A method and system for indoor layout generation based on dual rail constraints

Abstract

The application relates to the technical field of intelligent space layout generation, and discloses an indoor layout generation method and system based on double-track constraints, wherein the method comprises the following steps: acquiring indoor space structure data; constructing a current layout state and a candidate action set; scoring the candidate action based on a conditional scoring model; generating a score sequence after hard constraint modulation; calculating soft constraint energy; generating a candidate action probability distribution; and performing a step-by-step indoor layout generation task. Compared with the prior art which mainly relies on a single scoring model for layout generation, especially in a complex indoor scene with clear space non-placing areas and traffic constraints, it is difficult to guarantee layout legality and generation diversity at the same time. Due to the double-track constraint mechanism of hard constraint Boolean gate modulation and soft constraint energy hierarchical accumulation, the application realizes the probabilistic modeling of the prior legality screening and subsequent rationality differentiation of the candidate layout action, and improves the overall indoor layout generation quality.

Description

Technical Field

[0001] This invention relates to the field of intelligent space layout generation technology, and in particular to an indoor layout generation method and system based on dual-track constraints. Background Technology

[0002] Currently, most technologies related to interior layout generation rely on rule-based matching, single scoring models, or unified weighted constraint mechanisms to generate and filter candidate layout schemes. Their core process typically involves directly calculating a comprehensive score on the set of candidate actions and selecting or generating the layout result accordingly. However, in actual interior spaces, there are often multiple constraints simultaneously, such as areas where placement is prohibited, access path restrictions, functional zoning requirements, and style or usage preferences. These constraints differ significantly in their nature and strength.

[0003] Existing technologies generally treat different types of constraints at the same level, failing to effectively distinguish between layout legality conditions with absolute constraints and layout rationality conditions with relative preferences. This leads to illegal layout actions that do not meet spatial geometry or access requirements still participating in subsequent scoring or probability calculations during candidate layout generation, resulting in invalid layout results, unstable generation processes, or the need for additional post-processing steps for correction. Furthermore, in complex indoor scenarios with multiple coexisting layout preferences, existing technologies struggle to ensure layout legality while simultaneously considering the diversity and overall quality of the generated results. Especially during the step-by-step layout generation process, problems such as the accumulation of local decision-making conflicts and large fluctuations in layout quality easily arise, failing to fully meet the practical application needs of generating complex indoor spatial layouts.

[0004] Therefore, there is an urgent need for an indoor layout generation method that can separate and model different types of constraints without introducing global optimization or complex search mechanisms, and can achieve priority control of legality and probabilistic adjustment of rationality during the step-by-step layout generation process, so as to improve the stability, controllability and overall quality of the final layout result. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose an indoor layout generation method based on dual-track constraints. This method aims to solve the technical problem that existing technologies mainly rely on a single scoring model for layout generation, especially in complex indoor scenarios with clearly defined areas where placement is not permitted and access constraints exist. This makes it difficult to simultaneously ensure layout legality and generation diversity.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides an indoor layout generation method based on dual-track constraints.

[0007] The indoor layout generation method based on dual-track constraints includes: Step S10: Obtain indoor space structure data, and based on the indoor space structure data, perform the pre-modeling task of progressive layout generation using an explicit modeling method based on layout state, and output the current layout state corresponding to the t-th indoor layout generation instruction. and candidate action set ; Step S20: Based on the current layout state and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. ; Step S30: Based on the original scoring sequence of candidate actions Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. ; Step S40: Based on the hard-constrained modulated scoring sequence Following the hard-constraint modulated scores from high to low, the candidate action set is calculated sequentially using a hierarchical energy term accumulation method. Corresponding soft constraint energy ; Step S50: Joint hard-constrained modulation post-score sequence and soft constraint energy The vectorization fusion and normalization processes are performed to output the probability distribution of candidate actions in the current step, and the indoor layout generation task is executed based on the probability distribution of candidate actions in the current step.

[0008] Preferably, in step S10, indoor space structure data is acquired, and based on the indoor space structure data, a pre-modeling task of progressive layout generation is performed using an explicit modeling method based on layout state, and the current layout state is output. and candidate action set The steps specifically include: Step S101: Obtain indoor space structure data and construct the current layout state based on the indoor space structure data. , ;in, This represents the position, size, and orientation information of the furniture object after the application of the t-th interior layout generation command. Indicates the remaining available space; Indicates information about the gate and passageway; Indicates the wall and its fixed structure; Indicates the current generation stage identifier; Step S102: Based on the current layout state A candidate action enumeration and filtering mechanism based on generation stage identifier constraints is adopted for the current layout state. The executable layout operations are enumerated to generate a set of candidate actions. Among them, the candidate action set Each candidate action in the current layout state represents the current layout state. A specific layout generation operation that can be executed includes at least one of the following types: furniture category selection action, furniture instance placement action, furniture position determination action, and furniture orientation determination action; Step S103: Finally, output the current layout state. and candidate action set .

[0009] Preferably, in step S20, based on the current layout state and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. The steps specifically include: Step S201: First, construct the current layout state based on the indoor space structure data. , Perform explicit feature encoding to construct the input feature sequence for the conditional scoring model; Step S202: Pre-construct a conditional scoring model based on the Transformer encoding structure. The conditional scoring model includes an input embedding layer, used to map the input feature sequence to an embedding representation space of a unified dimension; a positional encoding layer, used to introduce relative positional information into each feature vector in the unified-dimensional embedding representation space to represent the structural relationship of data from different modalities; a multi-head self-attention encoding layer, used to model the global correlation between different modalities; a feedforward neural network layer, used to perform nonlinear mapping and expression enhancement on the features aggregated by the self-attention mechanism; a linear scoring mapping layer, used to map the encoded features corresponding to the nonlinear mapping and expression enhancement to continuous original scoring values; and an output layer, used to output the original scoring sequence of candidate actions based on the original scoring values. ; Step S203: Obtain the historical input feature sequence and the original score sequence of historical candidate actions. Take the historical input feature sequence as the input of the conditional scoring model and the original score sequence of historical candidate actions as the output of the conditional scoring model. Combine the cross-entropy loss function to perform the pre-training process of the conditional scoring model. Step S204: Input the input feature sequence into the pre-trained conditional scoring model, and the conditional scoring model outputs the original score sequence of candidate actions. .

[0010] Preferably, in step S201, the current layout state is first constructed based on the indoor space structure data. , The steps for performing explicit feature encoding to construct the input feature sequence of the conditional scoring model specifically include: Step S2011: Extract position parameters, size parameters, and orientation parameters from the furniture object's position information, size information, and orientation information, and map the position parameters, size parameters, and orientation parameters to the corresponding object state feature vectors. For the remaining placeable area Gate and passageway information and walls and fixed structures Extract geometric constraint features and map them to environmental constraint feature vectors. ;Identification of the current generation stage Perform stage embedding encoding to generate the current layout generation stage feature vector. ; Step S2012: For the candidate action set For each candidate action, extract the action type, target position parameters, and action attributes related to the generation stage to construct the corresponding action feature vector. ; Step S013: Based on object state feature vector Environmental constraint feature vector Feature vectors in the current layout generation stage and action feature vector Construct the input feature sequence for the conditional scoring model.

[0011] Preferably, in step S30, the candidate action original scoring sequence is used. Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. The steps specifically include: Step S301: Based on the current layout state The set of hard constraints for the current step is generated using preset geometric constraint determination rules and Boolean gating mapping. ; Step S302: Based on the current step's hard constraint set and candidate action original score sequence For the candidate action set The task of performing segmented hard-track modulation is adopted, and the output is a hard-constrained modulated scoring sequence. ; ; in, The scoring sequence after hard constraint modulation The score after hard constraint modulation of the i-th term; For action sets The corresponding i-th candidate action; The preset mandatory constraint reward bias coefficient; This is used to indicate that the probability of the candidate action is zero during the subsequent probability normalization process.

[0012] Preferably, in step S40, the soft constraint energy Continuous real numbers are used to describe the relative difference in rationality between candidate actions before and after application; soft constraint energy. The formula is expressed as: ; in, Represents the soft constraint energy of the i-th term; Let be the geometric deviation energy of the i-th term, which is a weighted sum of distance deviation, orientation deviation, and scale deviation. Let be the difference energy between the i-th furniture item and the target style vector; Let be the functional association deviation energy between the i-th candidate action and the already placed furniture; The energy of the effect of adding the i-th candidate action on the consistency of multi-room layout; , , and These are preset weighting coefficients.

[0013] Preferably, in step S50, the combined hard-constraint modulated scoring sequence and soft constraint energy The process involves vectorization fusion and normalization to output the probability distribution of candidate actions for the current step, and then executing the indoor layout generation task based on this probability distribution. Specifically, this includes: Step S501: Joint hard-constrained modulation post-score sequence and soft constraint energy Vectorization and normalization are performed to output the probability of candidate actions for the current step; among which, the hard-constrained modulated scoring sequence is used. The score after medium-hard constraint modulation is The item will have its probability of the candidate action in the current step set to zero; Step S502: Randomly sample all candidate action probabilities for the current step using the Softmax probability normalization statistical method, output the probability distribution of candidate actions for the current step, and execute the indoor layout generation task based on the probability distribution of candidate actions for the current step.

[0014] This invention also provides an indoor layout generation system based on dual-track constraints, comprising: The layout state and candidate action construction module is used to acquire indoor space structure data. Based on the indoor space structure data, it performs a pre-modeling task of progressive layout generation using an explicit modeling method based on layout state, and outputs the current layout state corresponding to the t-th indoor layout generation instruction. and candidate action set ; The conditional candidate action scoring module is used to score actions based on the current layout state. and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. ; Hard-constrained Boolean-gated modulation module, used to evaluate the original scoring sequence of candidate actions. Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. ; A hierarchical soft-constraint energy calculation module is used for calculating scores based on hard-constraint modulated post-score sequences. Following the hard-constraint modulated scores from high to low, the candidate action set is calculated sequentially using a hierarchical energy term accumulation method. Corresponding soft constraint energy ; The joint probability modulation and layout execution module is used to jointly perform hard-constrained modulation and scoring sequences. and soft constraint energy The vectorization fusion and normalization processes are performed to output the probability distribution of candidate actions in the current step, and the indoor layout generation task is executed based on the probability distribution of candidate actions in the current step.

[0015] The present invention also provides an indoor layout generation device based on dual-track constraints, comprising: a memory, a processor, and an indoor layout generation program based on dual-track constraints stored in the memory and executable on the processor. When the indoor layout generation program based on dual-track constraints is executed by the processor, it implements an indoor layout generation method based on dual-track constraints.

[0016] The present invention also provides a computer program product, including an interior layout generation program based on dual-track constraints, wherein the interior layout generation program based on dual-track constraints implements the interior layout generation method based on dual-track constraints when executed by a processor.

[0017] The beneficial effects of this invention are as follows: This invention separates the hard constraints and soft constraints of candidate actions into two tracks for modeling. In the layout generation process, it first explicitly eliminates candidate actions that do not meet the rules of spatial geometry, accessibility, and non-placement based on the hard constraint Boolean gating modulation mechanism. Then, it distinguishes the rationality of legal candidate actions based on the hierarchical soft constraint energy. This avoids the problem in the prior art where illegal layout actions participate in subsequent scoring or probability calculations, which leads to the failure or instability of the layout results. This significantly improves the legality and stability of the indoor layout generation process.

[0018] This invention constructs a candidate action probability distribution by vectorizing and normalizing the scoring sequence modulated by hard constraints and the soft constraint energy, and then performs a stepwise layout generation task based on the probability distribution. Without introducing global optimization or iterative search, it achieves a probabilistic balance between multiple layout preferences, effectively improving the diversity and overall generation quality of indoor layout results under complex constraints, and enhancing the adaptability and controllability of the layout generation method in different indoor space structure scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the first embodiment of an indoor layout generation method based on dual-track constraints according to the present invention.

[0021] Figure 2 This is a schematic diagram of an equipment for generating an indoor layout based on dual-track constraints according to the present invention. Detailed Implementation

[0022] 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. 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.

[0023] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the indoor layout generation method based on dual-track constraints of the present invention, which presents the first embodiment of the indoor layout generation method based on dual-track constraints of the present invention.

[0024] In the first embodiment, the indoor layout generation method based on dual-track constraints includes: Step S10: Obtain indoor space structure data, and based on the indoor space structure data, perform the pre-modeling task of progressive layout generation using an explicit modeling method based on layout state, and output the current layout state corresponding to the t-th indoor layout generation instruction. and candidate action set ; It should be noted that "explicit modeling based on layout state" refers to explicitly expressing the interior space structure information, the information of generated furniture objects, and the information of remaining available space in a unified layout state form. The layout state includes at least the geometric position, size parameters, orientation information, occupied space range of the placed furniture, and the set of remaining placeable areas formed therefrom. At the same time, the candidate action set is dynamically generated based on the current layout state and is used to describe the types of layout operations that can be executed in the current step, including optional furniture categories, optional placement areas, or optional placement methods, rather than an exhaustive enumeration of all possible layout actions.

[0025] It is understandable that by explicitly constructing the current layout state and generating a set of candidate actions in each generation step of the progressive layout generation, the subsequent candidate action scoring, constraint modulation, and probability calculation can directly perceive the current real layout environment, thereby ensuring that the layout generation process is consistent with the actual spatial conditions, avoiding the situation of ignoring placed furniture or spatial boundary conditions during the generation process, and improving the stability and continuity of the progressive layout generation process.

[0026] For example, in a comparative experiment of generating typical living room layouts, when the room size is about 20 square meters and there are fixed doors, windows and passageway constraints, in the layout schemes using the traditional overall generation method, more than 30% of the generated results have problems such as obstructed passage or overlapping furniture in the initial layout stage, requiring additional post-processing steps for correction. However, after adopting the explicit modeling of layout state and dynamic generation of candidate actions described in this step, in the layout results generated under the same experimental conditions, most schemes can maintain effective perception of the placed furniture and remaining space in the step-by-step generation process. The proportion of obvious spatial conflicts during the generation process is significantly reduced, the overall layout process is more stable, and the need for adjustment of the layout results in subsequent steps is significantly reduced.

[0027] Step S20: Based on the current layout state and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. ; It should be noted that the "conditional scoring model based on Transformer encoding structure" mentioned in this step refers to a sequence modeling network structure that takes the current layout state and the candidate action set as joint inputs. The Transformer encoding structure is used to jointly model the spatial relationship information contained in the layout state and the candidate action features, and outputs the original score value to represent the relative priority of each candidate action. The original score sequence only reflects the degree of conditional preference of the model for candidate actions under the condition of no constraints, and does not directly represent the legality of the candidate actions or the final selection result.

[0028] Understandably, by incorporating the current layout state as a conditional input into the candidate action scoring process, the scoring result of each candidate action is kept consistent with the currently generated layout. This allows the scoring results to dynamically reflect the impact of factors such as changes in spatial structure and the position of placed furniture on subsequent layout decisions, ensuring the continuity and contextual relevance of candidate action scoring during the progressive generation process.

[0029] It should be understood that, compared with the traditional method of independently scoring candidate layout actions based on fixed rules or static features, this invention explicitly introduces layout state information in the scoring stage through a conditional scoring model, so that the scoring of candidate actions is no longer divorced from the actual layout environment. This avoids the problem of the scoring results not matching the actual spatial conditions during the gradual generation of the layout, and provides a more reasonable and stable initial scoring basis for subsequent modulation processing based on hard and soft constraints.

[0030] Step S30: Based on the original scoring sequence of candidate actions Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. ; It should be noted that the "Boolean gating modulation based on hard constraints of candidate actions" mentioned in this step refers to determining the feasibility of each candidate action in the candidate action set according to predefined hard constraint rules, and applying Boolean gating processing to the original score of the corresponding candidate action based on the determination result. The hard constraint rules include at least spatial non-placement area constraints, furniture geometric collision constraints, passageway clearance constraints, and basic layout rules that must be met. The Boolean gating modulation is only used to distinguish whether candidate actions are allowed to enter the subsequent processing stage, and does not evaluate the superiority or inferiority relationship between candidate actions.

[0031] It is understandable that by introducing hard-constrained Boolean gating modulation after the candidate action scoring is completed, candidate actions that do not meet the spatial geometry or basic layout rules are explicitly marked as unselectable in the current generation step. This ensures that subsequent steps only perform further energy calculations and probability modeling on legal candidate actions, avoids illegal layout actions from continuously propagating in the generation process, and improves the reliability and stability of the step-by-step layout generation process.

[0032] For example, in a set of living room layout generation experiments containing main passageway constraints, when the furniture placement corresponding to a candidate action occupies the preset passageway area, the traditional unified scoring method may only give a certain degree of score penalty to the candidate action, and there is still a possibility that it will be selected in subsequent steps. However, after adopting the hard constraint Boolean gating modulation method described in this step, the candidate action is directly marked as unselectable when it is determined to violate the passageway constraint. Its corresponding score no longer participates in subsequent energy calculation and probability normalization processing in the current step, thereby completely avoiding the generation of obstructed passageway layout results in the early stage of generation and significantly reducing the need for subsequent layout correction.

[0033] Step S40: Based on the hard-constrained modulated scoring sequence Following the hard-constraint modulated scores from high to low, the candidate action set is calculated sequentially using a hierarchical energy term accumulation method. Corresponding soft constraint energy ; It should be noted that the "layered energy item accumulation method" mentioned in this step refers to, for candidate actions in the candidate action set that have been modulated by hard constraint Boolean gating, introducing multiple soft constraint energy items with different constraint attributes in descending order of their hard constraint modulated scores, and accumulating each soft constraint energy item according to a preset rule to obtain the soft constraint energy value corresponding to each candidate action; wherein, the soft constraint energy items are used to characterize relative constraint factors such as furniture spacing, functional area coordination, visual balance, or usage preferences, and the accumulation process only reflects the relative cost of the candidate action at the soft constraint level.

[0034] It is understandable that by calculating the soft constraint energy of candidate actions sequentially according to the scoring order after hard constraint modulation, higher priority candidate actions can complete the soft constraint evaluation in a more stable constraint environment. This avoids the interference of a large number of low priority or marginal candidate actions on the energy distribution during the soft constraint calculation stage, thereby improving the discriminability and stability of the soft constraint energy calculation results.

[0035] Step S50: Joint hard-constrained modulation post-score sequence and soft constraint energy The vectorization fusion and normalization processes are performed to output the probability distribution of candidate actions in the current step, and the indoor layout generation task is executed based on the probability distribution of candidate actions in the current step.

[0036] It should be noted that the "vectorized fusion processing and normalization processing" mentioned in this step refers to, for each candidate action, combining its corresponding hard constraint modulated score and soft constraint energy according to a preset fusion rule under a unified scale, and converting the fusion result into a relative probability distribution among candidate actions through normalization; wherein, the fusion processing is only used to characterize the relative selection tendency of different candidate actions under the combined effect of legality and rationality in the current step, rather than performing a global evaluation or optimal solution search on the candidate actions.

[0037] It is understandable that by introducing a candidate action representation based on probability distribution after completing hard constraint elimination and soft constraint energy calculation, the step-by-step layout generation process is transformed from a deterministic selection to a probability-driven execution process. This allows for a flexible balance between different soft constraint preferences while ensuring that all selected candidate actions meet the hard constraint conditions, which is beneficial to improving the stability and diversity of results in the layout generation process.

[0038] Example 2: Furthermore, the present invention provides an indoor layout generation system based on dual-track constraints, employing an indoor layout generation method based on dual-track constraints as described in the above embodiments, which can solve a technical problem related to indoor layout generation based on dual-track constraints. Compared with the prior art, the beneficial effects of the indoor layout generation system based on dual-track constraints provided by the present invention are the same as those of the indoor layout generation method based on dual-track constraints provided in the above embodiments, and other technical features of the indoor layout generation system based on dual-track constraints are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0039] Example 3: This invention provides an indoor layout generation device based on dual-track constraints. Please refer to... Figure 2An indoor layout generation device based on dual-track constraints includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the indoor layout generation method based on dual-track constraints described in Embodiment 1 above. An indoor layout generation device based on dual-track constraints in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle terminals (e.g., vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. An indoor layout generation device based on dual-track constraints is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. An indoor layout generation device based on dual-track constraints may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. Random access memory 1004 also stores various programs and data required for the operation of a dual-track constraint-based interior layout generation device. Processing device 1001, read-only memory 1002, and random access memory 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows a dual-track constraint-based interior layout generation device to communicate wirelessly or wiredly with other devices to exchange data. Although a dual-track constraint-based interior layout generation device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented or possessed alternatively.

[0040] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for generating an interior layout based on dual-track constraints. The computer program product provided by this invention can solve a technical problem related to generating an interior layout based on dual-track constraints. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the method for generating an interior layout based on dual-track constraints provided in the above embodiments, and will not be repeated here.

[0041] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0042] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0043] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for generating indoor layouts based on dual-track constraints, characterized in that, The methods include: Step S10: Obtain indoor space structure data, and based on the indoor space structure data, perform the pre-modeling task of progressive layout generation using an explicit modeling method based on layout state, and output the current layout state corresponding to the t-th indoor layout generation instruction. and candidate action set ; Step S20: Based on the current layout state and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. ; Step S30: Based on the original scoring sequence of candidate actions Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. ; Step S40: Based on the hard-constrained modulated scoring sequence Following the hard-constraint modulated scores from high to low, the candidate action set is calculated sequentially using a hierarchical energy term accumulation method. Corresponding soft constraint energy ; Step S50: Joint hard-constrained modulation post-score sequence and soft constraint energy The vectorization and normalization processes are performed to output the probability distribution of candidate actions in the current step, and the indoor layout generation task is executed based on the probability distribution of candidate actions in the current step.

2. The indoor layout generation method based on dual-track constraints as described in claim 1, characterized in that, In step S10, indoor space structure data is acquired. Based on the indoor space structure data, a pre-modeling task of progressive layout generation is performed using an explicit modeling method based on layout state, and the current layout state is output. and candidate action set The steps specifically include: Step S101: Obtain indoor space structure data and construct the current layout state based on the indoor space structure data. , ;in, This represents the position, size, and orientation information of the furniture object after the application of the t-th interior layout generation command. Indicates the remaining available space; Indicates information about the gate and passageway; Indicates the wall and its fixed structure; Indicates the current generation stage identifier; Step S102: Based on the current layout state A candidate action enumeration and filtering mechanism based on generation stage identifier constraints is adopted for the current layout state. The executable layout operations are enumerated to generate a set of candidate actions. Among them, the candidate action set Each candidate action in the current layout state represents the current layout state. A specific layout generation operation that can be executed includes at least one of the following types: furniture category selection action, furniture instance placement action, furniture position determination action, and furniture orientation determination action; Step S103: Finally, output the current layout state. and candidate action set .

3. The indoor layout generation method based on dual-track constraints as described in claim 2, characterized in that, In step S20, based on the current layout state and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. The steps specifically include: Step S201: First, construct the current layout state based on the indoor space structure data. , Perform explicit feature encoding to construct the input feature sequence for the conditional scoring model; Step S202: Pre-construct a conditional scoring model based on the Transformer encoding structure. The conditional scoring model includes an input embedding layer, used to map the input feature sequence to an embedding representation space of a unified dimension; a positional encoding layer, used to introduce relative positional information into each feature vector in the unified-dimensional embedding representation space to represent the structural relationship of data from different modalities; a multi-head self-attention encoding layer, used to model the global correlation between different modalities; a feedforward neural network layer, used to perform nonlinear mapping and expression enhancement on the features aggregated by the self-attention mechanism; a linear scoring mapping layer, used to map the encoded features corresponding to the nonlinear mapping and expression enhancement to continuous original scoring values; and an output layer, used to output the original scoring sequence of candidate actions based on the original scoring values. ; Step S203: Obtain the historical input feature sequence and the original score sequence of historical candidate actions. Take the historical input feature sequence as the input of the conditional scoring model and the original score sequence of historical candidate actions as the output of the conditional scoring model. Combine the cross-entropy loss function to perform the pre-training process of the conditional scoring model. Step S204: Input the input feature sequence into the pre-trained conditional scoring model, and the conditional scoring model outputs the original score sequence of candidate actions. .

4. The indoor layout generation method based on dual-track constraints as described in claim 3, characterized in that, In step S201, the current layout state is first constructed based on the indoor space structure data. , The steps for performing explicit feature encoding to construct the input feature sequence of the conditional scoring model specifically include: Step S2011: Extract position parameters, size parameters, and orientation parameters from the furniture object's position information, size information, and orientation information, and map the position parameters, size parameters, and orientation parameters to the corresponding object state feature vectors. For the remaining placeable area Gate and passageway information and walls and fixed structures Extract geometric constraint features and map them to environmental constraint feature vectors. ;Identification of the current generation stage Perform stage embedding encoding to generate the current layout generation stage feature vector. ; Step S2012: For the candidate action set For each candidate action, extract the action type, target position parameters, and action attributes related to the generation stage to construct the corresponding action feature vector. ; Step S2013: Based on object state feature vector Environmental constraint feature vector Feature vectors in the current layout generation stage and action feature vector Construct the input feature sequence for the conditional scoring model.

5. The indoor layout generation method based on dual-track constraints as described in claim 1, characterized in that, In step S30, based on the original scoring sequence of candidate actions... Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. The steps specifically include: Step S301: Based on the current layout state The set of hard constraints for the current step is generated using preset geometric constraint determination rules and Boolean gating mapping. ; Step S302: Based on the current step's hard constraint set and candidate action original score sequence For the candidate action set The task of performing segmented hard-track modulation is adopted, and the output is a hard-constrained modulated scoring sequence. ; ; in, The scoring sequence after hard constraint modulation The score after hard constraint modulation of the i-th term; For action sets The corresponding i-th candidate action; The preset mandatory constraint reward bias coefficient; This is used to indicate that the probability of the candidate action is zero during the subsequent probability normalization process.

6. The indoor layout generation method based on dual-track constraints as described in claim 5, characterized in that, In step S40, soft constraint energy Continuous real numbers are used to describe the relative difference in rationality between candidate actions before and after application; soft constraint energy. The formula is expressed as: ； in, Represents the soft constraint energy of the i-th term; Let be the geometric deviation energy of the i-th term, which is a weighted sum of distance deviation, orientation deviation, and scale deviation. Let be the difference energy between the i-th furniture item and the target style vector; Let be the functional association deviation energy between the i-th candidate action and the already placed furniture; The energy of the effect of adding the i-th candidate action on the consistency of multi-room layout; , , and These are preset weighting coefficients.

7. The indoor layout generation method based on dual-track constraints as described in claim 5, characterized in that, In step S50, the combined hard-constrained modulation scoring sequence and soft constraint energy The process involves vectorization fusion and normalization to output the probability distribution of candidate actions for the current step, and then executing the indoor layout generation task based on this probability distribution. Specifically, this includes: Step S501: Joint hard-constrained modulation post-score sequence and soft constraint energy Vectorization and normalization are performed to output the probability of candidate actions for the current step; among which, the hard-constrained modulated scoring sequence is used. The score after medium-hard constraint modulation is The item will have its probability of the candidate action in the current step set to zero; Step S502: Randomly sample all candidate action probabilities for the current step using the Softmax probability normalization statistical method, output the probability distribution of candidate actions for the current step, and execute the indoor layout generation task based on the probability distribution of candidate actions for the current step.

8. An indoor layout generation system based on dual-track constraints, applied to the indoor layout generation method based on dual-track constraints according to any one of claims 1 to 7, characterized in that, The indoor layout generation system based on dual-track constraints includes: The layout state and candidate action construction module is used to acquire indoor space structure data. Based on the indoor space structure data, it performs a pre-modeling task of progressive layout generation using an explicit modeling method based on layout state, and outputs the current layout state corresponding to the t-th indoor layout generation instruction. and candidate action set ; The conditional candidate action scoring module is used to score actions based on the current layout state. and candidate action set The candidate action scoring task is performed using a pre-defined conditional scoring model based on the Transformer encoding structure, and the original score sequence of the candidate actions is output. ; Hard-constrained Boolean-gated modulation module, used to evaluate the original scoring sequence of candidate actions. Perform a Boolean-gated modulation task based on hard constraints of candidate actions, and output the hard-constrained modulated score sequence. ; A hierarchical soft-constraint energy calculation module is used for calculating scores based on hard-constraint modulated post-score sequences. Following the hard-constraint modulated scores from high to low, the candidate action set is calculated sequentially using a hierarchical energy term accumulation method. Corresponding soft constraint energy ; The joint probability modulation and layout execution module is used to jointly perform hard-constrained modulation and scoring sequences. and soft constraint energy The vectorization and normalization processes are performed to output the probability distribution of candidate actions in the current step, and the indoor layout generation task is executed based on the probability distribution of candidate actions in the current step.

9. An indoor layout generation device based on dual-track constraints, characterized in that, The dual-track constraint-based indoor layout generation device includes: a memory, a processor, and a dual-track constraint-based indoor layout generation program stored in the memory and executable on the processor. When the dual-track constraint-based indoor layout generation program is executed by the processor, it implements the dual-track constraint-based indoor layout generation method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes an indoor layout generation program based on dual-track constraints, which, when executed by a processor, implements an indoor layout generation method based on dual-track constraints as described in any one of claims 1 to 7.

Images