponding similarity to generate a similarity mask for representing visual association. S433. The classification constraint mask, similarity mask and basic multi-channel mask are merged to form an enhanced multi-channel mask.

[0013] Optionally, the exhibit location prediction network adopts an encoder-decoder structure; the encoder is used to extract features from the enhanced multi-channel mask, and the decoder is used to output the location probability distribution map, where each pixel location corresponds to the probability distribution of a category label; The category labels include at least the wall area, the area outside the wall, the non-candidate layout area, and the candidate layout area.

[0014] Optional, the S500 includes: S510. Based on the location probability distribution map, perform connected region detection on the candidate arrangement area of ​​the current exhibit; S520. Select the target arrangement area of ​​the current exhibit based on the area of ​​the connected region; S530. Calculate the geometric center point of the target layout area, and calculate the perpendicular distance from the geometric center point to each wall of the exhibition hall; S540. Select the perpendicular point corresponding to the minimum perpendicular distance as the arrangement coordinate of the current exhibit, and update the enhanced multi-channel mask based on the arrangement coordinate.

[0015] Corresponding to the exhibition hall layout generation method, the present invention provides an exhibition hall layout generation system, which includes: The multi-channel conditional image acquisition module is used to acquire and input multi-channel conditional images of the exhibition hall, which includes an external mask, a wall mask, an internal mask, an entrance mask, and a center mask. The exhibition hall layout image generation module is used to input the multi-channel conditional image into a pre-constructed exhibition hall generation conditional diffusion model, encode its features and fuse it with a noisy image, and generate an exhibition hall layout image through denoising processing; The mask encoding module is used to extract spatial structure information based on the exhibition hall layout image and encode it into a basic multi-channel mask; the spatial structure information includes at least wall information, exhibition hall opening location information, and exhibition hall internal space distribution information. The exhibit location prediction module is used to input the enhanced multi-channel mask corresponding to the current exhibit into a pre-constructed exhibit location prediction network according to the importance order of the exhibits, so as to obtain the location probability distribution map of the current exhibit; wherein, the enhanced multi-channel mask is generated based on the basic multi-channel mask and the attribute information of the current exhibit. The exhibit location determination module is used to determine the current exhibit's layout coordinates based on the location probability distribution map and update the enhanced multi-channel mask; The iterative control module is used to repeatedly call the exhibit location prediction module and exhibit location determination module to perform corresponding functions based on the updated enhanced multi-channel mask, until the location prediction of all exhibits is completed and the exhibition hall layout result is generated.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) The external boundary, walls, internal space, entrance and center of the exhibition hall are encoded as spatial constraints by multi-channel conditional images. After being input into the exhibition hall generation conditional diffusion model, the noise image is guided to denoise, so that the corridor and sub-exhibition halls establish a spatial coupling relationship during the generation process, and output a reasonable and well-connected exhibition hall layout in one go, thereby improving the degree of automation. The spatial structure information of the exhibition hall layout is extracted and encoded into a basic multi-channel mask, and the position of exhibits is predicted by combining it with an enhanced multi-channel mask. After determining the current arrangement coordinates of the exhibits, the enhanced multi-channel mask is updated for the next exhibit position prediction. Through the sequential prediction mechanism, the arrangement of exhibits is simultaneously constrained by the inherent geometric structure of the exhibition hall and the importance level of the exhibits. While achieving the unity of spatial feasibility and display logic, it ensures that important exhibits occupy the best space first.

[0017] (2) An initial center mask is generated by S110 based on the geometric centroid of the outline of each sub-exhibition hall, providing a reasonable initial prior for the prediction of the exhibition hall center. Iterative prediction and correction are performed by combining the encoder-decoder structure of the exhibition hall center prediction network, gradually converging to the accurate center position of the sub-exhibition hall, overcoming the possible deviation of a single prediction, so that the final generated center mask accurately reflects the spatial centroid of the sub-exhibition hall, providing reliable spatial guidance for the diffusion model of exhibition hall generation conditions, thereby improving the structural rationality of the generated exhibition hall layout.

[0018] (3) The sliding window correction mechanism of S122-S124 is used to achieve accurate positioning and noise reduction of the center position of the exhibition hall in the predicted image. Specifically, S122 uses a sliding window of preset size to statistically analyze the pixel value distribution, S123 selects the target window, and S124 sets the pixels in the target window to preset values ​​to determine the center position. This transforms the continuous probability map predicted by the neural network into discrete and concentrated center point markers, eliminates the problem of center position ambiguity caused by prediction noise, and ensures that each sub-exhibition hall has clear and unique center coordinates, thereby improving the spatial accuracy of the multi-channel conditional image and providing clear structural anchors for subsequent layout generation.

[0019] (4) Through the collaborative design of the exhibition hall attribute encoder and the U-Net structure, the efficient generation of multi-channel conditional images into exhibition hall layout images is achieved. Specifically, the exhibition hall attribute encoder, composed of multi-layer convolutional neural networks, performs deep feature extraction on the multi-channel conditional images and maps them into conditional vectors. The multi-layer CNN encoder fully captures the spatial hierarchical features (external boundaries, walls, internal spaces, etc.) in the multi-channel mask. After the conditional vectors are injected into the U-Net denoising network, the noisy images are guided to gradually denoise and generate an exhibition hall layout containing walls and passages. The encoder-decoder structure of U-Net effectively integrates the multi-scale information of the conditional vectors and the noisy images, so that the generated exhibition hall layout images strictly follow the spatial constraints of the input, while maintaining the topological coherence of the walls and passages, thereby improving the controllability and structural rationality of the layout generation.

[0020] (5) By unifying the physical structure (walls, openings) and functional status (exhibit layout) of the exhibition hall as a basic multi-channel mask, complete and explicit spatial constraints are provided for the subsequent exhibit location prediction network, so that the exhibit layout takes into account both geometric feasibility (avoiding walls and utilizing openings) and functional coordination (avoiding already laid exhibits), thereby improving the practicality and rationality of the exhibition hall layout results.

[0021] (6) S410 obtains attribute information including importance value, size, classification value and image, S420 determines the position prediction order based on importance value to ensure priority layout of key exhibits, S430 dynamically enhances the basic multi-channel mask based on the attribute information of the current exhibit to make the prediction conditions of each exhibit adapt to its individual characteristics (size, classification, visual features), S440 inputs the position prediction network to obtain the probability distribution map, realizes the priority layout of important exhibits and personalized exhibition layout strategy, and improves the narrative logic and viewing experience of the exhibition.

[0022] (7) By generating a classification constraint mask based on classification value in S431, similar exhibits are guided to form clusters or sequences in space. By generating a similarity mask based on image feature vector similarity in S432, exhibits with similar visual styles are spatially associated. By merging the two with the basic multi-channel mask in S433, an enhanced multi-channel mask is formed. The synergistic effect makes the exhibition layout not only meet geometric feasibility, but also have thematic aggregation and visual fluency, thus enhancing the professionalism and artistic appeal of the exhibition.

[0023] (8) Through the network design of the encoder-decoder structure and the output definition of multi-class labels, the refined prediction of the exhibition layout location is realized. Specifically, the encoder extracts the deep features of the enhanced multi-channel mask image, and the decoder outputs the probability distribution of the class of each pixel location corresponding to the wall area, the area outside the wall, the non-candidate layout area, and the candidate layout area. By transforming the location prediction into a pixel-level classification problem, the candidate layout area and the non-candidate layout area are clearly distinguished, enabling the network to learn the layout distribution of the exhibition space. The output probability distribution map directly reflects the suitability of each location as a layout point, providing an interpretable and quantifiable decision basis for the subsequent determination of specific layout coordinates.

[0024] (9) Connectivity detection is performed on the candidate arrangement area in S510, the target arrangement area is selected based on the area in S520, the geometric center point and its perpendicular distance to each wall are calculated in S530, and the perpendicular point corresponding to the minimum perpendicular distance is selected as the arrangement coordinate and the mask is updated in S540. Based on connectivity detection and area screening, the spatial continuity of the arrangement area can be ensured; combined with the selection strategy of minimizing the perpendicular distance from the geometric center to the wall, exhibits can be arranged closer to the wall (in line with conventional exhibition habits); dynamic mask updates reflect the changes in the layout status in real time, ensuring the accuracy of exhibit arrangement and the coherence of layout iteration. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a simplified flowchart of an embodiment of the exhibition hall layout generation method of the present invention; Figure 2 This is a schematic diagram of a multi-channel conditional image in one embodiment of the exhibition hall layout generation method of the present invention; Figure 3 This is a schematic diagram of the network architecture of the exhibition hall center prediction network in one embodiment of the exhibition hall layout generation method of the present invention; Figure 4 This is a schematic diagram of the structure of the exhibition hall generation condition diffusion model in one embodiment of the exhibition hall layout generation method of the present invention; Figure 5 This is a schematic diagram of the similarity mask generation process in an embodiment of the exhibition hall layout generation method of the present invention; Figure 6 This is a schematic diagram of the network architecture of the exhibit location prediction network in one embodiment of the exhibition hall layout generation method of the present invention; Figure 7 This is a framework diagram of an embodiment of the exhibition hall layout generation system of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] like Figure 1 As shown, a method for generating an exhibition hall layout according to the present invention includes the following steps: S100: Acquire multi-channel conditional images of the exhibition hall, including external mask, wall mask, internal mask, entrance mask, and center mask; S200. Input the multi-channel conditional image into the pre-constructed exhibition hall generation conditional diffusion model, encode its features and fuse it with the noisy image, and generate the exhibition hall layout image through denoising processing; S300: Extract spatial structure information from the exhibition hall layout image and encode it into a basic multi-channel mask; the spatial structure information includes at least wall information, exhibition hall opening location information, and exhibition hall internal space distribution information. S400. Based on the importance order of the exhibits, the enhanced multi-channel mask corresponding to the current exhibit is input into the pre-constructed exhibit position prediction network to obtain the position probability distribution map of the current exhibit; wherein, the enhanced multi-channel mask is generated based on the basic multi-channel mask and the attribute information of the current exhibit. S500: Determine the current exhibit's layout coordinates based on the location probability distribution map, and update the enhanced multi-channel mask; specifically, update the mask of the already laid-out exhibits and the integrated mask of the already laid-out elements. S600: Based on the updated enhanced multi-channel mask, repeat S400 and S500 until the position prediction of all exhibits is completed, generating the exhibition hall layout result.

[0028] In this embodiment, the multi-channel conditional image is as follows: Figure 2 As shown, 2-(a) is the outer mask, 2-(b) is the wall mask, 2-(c) is the inner mask, 2-(d) is the entrance mask, and 2-(e) is the center mask. Each channel has a resolution of 64×64 pixels, and the linewidth of the wall mask and entrance mask is set to 1 pixel.

[0029] This invention encodes the external boundaries, walls, internal spaces, entrances, and center of the exhibition hall into spatial constraints using multi-channel conditional images. These constraints are then input into the exhibition hall generation conditional diffusion model to guide noise image denoising, establishing spatial coupling between corridors and sub-exhibition halls during the generation process. This results in a single output of a structurally sound and well-connected exhibition hall layout, enhancing automation. By extracting the spatial structure information of the exhibition hall layout and encoding it into a basic multi-channel mask, and combining this with an enhanced multi-channel mask, exhibit placement prediction is performed. After determining the current exhibit's layout coordinates, the enhanced multi-channel mask is updated for predicting the next exhibit's position. This sequential prediction mechanism ensures that exhibit placement is simultaneously constrained by the exhibition hall's inherent geometry and the exhibit's importance hierarchy, achieving a balance between spatial feasibility and display logic while ensuring that important exhibits occupy preferential spaces.

[0030] In this embodiment, S100 includes: S110. Construct an initial multi-channel conditional image, which includes an outer mask, a wall mask, an inner mask, an entrance mask, and an initial center mask; wherein, the geometric centroid is calculated based on the contour information of each sub-exhibition hall, and a marked area is generated in the initial center mask with the geometric centroid as the center; preferably, the marked area is a 7×7 pixel square area, the pixel value within the square area is set to 255, and the other positions are set to 0; S120. Input the current initial multi-channel conditional image into the pre-constructed exhibition hall center prediction network to generate a prediction image of the center position of a single sub-exhibition hall and perform correction processing to update the current initial center mask. S130, repeat S120 until the iteration termination condition is met, then use the current initial center mask as the center mask; S140. Merge the center mask, outer mask, wall mask, inner mask, and entrance mask to obtain a multi-channel conditional image of the exhibition hall.

[0031] In this embodiment, the initial center mask in S110 is a blank image.

[0032] Since the center of a sub-exhibition hall corresponds to only a single pixel in the image, it is easy to cause class imbalance. Therefore, in S110 of this invention, the geometric centroid is calculated based on the contour information of each sub-exhibition hall, and a marked region is generated in the initial center mask with the geometric centroid as the center, which can effectively improve the stability and robustness of the sub-exhibition hall center prediction.

[0033] This invention generates an initial center mask based on the geometric centroid of the outline of each sub-exhibition hall through S110, providing a reasonable initial prior for the prediction of the exhibition hall center. Iterative prediction and correction are performed by combining the encoder-decoder structure of the exhibition hall center prediction network, gradually converging to the accurate center position of the sub-exhibition hall. This overcomes the possible bias in a single prediction, so that the final generated center mask accurately reflects the spatial centroid of the sub-exhibition hall, providing reliable spatial guidance for the diffusion model of exhibition hall generation conditions, thereby improving the structural rationality of the generated exhibition hall layout.

[0034] In this embodiment, the exhibition hall center prediction network determines all sub-exhibition hall centers by predicting and updating them one by one. For the specific network architecture, please refer to [reference needed]. Figure 3 It employs an encoder-decoder structure. Preferably, the encoder is a ViT encoder, which divides the initial multi-channel conditional image into multiple image blocks of equal size, maps the image blocks into vector sequences through a linear projection layer, introduces positional encoding to preserve spatial position information, and uses a multi-layer Transformer Encoder to perform global feature modeling on the image block sequence; the decoder is preferably an ASPP decoder, which contains multiple parallel dilated convolution modules with dilation rates of 6, 12, 18, and 24, respectively, which can fuse multi-scale spatial information, restore the original spatial resolution through convolution and transposed convolution, and finally output the current initial center mask.

[0035] Furthermore, in this embodiment, S120 includes: S121. Input the current initial multi-channel conditional image into the pre-constructed exhibition hall center prediction network to generate a prediction image of the center location of a single sub-exhibition hall; S122. Traverse the prediction image through a sliding window of a preset size, and count the number of pixels with a preset value in each sliding window; preferably, the preset size is 7×7 pixels and the preset value is 255. S123. Select the sliding window corresponding to the maximum number of pixels as the target window; S124. Set the pixel values ​​in the target window to preset values, take the image area covered by the target window as the corrected center position of the sub-exhibition hall, update the pixel values ​​at the corresponding positions in the current initial center mask, and set the remaining positions to 0.

[0036] This invention achieves precise localization and noise reduction of the center position of the exhibition hall in the predicted image through a sliding window correction mechanism in steps S122-S124. Specifically, step S122 statistically analyzes the pixel value distribution through a sliding window of a preset size, step S123 selects a target window, and step S124 uniformly sets the pixels within the target window to preset values ​​to determine the center position. This transforms the continuous probability map predicted by the neural network into discrete, centralized center point markers, eliminating the ambiguity of the center position caused by prediction noise. This ensures that each sub-exhibition hall has clear and unique center coordinates, thereby improving the spatial accuracy of the multi-channel conditional image and providing clear structural anchors for subsequent layout generation.

[0037] In this embodiment, if the current initial center mask already contains all predicted sub-hall centers, the current initial center mask is used as input, and the image with all pixel values ​​of 255 is used as the prediction target; if the vast majority of pixel values ​​in the output of the hall center prediction network are 0, it is determined that the iteration termination condition is met, and the current initial center mask is used as the center mask.

[0038] In this embodiment, S200 includes: S210. Input the multi-channel conditional image into the exhibition hall attribute encoder, extract its features and map them into a conditional vector; preferably, extract features through multi-layer convolution, batch normalization and ReLU activation; downsample the feature map in each convolution layer; map the features into a 512-dimensional conditional vector through a fully connected layer; S220. Inject the conditional vector into the U-Net structured denoising network to perform progressive denoising on the noisy image and generate an exhibition hall layout image including walls and passageways.

[0039] Preferably, the noisy image is obtained by superimposing T-order Gaussian noise onto a real exhibition hall layout image.

[0040] Please refer to Figure 4 In this embodiment, the exhibition hall generation conditional diffusion model includes a pre-trained exhibition hall attribute encoder and a U-Net structure; the exhibition hall attribute encoder is composed of a multi-layer convolutional neural network. The exhibition hall attribute encoder extracts features from the multi-channel conditional images and maps them to conditional vectors, which are then injected into the denoising network of the U-Net structure. The U-Net network is used to perform progressive denoising: the encoder extracts features from the noisy image; the decoder progressively restores the spatial structure; skip connections fuse shallow location information and deep semantic information, finally generating an exhibition hall layout image including walls and passageways. Figure 4 middle, Represents a noisy image. The noise image at step t represents an intermediate state during the diffusion process, containing some noise and some structural information. This represents the noisy image at step t-1, which, after denoising, is closer to the actual layout of the exhibition hall. This represents the final generated image of the exhibition hall layout, including walls and passageways. This represents the forward-noise conditional distribution. Represents a condition vector. For U-Net network parameters; This represents the inverse denoising conditional distribution.

[0041] Preferably, the exhibition hall generation conditional diffusion model adopts a classifier-free guidance strategy: during the training phase, conditional information is discarded with a certain probability, so that the model can learn both conditional and unconditional distributions at the same time; during the sampling phase, the strength of conditional constraints is controlled by hyperparameters.

[0042] This invention achieves efficient generation of exhibition hall layout images from multi-channel conditional images through the collaborative design of an exhibition hall attribute encoder and a U-Net structure. Specifically, the exhibition hall attribute encoder, composed of a multi-layer convolutional neural network, performs deep feature extraction on the multi-channel conditional images and maps them into conditional vectors. The multi-layer CNN encoder fully captures the spatial hierarchical features (external boundaries, walls, internal spaces, etc.) in the multi-channel mask. After the conditional vectors are injected into the U-Net denoising network, the noisy image is guided to gradually denoise and generate an exhibition hall layout including walls and passages. The encoder-decoder structure of U-Net effectively integrates the multi-scale information of the conditional vectors and the noisy image, ensuring that the generated exhibition hall layout image strictly follows the spatial constraints of the input while maintaining the topological coherence of the walls and passages, thus improving the controllability and structural rationality of the layout generation.

[0043] In this embodiment, the wall information includes at least the wall outline information, wall length information, and wall inflection point coordinates; the exhibition hall opening information includes at least the exhibition hall entrance location information, exhibition hall exit location information, and exhibition hall window location information; the exhibition hall internal space distribution information includes at least the information of the exhibits that have been arranged. The basic multi-channel mask includes wall masks, door masks, window masks, interior masks, masks of already placed exhibits, and composite masks of already placed exhibits. Among them, the masks of already placed exhibits and the composite masks of already placed elements are updated in real time as the position of the exhibits is iteratively predicted; preferably, the composite mask of already placed elements includes at least exhibits, doors, windows, and walls.

[0044] This invention uses a unified multi-channel mask based on the physical structure (walls, openings) and functional state (exhibit layout) of the exhibition hall to provide complete and explicit spatial constraints for the subsequent exhibit location prediction network. This allows exhibit placement to consider both geometric feasibility (avoiding walls, utilizing openings) and functional coordination (avoiding already placed exhibits), thereby improving the practicality and rationality of the exhibition hall layout results.

[0045] In this embodiment, S400 includes: S410. Obtain attribute information for all exhibits. The attribute information includes at least importance values, dimensions, classification values, and images of the exhibits. The importance values ​​and classification values ​​can be preset by the designer or user.

[0046] S420. Based on the importance values ​​of all exhibits, determine the predicted order of the locations of all exhibits; preferably, sort them from high to low importance values. S430. Based on the position prediction order, for the current exhibit, dynamically enhance the basic multi-channel mask based on its attribute information to generate an enhanced multi-channel mask. S440. Input the enhanced multi-channel mask into the pre-constructed exhibit location prediction network to obtain the current exhibit location probability distribution map.

[0047] This invention obtains attribute information including importance value, size, classification value and image in S410, determines the position prediction order based on importance value in S420 to ensure priority layout of key exhibits in S430, dynamically enhances the basic multi-channel mask for the current exhibit based on its attribute information, so that the prediction conditions of each exhibit are adapted to its individual characteristics (size, classification, visual features) in S440, and inputs the position prediction network to obtain a probability distribution map, thereby realizing priority layout of important exhibits and personalized exhibition layout strategy, and improving the narrative logic and viewing experience of the exhibition.

[0048] Furthermore, in this embodiment, S430 includes: S431. Generate a classification constraint mask to represent classification similarity by taking the classification values ​​of the current exhibit and the already laid-out exhibits; normalize the classification value of the current exhibit to the interval [0, 1]; calculate the difference between the classification value of the current exhibit and the classification value of the already laid-out exhibits; the smaller the difference, the higher the pixel brightness of the corresponding position in the classification constraint mask. S432. Using a pre-trained convolutional neural network, based on the exhibit images of the current exhibit and the exhibit images of the already arranged exhibits, extract the corresponding image feature vectors and calculate the corresponding similarity to generate a similarity mask for representing visual associations; the implementation process of step S432 can be found in [link to S432]. Figure 5 Preferably, the similarity is cosine similarity; S433. Merge the classification constraint mask, similarity mask and basic multi-channel mask to form an enhanced multi-channel mask.

[0049] In this embodiment, the exhibits are specifically paintings.

[0050] This invention generates a classification constraint mask based on classification values ​​to represent classification similarity in S431, guiding similar exhibits to form clusters or sequences in space. It also generates a similarity mask based on image feature vector similarity to represent visual association, enabling exhibits with similar visual styles to form spatial associations. In S433, the two are combined with a basic multi-channel mask to form an enhanced multi-channel mask. The synergistic effect ensures that the exhibition layout not only meets geometric feasibility but also possesses thematic aggregation and visual fluency, enhancing the professionalism and artistic appeal of the exhibition.

[0051] In this embodiment, the exhibit location prediction network adopts an encoder-decoder structure; the encoder is used to extract features from the enhanced multi-channel mask, and the decoder is used to output a location probability distribution map, where each pixel location corresponds to a probability distribution of a category label; for details, please refer to... Figure 6 The input end is an enhanced multi-channel mask, with ① to ⑧ corresponding to the wall mask, interior mask, door mask, window mask, already placed exhibit mask, classification constraint mask, similarity mask, and comprehensive mask of already placed exhibits, respectively. The resolution of each mask is uniformly 128×128 pixels. The output end outputs the position of each pixel corresponding to the area outside the wall, the wall area, the non-candidate layout area, and the candidate layout area (corresponding to...). Figure 6 ⑨ to The probability distribution of categories.

[0052] In this embodiment, non-candidate layout areas mean that there are already elements in the exhibition hall layout, such as doors, windows, or exhibits that have already been placed.

[0053] This invention achieves refined prediction of exhibit placement locations through an encoder-decoder network design and multi-category label output definition. Specifically, the encoder extracts deep features from enhanced multi-channel mask images, and the decoder outputs the probability distribution of each pixel location corresponding to the wall area, the area outside the wall, the non-candidate layout area, and the candidate layout area. By transforming location prediction into a pixel-level classification problem, clearly distinguishing between candidate and non-candidate layout areas, the network can learn the layout feasibility distribution of the exhibition space. The output probability distribution map directly reflects the suitability of each location as a placement point, providing an interpretable and quantifiable decision-making basis for subsequently determining specific layout coordinates.

[0054] In this embodiment, S500 includes: S510. Based on the location probability distribution map, perform connected component detection on the candidate arrangement area of ​​the current exhibit; S520. Select the target placement area for the current exhibit based on the area of ​​the connected region; preferably select the connected region with the largest area as the target placement area. S530. Calculate the geometric center point of the target layout area, and calculate the perpendicular distance from the geometric center point to each wall of the exhibition hall; S540. Select the perpendicular point corresponding to the minimum perpendicular distance as the arrangement coordinate of the current exhibit, and update the enhanced multi-channel mask based on the arrangement coordinate.

[0055] This invention performs connectivity detection on candidate arrangement areas in step S510, selects target arrangement areas based on area in step S520, calculates the geometric center point and its perpendicular distance to each wall in step S530, and selects the perpendicular point corresponding to the minimum perpendicular distance as the arrangement coordinates and updates the mask in step S540. Based on connectivity detection and area filtering, the spatial continuity of the arrangement area can be ensured; combined with the selection strategy of minimizing the perpendicular distance from the geometric center to the wall, exhibits can be placed closer to the wall (in line with conventional exhibition practices); dynamic mask updates reflect changes in the layout state in real time, ensuring the accuracy of exhibit placement and the consistency of layout iteration.

[0056] Furthermore, the method of the present invention also includes: The generated exhibition hall layout image is traversed to identify room corners, corridor endpoints and their adjacent relationships, and the structural topology information of the exhibition space is constructed. This structural topology information is serialized into a structured data file, which can be used for 3D modeling, rendering or interactive editing.

[0057] like Figure 7 As shown, the present invention also provides an exhibition hall layout generation system, which includes: Corresponding to the exhibition hall layout generation method, the present invention provides an exhibition hall layout generation system, which includes: The multi-channel conditional image acquisition module 10 is used to acquire and input multi-channel conditional images of the exhibition hall, which includes an external mask, a wall mask, an internal mask, an entrance mask, and a center mask. The exhibition hall layout image generation module 20 is used to input multi-channel conditional images into a pre-constructed exhibition hall generation conditional diffusion model, encode their features and fuse them with noisy images, and generate exhibition hall layout images through denoising processing. The mask encoding module 30 is used to extract spatial structure information based on the exhibition hall layout image and encode it into a basic multi-channel mask; the spatial structure information includes at least wall information, exhibition hall opening location information and exhibition hall internal space distribution information. The exhibit location prediction module 40 is used to input the enhanced multi-channel mask corresponding to the current exhibit into the pre-constructed exhibit location prediction network according to the importance order of the exhibits to obtain the location probability distribution map of the current exhibit; wherein, the enhanced multi-channel mask is generated based on the basic multi-channel mask and the attribute information of the current exhibit. The exhibit location determination module 50 is used to determine the current exhibit's layout coordinates based on the location probability distribution map and update the enhanced multi-channel mask; The iterative control module 60 is used to repeatedly call the exhibit location prediction module 40 and the exhibit location determination module 50 to perform corresponding functions based on the updated enhanced multi-channel mask, until the location prediction of all exhibits is completed and the exhibition hall layout result is generated.

[0058] Preferably, the system may further include a serialization module for traversing the generated exhibition hall layout image, identifying room corners, corridor endpoints and their adjacent relationships, and constructing the structural topology information of the exhibition space. This structural topology information is serialized into a structured data file, which can be used for 3D modeling, rendering or interactive editing.

[0059] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0060] Furthermore, 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 apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0061] The foregoing description illustrates and describes preferred embodiments of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept by means of the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for generating exhibition hall layouts, characterized in that, Includes the following steps: S100: Acquire multi-channel conditional images of the exhibition hall, including external mask, wall mask, internal mask, entrance mask, and center mask; S200. Input the multi-channel conditional image into the pre-constructed exhibition hall generation conditional diffusion model, encode its features and fuse it with the noisy image, and generate an exhibition hall layout image through denoising processing; S300. Extract spatial structure information based on the exhibition hall layout image and encode it into a basic multi-channel mask; the spatial structure information includes at least wall information, exhibition hall opening location information, and exhibition hall internal space distribution information. S400. Based on the importance order of the exhibits, the enhanced multi-channel mask corresponding to the current exhibit is input into the pre-constructed exhibit position prediction network to obtain the position probability distribution map of the current exhibit; wherein, the enhanced multi-channel mask is generated based on the basic multi-channel mask and the attribute information of the current exhibit. S500: Determine the arrangement coordinates of the current exhibit based on the location probability distribution map, and update the enhanced multi-channel mask; S600: Based on the updated enhanced multi-channel mask, repeat S400 and S500 until the position prediction of all exhibits is completed, generating the exhibition hall layout result.

2. The exhibition hall layout generation method according to claim 1, characterized in that, S100 includes: S110. Construct an initial multi-channel conditional image, which includes the outer mask, the wall mask, the inner mask, the entrance mask, and the initial center mask; wherein, the geometric centroid is calculated based on the contour information of each sub-exhibition hall, and a marked area is generated in the initial center mask with the geometric centroid as the center; S120. Input the current initial multi-channel conditional image into the pre-constructed exhibition hall center prediction network to generate a prediction image of the center position of a single sub-exhibition hall and perform correction processing to update the current initial center mask; the exhibition hall center prediction network adopts an encoder-decoder structure; S130, repeat S120 until the iteration termination condition is met, then use the current initial center mask as the center mask; S140. Merge the central mask, outer mask, wall mask, inner mask, and entrance mask to obtain a multi-channel conditional image of the exhibition hall.

3. The exhibition hall layout generation method according to claim 2, characterized in that, S120 includes: S121. Input the current initial multi-channel conditional image into the pre-constructed exhibition hall center prediction network to generate a prediction image of the center position of a single sub-exhibition hall; S122. Traverse the predicted image through a sliding window of a preset size, and count the number of pixels with a preset value in each sliding window; S123. Select the sliding window corresponding to the maximum number of pixels as the target window; S124. Set the pixel value in the target window to the preset value, take the image area covered by the target window as the corrected center position of the sub-exhibition hall, and update the pixel value of the corresponding position in the current initial center mask.

4. The exhibition hall layout generation method according to claim 1, characterized in that, The exhibition hall generation conditional diffusion model includes a pre-trained exhibition hall attribute encoder and a U-Net structure; the exhibition hall attribute encoder is composed of a multi-layer convolutional neural network. S200 includes: S210. Input the multi-channel conditional image into the exhibition hall attribute encoder, extract its features and map them into a conditional vector; S220. Inject the conditional vector into the denoising network of the U-Net structure, perform progressive denoising processing on the noisy image, and generate the exhibition hall layout image including walls and passages.

5. The exhibition hall layout generation method according to claim 1, characterized in that, The wall information includes at least wall outline information, wall length information, and wall inflection point coordinates; the exhibition hall opening information includes at least exhibition hall entrance location information, exhibition hall exit location information, and exhibition hall window location information; the exhibition hall internal space distribution information includes at least information on the already arranged exhibits. The basic multi-channel mask includes wall mask, door mask, window mask, interior mask, mask of already laid-out exhibits, and integrated mask of already laid-out exhibits.

6. The exhibition hall layout generation method according to claim 1, characterized in that, The S400 includes: S410. Obtain attribute information for all exhibits, wherein the attribute information includes at least importance values, dimensions, classification values, and images of exhibits. S420. Based on the importance values ​​of all exhibits, determine the predicted order of the locations of all exhibits; S430. Based on the predicted position order, for the current exhibit, dynamically enhance the basic multi-channel mask based on its attribute information to generate an enhanced multi-channel mask. S440. Input the enhanced multi-channel mask into the pre-constructed exhibit location prediction network to obtain the current exhibit location probability distribution map.

7. The exhibition hall layout generation method according to claim 6, characterized in that, The S430 includes: S431. Generate a classification constraint mask to characterize classification similarity by combining the classification values ​​of the current exhibit and the classification values ​​of the already laid-out exhibits. S432. Based on the exhibit image of the current exhibit and the exhibit image of the already laid out exhibit, extract the corresponding image feature vectors of the two and calculate the corresponding similarity to generate a similarity mask for representing visual association. S433. The classification constraint mask, similarity mask and basic multi-channel mask are merged to form an enhanced multi-channel mask.

8. The exhibition hall layout generation method according to claim 1, characterized in that, The exhibit location prediction network adopts an encoder-decoder structure; the encoder is used to extract features from the enhanced multi-channel mask, and the decoder is used to output the location probability distribution map, where each pixel location corresponds to the probability distribution of a category label. The category labels include at least the wall area, the area outside the wall, the non-candidate layout area, and the candidate layout area.

9. The exhibition hall layout generation method according to claim 1 or 8, characterized in that, The S500 includes: S510. Based on the location probability distribution map, perform connected region detection on the candidate arrangement area of ​​the current exhibit; S520. Select the target arrangement area of ​​the current exhibit based on the area of ​​the connected region; S530. Calculate the geometric center point of the target layout area, and calculate the perpendicular distance from the geometric center point to each wall of the exhibition hall; S540. Select the perpendicular point corresponding to the minimum perpendicular distance as the arrangement coordinate of the current exhibit, and update the enhanced multi-channel mask based on the arrangement coordinate.

10. A system for generating exhibition hall layouts, characterized in that, include: The multi-channel conditional image acquisition module is used to acquire and input multi-channel conditional images of the exhibition hall, which includes an external mask, a wall mask, an internal mask, an entrance mask, and a center mask. The exhibition hall layout image generation module is used to input the multi-channel conditional image into a pre-constructed exhibition hall generation conditional diffusion model, encode its features and fuse it with a noisy image, and generate an exhibition hall layout image through denoising processing; The mask encoding module is used to extract spatial structure information based on the exhibition hall layout image and encode it into a basic multi-channel mask; the spatial structure information includes at least wall information, exhibition hall opening location information, and exhibition hall internal space distribution information. The exhibit location prediction module is used to input the enhanced multi-channel mask corresponding to the current exhibit into a pre-constructed exhibit location prediction network according to the importance order of the exhibits, so as to obtain the location probability distribution map of the current exhibit; wherein, the enhanced multi-channel mask is generated based on the basic multi-channel mask and the attribute information of the current exhibit. The exhibit location determination module is used to determine the current exhibit's layout coordinates based on the location probability distribution map and update the enhanced multi-channel mask; The iterative control module is used to repeatedly call the exhibit location prediction module and exhibit location determination module to perform corresponding functions based on the updated enhanced multi-channel mask, until the location prediction of all exhibits is completed and the exhibition hall layout result is generated.

Images