Industrial design variant generation method and system based on structured parameter model and feature decoupling

Abstract

The invention discloses an industrial design variant generation method and system based on a structured parameter model and feature decoupling, and the method comprises the following steps: obtaining an input image of a target design object in response to a variant generation request for the target design object, and calling a pre-trained visual semantic analysis model to analyze the input image, extracting a design semantic unit set comprising a structural layer, a morphological layer and a material process layer; and constructing a structured parameter model based on the design semantic unit set, and carrying out decoupling processing on the features of the design object. According to the method, the unstructured image visual features are mapped into mutually independent and quantifiable parameterized attributes by constructing the structured parameter model. According to the technical means, the defect of high feature coupling in an existing generative artificial intelligence model is effectively overcome.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided design, artificial intelligence content generation, and computer vision technology, specifically to a method and system for generating industrial design variants based on structured parametric models and feature decoupling. Background Technology

[0002] With the rapid development of generative artificial intelligence technology, the use of AI to generate images has become increasingly common in fields such as artistic creation and entertainment. General generative models, such as Midjourney and StableDiffusion, can quickly generate high-quality images based on text descriptions. However, in the professional field of industrial design, existing general generative technologies have significant limitations and are difficult to directly apply to actual production design processes.

[0003] First, there is a lack of structured control capabilities. Existing text-to-image or image-to-image technologies are mostly based on unstructured natural language descriptions. This descriptive approach is vague and imprecise, failing to meet the precise control requirements for specific attributes in industrial design. For example, in jewelry design, a designer might want to change only the "setting technique" or "metal texture" of the ring band, but must strictly maintain the "shape," "cut," and "size" of the main stone. General generative models often struggle to understand such strict geometric and logical constraints.

[0004] Secondly, design elements are heavily coupled. In existing diffusion model generation processes, images are often treated as a single pixel distribution. When attempting to modify local attributes via prompts (such as changing the product's color), it often leads to unexpected deformations of other key features (such as the shape of the brand logo or the proportions of the product's core functional structures). This strong coupling between features undermines the rigor of industrial design.

[0005] Finally, the generated results are highly random. Batch-generated results are often disordered random noise distributions, lacking a systematic deduction that conforms to design logic. For example, designers need a solution deduction based on a certain series of logic (such as a gradient from "modern minimalism" to "retro luxury"), but the variants generated by existing tools are often chaotic, making it extremely costly for designers to select effective solutions and failing to provide effective design assistance. To address this, a method and system for generating industrial design variants based on structured parametric models and feature decoupling are proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for generating industrial design variants based on structured parametric models and feature decoupling. By introducing visual semantic parsing, parametric decoupling modeling, and variant planning techniques, it achieves precise control over industrial design objects and the generation of diverse variants.

[0007] To achieve the above objectives, the first aspect of this disclosure is to solve the aforementioned technical problems.

[0008] One technical solution adopted in this application is: an industrial design variant generation method based on structured parametric models and feature decoupling, comprising the following steps: In response to a variant generation request for a target design object, an input image of the design object is obtained, and a pre-trained visual semantic parsing model is invoked to parse the input image and extract a set of design semantic units including structural, morphological, and material / process layers. Based on the set of design semantic units, a structured parametric model is constructed to decouple the features of the design object, map the visual features into mutually independent parametric attributes, and establish a hierarchical relationship topology between the attributes. The system receives constraint configuration instructions set by the user for the structured parameter model. The constraint configuration instructions divide each parameterized attribute into strongly constrained attributes and variable attributes, lock the parameter values ​​of the strongly constrained attributes, and define the variation space of the variable attributes. The variant programming algorithm is used to calculate the design variant path in the variant space, and a variant generation instruction set containing explicit differential descriptions is generated, the instruction set strictly maintaining the strong constraint property; Based on the variant generation instruction set, the generative model is driven to redraw or partially generate the design object, and output an industrial design variant image that satisfies the constraint configuration instructions.

[0009] In one implementation, the step of calling a pre-trained visual semantic parsing model to parse the input image and extract a set of semantic units includes: The input image is normalized and its background is weakened. The visual semantic parsing model is used to identify functional components and their combination relationships in a design object, rather than simply identifying object categories. According to the preset hierarchical rules, the parsing results are deconstructed into design semantic units. The design semantic units of the structure layer describe the basic components and their positions, the design semantic units of the morphology layer describe the geometric contours and proportions, and the design semantic units of the material and process layer describe the surface textures and physical properties.

[0010] In one implementation, the construction of a structured parametric model based on the set of design semantic units to decouple the features of the design object includes: The design semantic unit is transformed into a parameterized description in key-value pair form, forming an attribute list; Based on the logical structure of the design object, the attribute list is divided into a core structure layer, a component form layer, and a visual presentation layer. Establish a feature isolation mechanism to ensure that adjustments to any parameterized attribute value only affect the corresponding local features and do not cause unexpected deformations of other level attributes.

[0011] In one implementation, receiving the constraint configuration instruction set by the user for the structured parameter model includes: Provide an interactive interface to display the structured parameter model, and in response to user selections, mark the brand asset characteristics or core functional structure of the design object as strongly constrained attributes; In response to user adjustments, set the range of values ​​or style preferences for variable attributes; Upon receiving the constraint configuration, the system freezes the feature vector corresponding to the strong constraint attribute, keeping it constant during subsequent generation.

[0012] In one implementation, the computation of design variant paths within the variation space using a variant programming algorithm includes: The variation space of the variable attribute is calculated, and candidate attribute combinations with significant visual differences are selected based on the maximum design distance algorithm; Based on the preset style compatibility rules, combinations that not only meet the constraints but also violate the logic of industrial design are eliminated; The filtered attribute combinations are converted into differential descriptions in natural language form, and combined with the descriptions of the locked, strongly constrained attributes to synthesize the variant generation instruction set.

[0013] In one implementation, driving the generative model to redraw or partially generate the design object based on the variant generation instruction set includes: Based on the structured parameter model and the constraint configuration instructions, a region mapping relationship is established, and the design object is divided into a core structural region, a functional key region, and a variable shape region. Generate a region mask corresponding to the strong constraint attribute. During the generation process, a locking strategy is implemented on the pixel region covered by the region mask to prevent redrawing or structural modification. Within the variable shape region, pixel generation is guided by the differentiated description in the variant generation instruction set to achieve directional evolution of local features.

[0014] A second aspect of this disclosure, in order to solve the above-mentioned technical problems, is another technical solution adopted by this application: an industrial design variant generation system based on structured parametric models and feature decoupling, comprising: The semantic parsing module is configured to acquire the input image of the design object and call a pre-trained visual semantic parsing model to parse the input image and extract a set of design semantic units including structural layer, morphological layer and material process layer. The parameter model construction module is configured to construct a structured parameter model based on the set of design semantic units, decouple the features of the design object, map the visual features into mutually independent parameterized attributes, and establish a hierarchical relationship topology between the attributes. The constraint control module is configured to receive constraint configuration instructions set by the user for the structured parameter model. The constraint configuration instructions divide each parameterized attribute into strongly constrained attributes and variable attributes, lock the parameter values ​​of the strongly constrained attributes, and define the variation space of the variable attributes. The planning module is configured to use a variant planning algorithm to calculate design variant paths within the variant space and generate a variant generation instruction set containing explicit differential descriptions. The generation execution module is configured to drive the generative model to redraw or partially generate the design object based on the variant generation instruction set, and output an industrial design variant image that satisfies the constraint configuration instructions.

[0015] A third aspect of this disclosure is as follows: To solve the above-mentioned technical problems, another technical solution adopted by this application is: an electronic device, including a processor, a memory, and a communication interface, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned industrial design variant generation method and system based on structured parameter model and feature decoupling.

[0016] A fourth aspect of this disclosure is that, in order to solve the above-mentioned technical problems, another technical solution adopted by this application is: a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the industrial design variant generation method and system based on structured parameter model and feature decoupling as described above.

[0017] Compared with the prior art, the present invention has the following advantages: 1. This invention maps unstructured image visual features into independent, quantifiable parameterized attributes by constructing a structured parametric model. This technique effectively overcomes the high coupling of features in existing generative artificial intelligence models. By establishing a hierarchical association topology and feature isolation mechanism between attributes, this invention enables precise control and targeted modification of specific design details such as materials, shapes, or processes without disrupting the overall design consistency.

[0018] 2. This invention proposes a strong constraint attribute locking mechanism that can automatically identify key asset areas (such as brand logos, core functional components, and main stone geometry) in a design object based on semantic parsing results, and generate high-precision protective masks. During the generation process, this mechanism forcibly preserves the pixel-level original state or structural logic of the aforementioned key features, effectively avoiding the risk of random tampering during the artificial intelligence generation process, and ensuring that the generated variant solutions strictly comply with brand visual specifications and product functional requirements, thus possessing the feasibility for application in actual industrial design and production processes.

[0019] 3. This invention introduces an intelligent decision-making mechanism based on a variant programming algorithm. It calculates the combination space of variable attributes and uses a maximum design distance algorithm to filter solutions. This mechanism changes the disordered and random blind generation mode of traditional generation methods, automatically eliminating highly similar or redundant solutions that violate design logic. This ensures that the output batch of variants has significant differences and rationality in visual style and attribute combinations. This not only enables systematic deduction of design solutions but also significantly reduces the manpower cost for designers to filter invalid solutions later, significantly improving the efficiency of design iteration and exploration. Attached Figure Description

[0020] 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.

[0021] Figure 1 This is a flowchart illustrating the industrial design variant generation method based on structured parametric model and feature decoupling of the present invention. Figure 2 This is a block diagram of the industrial design variant generation system based on structured parameter model and feature decoupling in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device 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] Figure 1This is a flowchart illustrating an industrial design variant generation method based on structured parametric models and feature decoupling, according to an embodiment of the present invention. It should be noted that if substantially the same result is achieved, the method of this application is not necessarily identical. Figure 1 The sequence of processes shown is limited.

[0024] Example 1: like Figure 1 The industrial design variant generation method shown, based on structured parametric models and feature decoupling, includes the following steps: In step S10, in response to a variant generation request for the target design object, an input image of the design object is obtained, and a pre-trained visual semantic parsing model is invoked to parse the input image and extract a set of design semantic units including structural layer, morphological layer and material process layer. In this embodiment of the disclosure: (a) Design of input data acquisition and standardization processing: The system first receives at least one input image of the design object, which can be any one or more of the following forms: Industrial design sketches; Product renderings have been completed; Photos of the actual product; Two-dimensional view exported from CAD software.

[0025] To ensure the stability of subsequent semantic parsing, the system performs uniform standardization processing on the input images, including: Image size normalization processing; Lighting and color distribution correction; Background weakening or masking of non-design areas.

[0026] This standardization process is only used to improve the accuracy of semantic parsing and does not involve any generation or modification of the design.

[0027] (ii) Invocation of semantic parsing model based on design domain: The system calls a visual semantic parsing model that has been pre-trained and fine-tuned for a specific industrial design field to parse the standardized design images.

[0028] Unlike traditional object detection or segmentation techniques, the parsing target of this model is not "object category," but rather the design of semantic units, specifically including: Identification of the existence of each functional component in the design object; Identification of the combination relationships between components; Identification of the design intent type carried by the component.

[0029] The model outputs not an image mask or classification label, but a set of semantic candidates.

[0030] (III) Designing a hierarchical deconstruction of semantic units: The system deconstructs the parsing results according to the preset semantic hierarchy rules, dividing them into at least the following levels: Structural layer semantic units: used to describe the basic components in a design object that cannot be changed arbitrarily, such as the main frame, load-bearing structure, and the location of functional modules.

[0031] Morphological semantic units: used to describe the geometric morphological features of each component, such as contour type, cross-sectional form, and proportional relationship.

[0032] Material and process layer semantic unit: used to describe material properties, surface treatment methods, processing characteristics, etc.

[0033] Style and visual layer semantic unit: used to describe the overall or partial design style orientation, decorative language and visual emotional characteristics.

[0034] Each semantic unit is assigned a clear semantic identifier and category identifier for subsequent parameter modeling.

[0035] (iv) Controllability labeling and independence judgment of semantic units: The system further performs a controllability analysis on each semantic unit to determine whether the semantic unit possesses the following characteristics: Can it be changed independently without destroying the overall structure? Is it permissible for the designer to make individual adjustments during the industrial design process? Does it have a strong dependency on other semantic units?

[0036] Based on the above judgment, the system labels the semantic unit as: Semantic units can be controlled independently; Restricted control semantic units; It is not recommended to change semantic units.

[0037] The annotation results directly affect the feasibility and security of subsequent variant generation.

[0038] (v) Output format of semantic-level feature deconstruction results: The final output of step S10 is not an image, but a set of semantic features for design, which includes at least: The name of each semantic unit; Semantic level; Description of the logical location within the design object; Description of dependencies with other semantic units; Controllability labeling information.

[0039] The output will be directly passed to step S20 for constructing a structured design parameter model.

[0040] In this embodiment of the disclosure: Example background: The input is a product rendering of a single main stone ring.

[0041] Implementation process: The system receives the ring rendering image and completes image standardization processing; The ring is analyzed using a semantic parsing model specific to the field of jewelry design. The system identifies and deconstructs the following semantic units: Structural layer semantic units: ring band, main stone support structure, setting position relationship; Morphological semantic units: ring band cross-sectional shape, main stone cut outline; Material layer semantic units: metal type, metal surface gloss state; Style layer semantic unit: The overall style orientation is "simple and modern".

[0042] The system further determines that the cut and size of the main stone are constrained semantic units. The cross-sectional shape of the ring arm is a semantically controllable unit; The metallic material can vary within a limited set.

[0043] Output: The system outputs a complete set of semantic features for the ring design, which is used for subsequent structured modeling and variant generation control.

[0044] In step S20, a structured parameter model is constructed based on the set of design semantic units, the features of the design object are decoupled, the visual features are mapped into mutually independent parameterized attributes, and a hierarchical relationship topology between the attributes is established. In this embodiment of the disclosure: (a) Generation of semantic features to parameter terms: The system receives the set of design semantic features output from step S10 and generates one or more parameterized descriptive terms for each semantic unit.

[0045] The mapping process follows these principles: A semantic unit — corresponds to at least one parameter item; Parameters must have a clear name, category, and value definition; The parameters do not describe the image, but rather the design decision space.

[0046] For example, the semantics of "ring arm shape" are converted to the parameter "ring arm cross-section type"; The semantic meaning of "overall style leans towards modern" is converted to the "design style category" parameter.

[0047] (ii) Categorized hierarchical modeling of parameter items: The system performs hierarchical categorization modeling on all generated parameter items, classifying them into at least the following categories: Structural parameters: Describe the structural relationships, component locations, and assembly methods of the design objects; It is usually directly related to functionality and engineering feasibility.

[0048] Morphological parameters: Describe the geometric contours, proportions, and straight-curve relationships of the components; The basic form that determines the appearance of the design.

[0049] Material and process parameters: Describe the material type, surface treatment method, and processing characteristics; It is directly related to manufacturing and cost.

[0050] Style and visual parameters: Describe the design style, decorative language, and visual orientation; Used to control the overall or partial aesthetic direction.

[0051] Through this hierarchical modeling, the system is able to apply different control strategies to different categories of parameters in subsequent steps.

[0052] (iii) Explicit definition of the parameter value space: For each parameter, the system does not simply record a "current value," but explicitly defines its value space, including: Discrete selectable sets (such as style type, material type); Continuous range (e.g., scale, size range); A hybrid form of enumeration and rule constraints.

[0053] The sources for defining this value space include: Industry design standards; Historical design database; Designer experience rules; The system's preset security boundaries.

[0054] This ensures that the parametric model does not rely on random generation, but always operates within a controllable design space.

[0055] (iv) Modeling of dependencies and constraints between parameters: The system further analyzes the logical and engineering dependencies between parameter items and explicitly records them in the model, including but not limited to: Master-slave relationship (a change in one parameter affects another parameter); Exclusion relationship (two parameters cannot simultaneously take certain values); Interlocking relationships (multiple parameters need to change in tandem).

[0056] This dependency relationship is not determined during the generation stage, but is solidified in advance at the model layer to avoid illegal combinations from the source.

[0057] (v) Unified storage format for structured design parameter models: After completing the above modeling, the system organizes all parameter items and their relationships into a unified structured design parameter model, characterized by: It can be directly parsed and calculated by the program; It can be repeatedly called and modified in subsequent steps; It can be used as an intermediate representation layer for generating instructions.

[0058] This model becomes the sole authoritative data source for all subsequent steps (constraint configuration, variant programming, and generative control).

[0059] In this embodiment of the disclosure: Example background: The input is a design rendering of a beverage packaging.

[0060] Implementation process description: 1. Semantic feature input: The semantic features output in step S10 include: Brand logo area; Product name text area; Main visual graphic area; Background color and texture style.

[0061] 2. Parametric model construction results: The system constructs the following structured design parameter model: (1) Brand equity layer parameters: Logo graphics: fixed parameters, cannot be changed; Product name text: Fixed parameters, cannot be changed; Brand standard color: Limited set of parameters.

[0062] (2) Product information layer parameters: Capacity label location: Limited options available; Function description text style: Variable parameters.

[0063] (3) Visual representation layer parameters: Background main color: Variable parameter (limited color scheme); Texture type: Discrete set parameters; Decorative graphic style: enumerated parameters.

[0064] 3. Example of parameter dependency: Brand standard color parameters—constrain the range of background main color values; Some texture types exclude highly saturated background colors; Decorative graphic style — linked to the overall design style parameters.

[0065] Output: The system outputs a complete and computable packaging design parameter model, which can be used for constraint setting and variant planning in subsequent steps.

[0066] In step S30, a constraint configuration instruction set by the user for the structured parameter model is received. The constraint configuration instruction divides each parameterized attribute into strong constraint attributes and variable attributes, locks the parameter values ​​of the strong constraint attributes, and defines the variation space of the variable attributes. In this embodiment of the disclosure: (a) Predefined and classified constraint state types: The system predefines multiple attribute constraint state types at the parameter model layer, including at least the following three categories: Strong constraint state, variable constraint state, and exclusive constraint state; Each constraint state has a clear engineering semantic and execution meaning, rather than a simple label.

[0067] (II) Implementation methods for configuring strong constraint attributes: 1. Rules for defining strongly constrained attributes: The system will automatically or manually mark parameters that meet any of the following conditions as strongly constrained attributes: It is directly related to brand identity and core product functions; Arbitrary modifications are not permitted in the industrial design process; Changes will cause the functionality or recognizability of the designed object to fail.

[0068] 2. Execution method of strong constraints: For parameter terms marked as strongly constrained, the system executes the following control strategy: In subsequent variant planning phases, this parameter should be prohibited from participating in combinatorial calculations; During the generation and execution phase, the design feature corresponding to this parameter is forcibly written into the "immutable region description" of the generation instruction; If the attribute is detected to be corrupted during the result verification phase, the generated result is determined to be invalid.

[0069] 3. Strongly constrained model writing method: Strong constraint states are explicitly written into the structured design parameter model as field-level immutable markers for direct use by subsequent modules.

[0070] (III) Implementation methods for constraint configuration of variable attributes: 1. Principles for limiting mutable attributes: The system does not allow "completely free change," but rather configures limits on the range of change for variable attributes, including: Optional discrete set; Range of numerical variation; Style category-restricted set.

[0071] 2. Change control mechanism for variable attributes: The system records the following information for each variable attribute: Whether participation in variant planning is allowed; Maximum range of change or level of change; Does it need to change in conjunction with other attributes?

[0072] In subsequent steps, the system performs variant planning only within this constraint to avoid generating meaningless or unmanufacturable solutions.

[0073] (iv) Implementation method for constraint configuration of exclusion attribute combinations: 1. The source of exclusionary relationships: Exclusionary constraints are not subjectively set, but rather derive from at least one of the following bases: Engineering feasibility rules; Aesthetic or design guidelines; Prohibited combinations for brands or industries; Historical design failures.

[0074] 2. Configuration method for exclusion rules: The system records exclusion relationships in the form of parameter pairs or parameter groups, for example: When attribute A has a value of X, attribute B is prohibited from having a value of Y; If attribute C is enabled, then attribute D must be disabled.

[0075] 3. Execution phase of exclusion constraints: During the variant programming phase, illegal combinations are directly pruned. During the instruction reorganization phase, conflicting descriptions should be avoided. During the result verification phase, it serves as one of the consistency criteria.

[0076] (v) Unified writing and version management of constraint states: After completing the constraint configuration, the system writes all constraint state information into the structured design parameter model and records it: Constraint configuration source (user settings / system rules); Constraint effective time; Constraint version number.

[0077] This mechanism ensures that the constraint logic is reproducible, auditable, and iterative, avoiding uncontrollable behavior caused by "temporary rules".

[0078] In this embodiment of the disclosure, (taking a jewelry ring design as an example) Example background: The input is a single main stone ring design, and steps S10 and S20 have been completed.

[0079] Implementation process description: 1. Strong constraint attribute configuration: The system configures the following parameters as strong constraints: Main stone shape (round cut); Main stone color (colorless and transparent); The proportions of the main stone size.

[0080] Reason: The above attributes are directly related to the core value of the product and the wearing experience, and cannot be compromised in variations.

[0081] 2. Variable attribute configuration: The system configures the following parameters as variable attributes: Ring arm cross-sectional shape (circular / square / flat); Ring band surface finish (mirror / brushed / matte); The presence or absence of decorative elements on the ring arm.

[0082] It also limits the range of variations and prohibits combinations that exceed manufacturing specifications.

[0083] 3. Exclusion constraint configuration: The system is configured with the following exclusion rules: If the metal material is rose gold, it is forbidden to combine it with certain highly saturated colored gemstones; If the band has a minimalist style, complex patterns and decorations are prohibited.

[0084] Output: The system outputs a structured design parameter model with complete constraint state definitions, which will serve as the only valid input for variant programming in step S40.

[0085] In step S40, the design variant path is calculated in the variation space using a variant planning algorithm to generate a variant generation instruction set containing explicit differentiation descriptions; In this embodiment of the disclosure: (a) Extraction and grouping of variable attribute sets: The system first automatically extracts the attribute items marked as "variable state" in step S30 from the structured design parameter model, and groups them according to their design semantics, for example: Appearance-related attribute groups; Style-related attribute groups; Material and surface treatment properties group; Decoration and detail attribute group.

[0086] By grouping and processing, the system avoids the disordered combination of attributes at different semantic levels, thus preventing design logic chaos from the source.

[0087] (II) Establishing the Design Orientation Dimension: Based on an industrial design knowledge base and historical design data, the system establishes several design orientation dimensions for variable attributes, such as style orientation dimensions (minimalist, retro, modern, and decorative). Visual complexity orientation; Industrial and decorative aesthetics; Emotional expression orientation (rational, gentle, tense).

[0088] Each variant scheme must form a relatively stable and consistent combination of orientations across these orientation dimensions, rather than conflicting with each other.

[0089] (III) Generation mechanism of differentiated variant paths: The system, based on the premise of "consistency in design orientation," performs combined planning for variable attributes. Specific execution methods include: Each variant path selects only one set of core orientations; The values ​​of all variable attributes must match this orientation; Attribute values ​​with conflicting styles or visuals are not allowed in the same path.

[0090] Through this mechanism, the system generates not a "parameter permutation and combination", but a solution path with a clear design direction.

[0091] (iv) Variation assessment and screening: To avoid generating multiple highly similar solutions, the system performs a difference assessment on the planned variant paths. The assessment includes: Are the core attribute values ​​duplicated? The degree of change in the overall shape and outline; Distinctiveness of style expression; Distinguishability at the user perception level.

[0092] When two variant paths are highly similar in key attributes and overall design orientation, the system automatically retains the more representative one and eliminates redundant solutions.

[0093] (v) Pruning of illegal or low-value combinations: During the variant programming process, the system invokes the exclusion constraint rules configured in step S30 in real time to prune the following cases: Combinations of attributes that violate exclusion rules; Combinations that do not violate constraints but whose design expression is unclear; A combination that lacks rationality in terms of manufacturing, wearing, or use.

[0094] The pruning process is completed during the planning phase, rather than being remedied after generation.

[0095] (vi) Structured output of variant programming results: After the planning is completed, the system outputs a set of significantly different and controlled-number variant solutions, each of which includes: A clear design orientation is stated; The corresponding set of variable attribute values; A complete parameter structure that maintains consistency with strong constraint properties.

[0096] These results are output in the form of structured data, providing deterministic input for subsequent generation and execution.

[0097] In this embodiment of the disclosure, (taking a jewelry ring design as an example) Example premise: The following constraint configuration has been completed in step S30: The shape and size of the main stone are strictly constrained; The shape, surface finish, and degree of decoration of the ring band are variable attributes; Several exclusion combination rules have been established.

[0098] The implementation process of step S40: 1. Extracting mutable properties The system extracts the following variable attributes: Ring arm cross-sectional shape; Ring arm surface finish; Complexity of decorative elements.

[0099] 2. Establish a design orientation The system establishes three clear design orientations: Minimalist approach; Retro style; Modern industrial orientation.

[0100] 3. Generate differential variant paths: The system plans three valid variant paths: Option 1: Minimalist approach; the ring band has a geometric cross-section and a mirror finish, without any decorative elements; Option 2: Vintage Style: The ring band features a softly curved cross-section and a finely textured surface with subtle embellishments; Option 3: Modern Industrial Approach: The ring band has a simple structure, uses a matte metallic texture, and has very few decorative elements.

[0101] 4. Difference Filtering Results: The system confirmed that the above three paths have significant differences in appearance, style and user perception, and all of them were retained and output.

[0102] In step S50, based on the variant generation instruction set, the generative model is driven to redraw or partially generate the design object, and output an industrial design variant image that satisfies the constraint configuration instructions.

[0103] In the disclosed embodiments: (a) Hierarchical decomposition of generation instructions: The system first breaks down the generated instructions into at least the following three logical levels: Global immutable constraint layer, local region control layer, and variant difference description layer; The scope and modification permissions of each layer are predefined in the system, and the model generation must be executed in sequence.

[0104] (ii) Construction method of immutable constraint description: 1. Content Source: The description of immutable constraints is directly derived from the attribute items marked as strong constraints in step S30, including but not limited to: Core structural form; Key proportional relationships; Core functions or identifying features.

[0105] 2. Implementation method: The system writes these attributes into the generation command in a "do not modify" manner and clearly defines their scope of application, for example: Clearly specify which structural regions must remain unchanged; It is explicitly required that the generated results maintain consistency with the input structural features.

[0106] 3. Technical effects: This layer of instructions has the highest priority during execution, and no subsequent instructions may override or weaken the constraints of this layer.

[0107] (III) Organization of local area control information: 1. Basis for regional division: The system, based on a structured design parametric model, divides the design object into multiple functional or aesthetic areas, for example: Main structural area; Decorative areas; Auxiliary structural region; Background or non-core areas.

[0108] 2. Control method: The system specifies the following for different regions: Does the generated model allow for appearance changes? Allowed types of changes (shape, texture, process, etc.); The magnitude of the change.

[0109] These area control information are written separately into the corresponding sections of the generation instruction.

[0110] 3. Technical effects: When the generated model is executed, it can only make changes within the permitted areas to avoid "the whole being redrawn" or "the core structure being destroyed".

[0111] (iv) Generation of differentiated descriptions based on variant paths: 1. Source of information on differences: The differentiated description is strictly derived from a specific variant path determined in step S40, including: A clear design orientation; The corresponding variable attribute values; The combined results have passed the constraint verification.

[0112] 2. Implementation method: The system transforms this difference information into specific generation requirements for local regions, for example: Stylistic expression tendencies; Surface treatment methods; Decorative complexity level.

[0113] Each variant path corresponds to a separate set of differentiated description instructions.

[0114] (v) Instruction conflict detection and reassembly verification: Before generating the final output of the instruction, the system performs a structured check to ensure that: Differential descriptions do not cover immutable constraints; Local area control did not violate the exclusion rule; There are no self-contradictory descriptions within the same instruction layer.

[0115] If a conflict is detected, the system automatically backtracks to step S40 or step S30 to make corrections, instead of forcibly generating the new system.

[0116] (vi) Generate structured output of instructions: Ultimately, the system outputs not a natural language prompt, but rather: A generation instruction structure consisting of multiple logical segments; Each section corresponds to a different level of control; It can be generated, parsed, and executed sequentially.

[0117] This instruction structure serves as the only valid input for generation and execution in step S60.

[0118] In this embodiment of the disclosure (taking a ring design as an example): Example premise: A variation of the "modern minimalist style" has been planned.

[0119] Example of the implementation process of step S50: 1. Description of immutable constraints: The system explicitly instructs: The shape, size, and setting position of the main stone must remain unchanged from the original design; The proportions of the main structure must not be changed.

[0120] 2. Local area control: System specifies: No visual modifications are permitted in the main stone area; Variations in shape and surface finish are permitted in the ring arm area; Non-core background areas can be softened.

[0121] 3. Differentiated Description: The system writes the following to the ring arm area: The overall style leans towards modern minimalism; The surface treatment features low reflection and a simple, textured finish. Complex patterns or decorative protrusions are prohibited.

[0122] Final result: When the generative model is executed: The main stone and core structure must be strictly preserved; The modern minimalist style is only reflected in the band area; It will not generate random decorations that are unrelated to the planned path.

[0123] Step S60: If the input image is a design object containing a brand logo, then when constructing the structured parameter model, the brand logo is automatically identified as the highest priority strong constraint attribute. During the variant planning phase, if a conflict is detected between the generated variant instruction set and the visual specifications of the brand identity, a rollback mechanism is automatically triggered or the variant instruction set is corrected.

[0124] Example 2: like Figure 2As shown, to solve the above-mentioned technical problems, based on Embodiment 1, another technical solution adopted in this application is: an industrial design variant generation system based on structured parameter model and feature decoupling, comprising: The semantic parsing module is configured to acquire the input image of the design object and call a pre-trained visual semantic parsing model to parse the input image and extract a set of design semantic units including structural layer, morphological layer and material process layer. The parameter model construction module is configured to construct a structured parameter model based on the set of design semantic units, decouple the features of the design object, map the visual features into mutually independent parameterized attributes, and establish a hierarchical relationship topology between the attributes. The constraint control module is configured to receive constraint configuration instructions set by the user for the structured parameter model. The constraint configuration instructions divide each parameterized attribute into strongly constrained attributes and variable attributes, lock the parameter values ​​of the strongly constrained attributes, and define the variation space of the variable attributes. The planning module is configured to use a variant planning algorithm to calculate design variant paths within the variant space and generate a variant generation instruction set containing explicit differential descriptions. The generation execution module is configured to drive the generative model to redraw or partially generate the design object based on the variant generation instruction set, and output an industrial design variant image that satisfies the constraint configuration instructions.

[0125] For other details regarding the implementation techniques of each module in the above embodiments, please refer to the description in the above embodiments of the method and system for generating industrial design variants based on structured parameter models and feature decoupling, which will not be repeated here.

[0126] 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-type 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.

[0127] Example 3: To address the aforementioned technical problems, based on Embodiment 1, this application adopts another technical solution: an industrial design variant generation method based on structured parametric models and feature decoupling. This method transforms unstructured image information into structured parametric models, achieving refined control and intelligent variant generation of industrial design objects. The method specifically includes the following steps: Step S1: Design the acquisition and standardization processing of input data: First, the system receives a variant generation request for the target design object. The input data is usually one or more images, in forms including but not limited to industrial design sketches, rendered images of completed products, photographs of physical products, or 2D views exported from CAD software.

[0128] To ensure the stability of subsequent semantic parsing, the system performs uniform standardization processing on the input images. Specifically, this includes: Image size normalization: Adjust input images of different resolutions to a standard size (e.g., 1024x1024 pixels) to fit the input of the semantic parsing model, while maintaining the original aspect ratio and filling blank areas.

[0129] Illumination and color distribution correction: Through histogram equalization or white balance algorithms, eliminate color casts or overexposure / underexposure problems caused by the shooting environment in the input image, and restore the inherent color of the design object.

[0130] Background weakening: A saliency detection algorithm is used to identify the main subject area. Background areas outside the design are blurred, lightened, or replaced with a solid color to reduce background noise interference with subsequent analysis. It should be noted that this standardization process is only used to improve analysis accuracy and does not involve generating or modifying any design content.

[0131] Step S2: Semantic parsing and hierarchical deconstruction based on the design domain: The system calls a pre-trained and fine-tuned visual semantic parsing model for specific industrial design fields (such as jewelry, packaging, and 3C products) to parse the standardized design images.

[0132] Unlike traditional object detection methods that only identify "object categories," the model in this invention aims to extract "design semantic units." The parsing process includes: Component identification: Identify the existence of each functional component in the design object (e.g., the band, setting, and main stone of a ring; the cap, body, and label of a package).

[0133] Relationship identification: Identify the spatial position and combination relationship between components (e.g., the setting is located above the band, and the main stone is set inside the setting).

[0134] Intent recognition: Identify the type of design intent carried by a component (e.g., decorative lines, functional buttons).

[0135] Subsequently, the system deconstructs the parsing results into a set of design semantic units according to preset hierarchical rules, dividing them into at least the following three levels: Structural layer semantic units: These describe the fundamental, non-changeable components of a design object, such as the main frame, load-bearing structure, and the location of functional modules. This is the skeleton of the product.

[0136] Morphological semantic units: used to describe the geometric morphological characteristics of each component, such as outline type (circle, square), cross-sectional form, and proportional relationship, which is the product's shape.

[0137] Material and process layer semantic unit: used to describe material properties (gold, silver, plastic), surface treatment methods (brushed, polished, matte), and processing characteristics. This is the texture of the product.

[0138] Step S3: Constructing a structured parametric model and decoupling features: Based on the extracted set of design semantic units, the system constructs a "structured parametric model." This model serves as a bridge connecting visual images and generative control.

[0139] Specifically, the system transforms the design semantic units into parameterized descriptions in key-value pair form, forming an attribute list.

[0140] For example, for a ring, the attribute list might include: {Main stone type: diamond, main stone cut: round brilliant, band material: 18K gold, band finish: polished}.

[0141] During this process, the system performs the core feature decoupling operation. Feature decoupling refers to separating visually entangled features into independent parameterized attributes. For example, in the original image, "golden curved ring arm" is a whole visual information, which is decoupled and split into two independent attributes: "shape: curved" and "material: gold".

[0142] Simultaneously, the system establishes a hierarchical topology for the relationships between attributes. This topology defines the dependencies between attributes. For example, the "setting size" parameter depends on the "main stone size" parameter; the "surface texture" parameter is attached to the "ring band shape" parameter. By establishing this topological relationship, it can be ensured that when a certain attribute is modified subsequently, the system can automatically determine whether other attributes need to be adjusted accordingly, or whether physical constraints have been violated.

[0143] Step S4: Constraint Configuration and Strong / Weak Constraint Partitioning: The system provides an interactive interface that displays the constructed structured parametric model. Users (designers) use this interface to set constraint configuration commands.

[0144] The system supports dividing each parameterized attribute into two categories: Strongly constrained attributes: Features that the user specifies must remain strictly unchanged. Examples include brand logos, core functional components of a product, and the shape of the main stone. The system will automatically "lock" the feature vectors or image regions corresponding to these attributes.

[0145] Variable attributes: Features that users are allowed to change. Users can further define their variation space, for example, specifying that the "ring style" can vary between "minimalist, retro, and organic forms", or the "material" can vary between "rose gold and white gold".

[0146] In response to user configuration, the system automatically freezes the feature data corresponding to strongly constrained attributes, keeping them constant during subsequent generation processes, and only granting access to adjust variable attributes.

[0147] Step S5: AIPlanner variant planning and instruction generation: After obtaining the constraint configuration, the system does not directly perform random generation, but instead launches an intelligent planner to perform variant planning. This step aims to solve the problem of "homogeneous" or "illogical" generated results.

[0148] AIPlanner's execution logic includes: Spatial computation: Computing the space of all possible combinations of variable properties.

[0149] Design Distance Filtering: Using a design distance maximization algorithm, candidate attribute combinations with significant visual differences are filtered in the combination space. For example, if two schemes are generated, scheme A is "frosted and thin ring arm" and scheme B is "frosted and slightly coarse ring arm". Since the difference between the two is too small, AIPlanner will eliminate one of them and retain scheme C, "highlight and irregular ring arm", which has a greater difference.

[0150] Logical filtering: Combines preset style compatibility rules or industrial design logic to eliminate unreasonable combinations (for example, "wood material" combined with "casting process" may be considered unreasonable and eliminated).

[0151] Instruction synthesis: Each set of filtered attributes is converted into a natural language description or control code that the generative model can understand. Simultaneously, descriptions of locked, strongly constrained attributes are also incorporated to form a complete variant generation instruction set.

[0152] Each instruction in the instruction set contains: "Keep the [strong constraint description] unchanged, and modify the [variable attribute] to the [specific differential description]".

[0153] Step S6: Region Locking and Generation Execution: Based on the variant generation instruction set, the system drives generative models (such as StableDiffusion and ControlNet) to perform the final image generation.

[0154] To ensure that strong constraint properties are not violated, the system employs a region mapping and locking mechanism: Region mapping: Based on the structured parametric model, the design object is divided into core structural regions, key functional regions, and variable shape regions on the image.

[0155] Mask generation: Generate high-precision pixel-level masks for regions with strong constraint attributes (such as the main stone region).

[0156] Locking and Redrawing: During the generation process, the system implements a locking strategy for the areas covered by the mask, prohibiting the model from redrawing pixels or structurally modifying them. Pixel generation is guided only within the "variable shape region" based on the differentiated descriptions in the instruction set.

[0157] Ultimately, the system outputs a series of industrial design variant images that retain the core design assets while presenting diverse design styles across specified dimensions.

[0158] Example 4: like Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the electronic device in the embodiment of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0159] like Figure 3 As shown, an electronic device includes a processor, a memory, and a communication interface. The memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned industrial design variant generation method and system based on structured parameter model and feature decoupling of the various embodiments of this disclosure. The electronic device can exchange data with other devices or systems through the communication interface, enabling real-time updating and sharing of drug information.

[0160] The processor in the aforementioned electronic device serves as its core, responsible for executing the computer program stored in the memory to implement various functions of the paperless conference terminal's intelligent interaction method. The processor can employ a high-performance multi-core CPU or a dedicated chip to meet the demands of complex calculations and real-time processing. The memory stores the operating system, applications, data, and computer programs. In this embodiment, the memory stores the computer program implementing the paperless conference terminal's intelligent interaction method. The memory can be RAM, ROM, Flash memory, or other types of non-volatile memory. The communication interface connects the electronic device to other devices or networks, enabling data transmission and exchange. In this embodiment, the communication interface supports multiple communication protocols and interface standards, such as Wi-Fi, Bluetooth, USB, and Ethernet, to meet communication needs in different scenarios.

[0161] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0162] Example 5: According to an embodiment of the present disclosure, a computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the various functions of the industrial design variant generation method and system based on structured parameter model and feature decoupling of the foregoing embodiments of the present disclosure.

[0163] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0164] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0165] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating industrial design variants based on structured parametric models and feature decoupling, characterized in that, Includes the following steps: In response to a variant generation request for a target design object, an input image of the design object is obtained, and a pre-trained visual semantic parsing model is invoked to parse the input image and extract a set of design semantic units including structural, morphological, and material / process layers. Based on the set of design semantic units, a structured parametric model is constructed to decouple the features of the design object, map the visual features into mutually independent parametric attributes, and establish a hierarchical relationship topology between the attributes. The system receives constraint configuration instructions set by the user for the structured parameter model. The constraint configuration instructions divide each parameterized attribute into strongly constrained attributes and variable attributes, lock the parameter values ​​of the strongly constrained attributes, and define the variation space of the variable attributes. The variant planning algorithm is used to calculate the design variant path in the variant space, generating a variant generation instruction set containing explicit differentiation descriptions; Based on the variant generation instruction set, the generative model is driven to redraw or partially generate the design object, and output an industrial design variant image that satisfies the constraint configuration instructions.

2. The method according to claim 1, characterized in that, The process of calling a pre-trained visual semantic parsing model to parse the input image and extract a set of semantic units includes: The input image is normalized and its background is weakened. The visual semantic parsing model is used to identify the functional components and their combination relationships in the design object; According to the preset hierarchical rules, the parsing results are deconstructed into design semantic units. The design semantic units of the structure layer describe the basic components and their positions, the design semantic units of the morphology layer describe the geometric contours and proportions, and the design semantic units of the material and process layer describe the surface textures and physical properties.

3. The method according to claim 1, characterized in that, The construction of a structured parametric model based on the set of design semantic units, and the decoupling of the features of the design object, includes: The design semantic unit is transformed into a parameterized description in key-value pair form, forming an attribute list; Based on the logical structure of the design object, the attribute list is divided into a core structure layer, a component form layer, and a visual presentation layer. Establish a feature isolation mechanism to ensure that any adjustment to the value of any parameterized attribute affects the corresponding local features.

4. The method according to claim 1, characterized in that, The step of receiving the constraint configuration instructions set by the user for the structured parameter model includes: Provide an interactive interface to display the structured parameter model, and in response to user selections, mark the brand asset characteristics or core functional structure of the design object as strongly constrained attributes; In response to user adjustments, set the range of values ​​or style preferences for variable attributes; Upon receiving the constraint configuration, the system freezes the feature vector corresponding to the strong constraint attribute.

5. The method according to claim 4, characterized in that, The calculation of design variant paths within the variation space using the variant programming algorithm includes: Calculate the variation space of the variable attribute and select attribute combinations based on the maximum design distance algorithm; Combined with preset style compatibility rules; The filtered attribute combinations are converted into differential descriptions in natural language form, and combined with the descriptions of strongly constrained attributes, to synthesize the variant generation instruction set.

6. The method according to claim 5, characterized in that, The process of driving the generative model to redraw or partially generate the design object based on the variant generation instruction set includes: Based on the structured parameter model and the constraint configuration instructions, a region mapping relationship is established, and the design object is divided into a core structural region, a functional key region, and a variable shape region. Generate a region mask corresponding to the strong constraint attribute, and execute a locking strategy on the pixel region covered by the region mask during the generation process; Within the variable shape area, pixel generation is guided by the differentiated description in the variant generation instruction set.

7. The method according to claim 1, characterized in that, The method further includes: If the input image is a design object containing a brand logo, then when constructing the structured parametric model, the brand logo will be automatically identified as the highest priority strong constraint attribute. During the variant planning phase, if a conflict is detected between the generated variant instruction set and the visual specifications of the brand identity, a rollback mechanism is automatically triggered or the variant instruction set is corrected.

8. An industrial design variant generation system based on structured parametric models and feature decoupling, applied to the industrial design variant generation method based on structured parametric models and feature decoupling as described in any one of claims 1-7, characterized in that, include: The semantic parsing module is configured to acquire the input image of the design object and call a pre-trained visual semantic parsing model to parse the input image and extract a set of design semantic units including structural layer, morphological layer and material process layer. The parameter model construction module is configured to construct a structured parameter model based on the set of design semantic units, decouple the features of the design object, map the visual features into mutually independent parameterized attributes, and establish a hierarchical relationship topology between the attributes. The constraint control module is configured to receive constraint configuration instructions set by the user for the structured parameter model. The constraint configuration instructions divide each parameterized attribute into strongly constrained attributes and variable attributes, lock the parameter values ​​of the strongly constrained attributes, and define the variation space of the variable attributes. The planning module is configured to use a variant planning algorithm to calculate design variant paths within the variant space and generate a variant generation instruction set containing explicit differential descriptions. The generation execution module is configured to drive the generative model to redraw or partially generate the design object based on the variant generation instruction set, and output an industrial design variant image that satisfies the constraint configuration instructions.

9. An electronic device, characterized in that, The system includes a processor, a memory, and a communication interface. The memory stores a computer program, and when the processor executes the computer program, it implements the industrial design variant generation method based on structured parameter model and feature decoupling as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the industrial design variant generation method based on structured parametric model and feature decoupling as described in any one of claims 1 to 7.

Images