Intelligent grading method for sports uniform patterns using neural networks and computing device

Abstract

The embodiment of the present disclosure provides a sports uniform pattern intelligent grading method using a neural network and a computing device, and relates to the technical field of computer-aided design of sports uniforms. Embodiments are disclosed for realizing intelligent grading of sports uniform patterns and text on multi-size cutting pieces through cooperation of multiple neural networks. The method comprises: receiving input, the input comprising an initial design of a sports uniform and a multi-source rule set; processing the multi-source rule set through a rule reasoning neural network, automatically deciding based on rule priority and current design context in the case of multi-rule conflict, and outputting unified design constraints; through a geometry and visual optimization neural network, performing visual equivalent scaling, deformation and optical compensation on the pattern and text based on the initial design, the unified design constraints and the producibility process constraints, and generating a multi-size optimization design scheme; through a consistency correction neural network, automatically detecting, correcting and checking the consistency of the design scheme, and outputting a piece production file.

Description

Technical Field

[0001] This disclosure generally relates to the fields of computer-aided design, artificial intelligence, and digital manufacturing, and particularly to a method and computing device for intelligent scaling of sports uniform patterns on multi-size fabric pieces using a trained neural network to achieve intelligent adaptation, optimization, and consistency correction of sports uniform patterns. Background Technology

[0002] In the production process of professional sports uniforms, "design pattern rendering" is a crucial step. This step requires precisely adapting the designer's finalized graphic visual scheme (including team logos, sponsor logos, stripes, and fixed text) to each garment piece (such as the front and back panels, sleeves, etc.) across the entire size range (e.g., XS to 5XL). Currently, this process mainly relies on manual operation by designers, resulting in low efficiency, poor visual consistency, high compliance risks, and difficulty in ensuring pattern continuity across garment pieces. Existing automated tools lack the ability to handle complex and fuzzy constraints and perform advanced visual optimization and automatic correction. Summary of the Invention

[0003] In view of the above problems, this disclosure provides a method and computing device for intelligent gradation of sports uniform patterns using neural networks to solve the above technical problems.

[0004] In a first aspect, embodiments of this disclosure provide an intelligent scaling method for sports uniform patterns, executed by a computing device. The method includes: receiving input, the input including an initial design of the sports uniform and a multi-source rule set; processing the multi-source rule set through a rule inference neural network, automatically making decisions based on rule priority and the current design context when multiple rules conflict, and outputting unified design constraints after conflict resolution; using a geometric and visual optimization neural network, based on the initial design, the unified design constraints, and manufacturability process constraints, performing visually equivalent scaling, deformation, and optical compensation on the proportions, positions, and deformations of the inherent patterns and text of the sports uniform to maintain visual consistency under non-uniform size variations, generating multiple optimized design schemes covering multiple different sizes; using a consistency correction neural network, automatically detecting and correcting the multiple optimized design schemes, the correction including correcting color shifts, uneven margins, character breaks, or misalignments at seams across fabric pieces, and performing cross-fabric pattern continuity correction and full-size batch consistency verification; and outputting verified production files corresponding to each size of fabric piece.

[0005] Optionally, the rule reasoning neural network is a model based on an attention mechanism or a graph neural network, which is trained to learn the implicit relationships and conflict resolution patterns among multi-source rules and adjust the decision weights according to the current design context.

[0006] Optionally, the manufacturability process constraints include a quantitative representation of process requirements, provided by a manufacturability constraint neural network, wherein the manufacturability constraint neural network is trained to transform the process requirements into constraints that the geometric and visual optimization neural network can handle. The process requirements include piece splicing relationships, seam direction, thermal transfer or screen printing process tolerances, color difference ranges, and overprint avoidance requirements.

[0007] Optionally, the manufacturability constraint neural network is a convolutional neural network or a graph neural network, which is trained to predict the deformation, displacement or color deviation of the pattern that may occur in production based on the input pattern features, fabric material and process parameters, and quantifies the prediction results into constraints.

[0008] Optionally, the joint optimization processing of the geometric and visual optimization neural network is achieved by minimizing a multi-objective function, which simultaneously optimizes the readability score of the pattern and text, the consistency of the brand logo, and the safe distance of the design element relative to the edge of the fabric piece.

[0009] Optionally, the geometry and vision optimization neural network is a conditional generative adversarial network, a differentiable rendering pipeline, or a diffusion model, which is trained to directly output the coordinates, deformation, and attribute parameters of the optimized pattern and text under given design input, target size parameters, and various constraints.

[0010] Optionally, the method further includes: performing visual enhancement processing on the multiple optimized design schemes through an intelligent cascading neural network to optimize the visual equivalent scaling and deformation effect. The intelligent cascading neural network is a combination of a super-resolution network and a spatial transformation network, which is trained to learn a nonlinear mapping from a baseline size design to a target size design, and to enhance the details of small-size patterns and normalize the proportions of large-size patterns.

[0011] Optionally, the method further includes: automatically projecting the optimized full-body design onto the coordinate system of each piece corresponding to each size through a piece mapper, and preserving the alignment relationship at the seams and the integrity of the continuous pattern across the pieces during the projection process.

[0012] Optionally, multiple neural networks can be trained jointly or independently in an end-to-end or phased manner, with training data including historical design cases, corresponding multi-size production documents, rule documents, and production quality inspection reports.

[0013] Secondly, embodiments of this disclosure provide a computing device, including: one or more processors; and a memory storing computer-executable instructions, which, when executed by the one or more processors, implement the intelligent grading method for sports uniform patterns according to the first aspect of this disclosure.

[0014] In one or more embodiments, the disclosed method includes receiving input comprising an initial design and a multi-source rule set. The multi-source rule set is processed by a rule-based inference neural network, which automatically makes decisions based on rule priority and the current design context when multiple rules conflict, and outputs a unified design constraint after conflict resolution. A geometric and visual optimization neural network is used to perform visually equivalent scaling, deformation, and optical compensation on the proportions, positions, and deformations of the inherent patterns and text of the sportswear, based on the initial design, the unified design constraint, and manufacturability process constraints, to maintain visual consistency under non-uniform size variations, generating multiple optimized design schemes covering multiple different sizes. A consistency correction neural network automatically detects and corrects the multiple optimized design schemes, including correcting color shifts, uneven margins, character breaks, or misalignments at seams across fabric pieces, and performing cross-fabric pattern continuity correction and full-size batch consistency verification. Finally, a verified production file corresponding to each size fabric piece is output. This method, through the collaboration of multiple neural networks, receives the initial design and multi-source rules, automatically completes rule conflict decision-making, multi-constraint joint optimization, visual equivalent scaling and deformation, automatic correction and consistency verification, and finally outputs high-quality fragmented production files.

[0015] These or other aspects of this disclosure will become more apparent in the following description of embodiments. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 The architecture and data flow diagram of a sports uniform pattern intelligent cascading system according to one or more embodiments are shown.

[0018] Figure 2 A schematic diagram of an exemplary architecture for a rule-based reasoning neural network is shown.

[0019] Figure 3 A schematic diagram of an exemplary architecture of an intelligent hierarchical neural network is shown.

[0020] Figure 4 A flowchart of a method for intelligently grading sports uniform patterns according to one or more embodiments is shown.

[0021] Figure 5 A structural block diagram of an exemplary computing device that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0022] Embodiments of this disclosure are described in detail below, with examples of embodiments shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this disclosure, and should not be construed as limiting this disclosure.

[0023] To enable those skilled in the art to better understand the solutions disclosed herein, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0024] In this disclosure, it should be noted that relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0025] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0026] In the description of embodiments in this disclosure, terms such as "example" or "for example" are used to indicate exemplification, illustration, or description. Any embodiment or design described as "example" or "for example" in this disclosure is not to be construed as being more preferred or having more advantages than another embodiment or design. The use of terms such as "example" or "for example" is intended to present relative concepts in a clear manner.

[0027] Furthermore, in this disclosure, "multiple" refers to two or more. Therefore, in this disclosure, "multiple" can also be understood as "at least two." "At least one" can be understood as one or more, such as one, two, or more. For example, including at least one means including one, two, or more, and is not limited to which ones are included. For instance, including at least one of A, B, and C could mean including A, B, C, A and B, A and C, B and C, or A and B and C.

[0028] It should be noted that in this embodiment of the disclosure, "and / or" describes the relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. In addition, the character " / ", unless otherwise specified, generally indicates that the associated objects before and after it are in an "or" relationship.

[0029] It should be noted that in the embodiments of this disclosure, "connection" can be understood as electrical connection, and the connection between two electrical components can be a direct or indirect connection between the two electrical components. For example, the connection between A and B can be a direct connection between A and B, or an indirect connection between A and B through one or more other electrical components.

[0030] Figure 1 The architecture of a sports uniform pattern intelligent cascading system 100 according to an embodiment of the present disclosure is shown. The system 100 can be implemented on one or more computing devices, and its core is a collaborative workflow consisting of multiple specially trained neural networks. The system is configured to receive an initial design 101 and an associated multi-source rule set 102 as input, and after a series of intelligent processing steps, ultimately output a segmented production file 190 that can be directly used for production.

[0031] like Figure 1As shown, the system mainly includes a rule-based reasoning neural network 110, a geometry and visual optimization neural network 120, a manufacturability constraint neural network 130, a consistency correction neural network 140, an optional intelligent cascading neural network 150, and a piece mapper 160. Each component is connected through a clearly defined data interface. Specifically, the rule-based reasoning neural network 110 processes a multi-source rule set 102 and outputs a unified design constraint 111. The geometry and visual optimization neural network 120 receives the initial design 101, the unified design constraint 111, the target size parameters 103, and the manufacturability process constraints 131 provided by the manufacturability constraint neural network 130, generating a preliminary optimized design scheme 121 for multiple sizes. Optionally, the intelligent cascading neural network 150 can perform visual perception optimization on scheme 121 to obtain an enhanced scheme 151. The piece mapper 160 projects the full-body design scheme (121 or 151) onto the coordinate system of each independent pattern piece, generating a piece design scheme 161. Finally, the consistency correction neural network 140 automatically detects, corrects, and verifies the consistency of the sharding design scheme 161, and outputs the final sharding production file 190.

[0032] The rule-reasoning neural network 110 is configured to process multi-source, potentially conflicting design rule texts from international sports federations, regional leagues, clubs, and brands. Its core function is to understand rule semantics and, when conflicts occur, automatically make decisions based on learned priority rules and the current design context (e.g., specific event level, regional regulations), outputting a set of conflict-free, unified, and quantifiable design constraint vectors (unified design constraints 111). In one embodiment, such as... Figure 2 As shown, the neural network can employ a Transformer-based encoder architecture. The input layer 210 transforms textual rule clauses (e.g., "back number height not less than 25cm") into sequence feature vectors through word embeddings and positional encoding. The encoder 220 consists of multiple stacked Transformer layers, whose internal self-attention mechanism automatically learns the semantic relationships and potential conflicts between different rule clauses. The decision head 230 receives the encoder's output and, combined with the current design context (e.g., a context-encoded vector representing "UEFA Champions League"), generates the final unified design constraint vector. This vector, in a structured numerical form, precisely represents the boundary conditions such as position, size, and color that each pattern element (e.g., number, logo) must satisfy. In another embodiment, the neural network can be implemented using a graph neural network (GNN). In this model, each rule and its associated design elements are modeled as graph nodes, and the relationships between rules, such as references, overriding, and contradictions, are modeled as edges. Through message passing and aggregation mechanisms on the graph, the network can infer the final effective rule subgraph and extract the corresponding constraints.

[0033] The geometric and visual optimization neural network 120 is the core generator of the system. It is configured to intelligently scale, deform, and reposition patterns and text in the initial design under multiple constraints. Its inputs include the initial design, target size parameters, unified design constraints 111 output by the rule-based inference neural network 110, and manufacturability process constraints 131 quantified by the manufacturability constraint neural network 130. Its output is an optimized design scheme for each target size. The optimization objective of this neural network is achieved by minimizing a multi-objective function. This objective function simultaneously optimizes several key metrics: readability score, which evaluates the clarity and recognizability of patterns and text (especially numbers and names) at simulated match viewing distances based on a human visual perception model; brand identity consistency, which measures the consistency of the visual characteristics of core brand elements (such as team logos, specific stripes) with the original design or brand standards after scaling and deformation; and the safe distance of design elements relative to the fabric edge, which penalizes pattern elements that are too close to the fabric cutting edge to ensure a safety margin during production. In one embodiment, the neural network can be constructed as a differentiable rendering pipeline. First, design elements (pattern outlines, text paths) are parameterized. Then, these parameters are rendered as images or vector graphics using a differentiable rasterizer or vector renderer. The optimization process directly updates these design parameters using gradient descent to minimize the aforementioned multi-objective function, where the manufacturability constraint 131 is integrated as a penalty term into the objective function. In another embodiment, the neural network can employ a Conditional Generative Adversarial Network (CGAN). The generator directly generates optimized design data based on the initial design, size parameters, and constraint vectors; the discriminator determines whether the generated design simultaneously satisfies visual realism and all constraints. Alternatively, a diffusion model can also be used for this purpose, generating constrained designs through conditional guidance during the denoising process.

[0034] The manufacturability constraint neural network 130, acting as the system's "process knowledge base," transforms the specific requirements of downstream production processes (i.e., process requirements) into mathematical models and provides quantified manufacturability process constraints 131 to the geometric and visual optimization neural network 120. These process requirements include the splicing relationships between fabric pieces, seam direction, expected tolerances for heat transfer or screen printing, allowable color difference ranges (ΔE values), and areas to be avoided during color overprinting. In one embodiment, this neural network can be a convolutional neural network. Its inputs are local image patches of the design pattern, feature encodings of the fabric material, and specific process parameter encodings. After training, the network can output a probability map of the deformation displacement field that may occur in the region during production (e.g., predicting the possible (Δx, Δy) offset of each pixel after hot pressing) or the occurrence of specific defects (such as bleeding or misregistration). These predictions are directly quantified into constraint values. In another embodiment, a graph neural network can be used to simulate the complete production process chain from printing, cutting to sewing, modeling each process and its parameters as a time sequence diagram, predicting and quantifying the cumulative effect of errors in the process, thereby providing more systematic constraints.

[0035] The consistency correction neural network 140 is the final automatic quality inspection and repair module. It is configured to batch process the piecewise mapped design, automatically detecting defects such as color shifts, uneven margins, virtual breaks caused by characters precisely positioned at the seam line, and misalignment of patterns across fabric pieces at the seam, and executing automatic correction algorithms. For cross-fabric pattern continuity correction, in one embodiment, the neural network can employ a U-Net-structured image inpainting network. The input is images of corresponding pattern regions on the two fabric pieces to be sewn. The network is trained to generate a visually continuous complete pattern with pixel-level smooth transitions at the seam boundary, thereby compensating for preset process seam tolerances. In another embodiment for vector graphics processing, the neural network can employ a graph convolutional network. In this model, vector elements of the patterns on each fabric piece (such as control points of Bézier curves) are treated as graph nodes, and strong connections are established between corresponding nodes on both sides of the seam boundary. Through graph convolution operations, the position coordinates of these nodes are collaboratively adjusted to achieve geometrically precise alignment, thereby eliminating misalignment.

[0036] The Intelligent-Level Neural Network 150 is an optional enhancement module specifically designed for achieving finer visual equivalent scaling and deformation. It can be a combination of a super-resolution network and a spatial transformation network. Figure 3As shown, the deformation prediction subnetwork 310 learns the complex, nonlinear spatial transformation mapping from the baseline size to the target size. The super-resolution enhancement subnetwork 320 is specifically designed for downsampling (corresponding to smaller sizes), responsible for recovering and enhancing the fine details and texture clarity of the pattern from lower-resolution features to overcome the blurring problem caused by simple scaling. The two work together to ensure that designs of different sizes are visually equivalent, for example, by optically compensating for "thickening" the numbers in smaller sizes.

[0037] The panel mapper 160 can be implemented based on computational geometry. It utilizes a predefined digital model of the panel template and a seam relation table to establish a transformation relationship for each size from a unified "whole-body" coordinate system to the local coordinate system of each individual panel (front, back, sleeves, etc.). Its function is to automatically and accurately "project" the optimized whole-body design onto each panel, ensuring that during the projection process, continuous patterns spanning multiple panels maintain mathematical coordinate continuity and are precisely aligned at panel seam boundaries.

[0038] The aforementioned neural networks can be trained jointly or independently using an end-to-end or phased approach. In phased pre-training, each neural network can first be pre-trained independently on relevant tasks. For example, a rule-based reasoning neural network can be pre-trained on a legal text implication dataset; a productivity constraint neural network can be pre-trained on an industrial defect detection dataset; and an intelligent cascading neural network can be pre-trained on an image super-resolution dataset.

[0039] After acquiring system-level training data (including historical design cases, corresponding perfect multi-size production documents, relevant rule documents, process parameters, and quality inspection reports), joint training can be performed. Taking a simplified joint training framework involving neural networks 110, 120, 130, and 140 as an example, its forward propagation process is as follows: Input a training sample {initial design D, rule set R, process parameters P, target size S, true production document T}. Rule inference neural network 110 processes R and outputs constraint C. Manufacturability constraint neural network 130 processes (D, P) and outputs process risk prediction M. Geometric and visual optimization neural network 120 generates design scheme G based on (D, S, C, M). Consistency correction neural network 140 corrects G to obtain the final output G'.

[0040] The total loss L_total for joint training is a weighted sum of multiple losses used to drive parameter updates, including: reconstruction loss (L_recon), which compares the difference between G' and the ground truth T, such as using L1 loss or the mean squared error of control point coordinates; rule compliance loss (L_rule), which calculates the degree of violation of constraint C by G'; productiveability loss (L_prod), which inputs G' into network 130 to calculate its process risk score, encouraging low-risk designs; perceptual loss (L_perceptual), which uses a pre-trained network (such as VGG) to extract features from G' and T and calculate the differences, maintaining high-level visual consistency; and adversarial loss (L_adv) if a GAN architecture is used, to make the generated distribution approximate the real design distribution. The total loss can be expressed as: L_total = α*L_recon + β*L_rule + γ*L_prod + δ*L_perceptual + ε*L_adv. The gradient of L_total with respect to the trainable parameters of each neural network is calculated using the backpropagation algorithm, and the parameters are updated using an optimizer (such as Adam). End-to-end training allows the gradient to propagate back from the consistency correction neural network 140 all the way to the rule inference neural network 110, enabling all components to collaboratively optimize the final output quality.

[0041] Figure 4 A flowchart is shown for a method for intelligently grading sports uniform patterns according to one or more embodiments. The method can be executed by the aforementioned system 100.

[0042] Step S410: Receive input.

[0043] Specifically, the input includes the initial design of the sports uniform and a multi-source rule set. The initial design defines the inherent patterns, text, and initial layout of the uniform. In a specific implementation, the initial design is typically the inherent patterns, text, and initial layout of the uniform's base size. The multi-source rule set includes design specifications from different authoritative organizations, which may include, but are not limited to, multi-source, potentially conflicting design rule texts from international sports federations, regional leagues, clubs, and brands. In a specific implementation, rules related to the current design can be selected from a pre-set rule library, or rule texts related to the current design can be uploaded. This embodiment of the disclosure does not limit the input method.

[0044] Step S420: Process the multi-source rule set through a rule-based reasoning neural network.

[0045] Specifically, the rule-reasoning neural network parses the rule text and identifies potential rule conflicts. Based on its internally learned rule priority logic (e.g., international competition rules may take precedence over regional rules), and combined with the current task context information (such as the specified competition level), the rule-reasoning neural network automatically makes decisions, resolves conflicts, and outputs a unified, quantified set of design constraints. These design constraints clearly define the boundary values ​​of parameters such as size, position, and color that each design element must satisfy across all sizes.

[0046] In one embodiment, such as Figure 2 As shown, the rule-reasoning neural network can employ a Transformer-based encoder architecture. The input layer 210 transforms textual rule clauses (e.g., "back number height not less than 25cm") into sequence feature vectors through word embeddings and positional encoding. The encoder 220 consists of multiple stacked Transformer layers, whose internal self-attention mechanism automatically learns the semantic relationships and potential conflicts between different rule clauses. The decision head 230 receives the encoder's output and, combined with the current design context (e.g., a context-encoded vector representing "UEFA Champions League"), generates the final unified design constraint vector. This vector, in a structured numerical form, precisely represents the boundary conditions such as position, size, and color that each pattern element (e.g., number, logo) must satisfy.

[0047] In another embodiment, the rule-reasoning neural network can be implemented using a graph neural network (GNN). In this model, each rule and its associated design elements are modeled as graph nodes, and the relationships between rules, such as references, overriding, and contradictions, are modeled as edges. Through message passing and aggregation mechanisms on the graph, the network can infer the final effective rule subgraph and extract the corresponding constraints.

[0048] Step S430: Visual equivalent scaling, deformation, and optical compensation are performed using a geometric and visual optimization neural network.

[0049] Specifically, the geometry and vision optimization neural network receives the initial design, target size parameters, the unified design constraints, and manufacturability process constraints provided by the manufacturability constraint neural network. The core task of the geometry and vision optimization neural network is to perform visually equivalent scaling, deformation, and optical compensation. This means that scaling is not a simple geometrical transformation, but rather introduces nonlinear adjustments based on a human eye perception model. For example, for a small-sized (e.g., XS) team logo, the network might predict a scaling factor larger than the geometrical scaling ratio and might simultaneously thicken its outline lines to compensate for any loss of visual clarity and recognizability due to the reduced physical size, thus giving it an "equivalent" visual presence to the viewer compared to a base-sized (e.g., L) team logo. This process is achieved through joint optimization of the proportions, positions, and deformations of the pattern and text, ultimately generating multiple optimized design schemes covering all target sizes. In one specific implementation, as described above, the optimization process of the geometry and vision optimization neural network is driven by minimizing a multi-objective function. This function simultaneously evaluates and optimizes the generated design across multiple aspects: assessing the readability of patterns and text at simulated viewing distances; measuring the consistency of the brand logo before and after transformation; and ensuring that all design elements maintain sufficient safety distance from the edges of the cut pieces. Manufacturability constraints (such as printing tolerances) are also integrated into the optimization objectives to ensure the feasibility of the design.

[0050] The manufacturability process constraints include a quantitative representation of process requirements, provided by a manufacturability constraint neural network. This network is trained to transform the process requirements into constraints that the geometric and visual optimization neural network can handle. These process requirements include panel splicing relationships, seam direction, tolerances for heat transfer or screen printing, color difference ranges, and overprint avoidance requirements. In one specific implementation, as described above, the manufacturability constraint neural network 130 serves as the system's "process knowledge base," transforming the specific requirements of downstream production processes (i.e., process requirements) into mathematical models and providing quantified manufacturability process constraints 131 to the geometric and visual optimization neural network 120. These process requirements include panel splicing relationships, seam direction, expected tolerances for heat transfer or screen printing, allowable color difference ranges (ΔE values), and areas to be avoided during color overprinting. In one embodiment, this neural network can be a convolutional neural network. Its inputs are local image patches of the design pattern, feature encodings of the fabric material, and specific process parameter encodings. The network, once trained, can output a probability map of the deformation displacement field that may occur in the region during production (e.g., predicting the possible (Δx, Δy) offset of each pixel after hot pressing) or the occurrence of specific defects (such as bleeding or misregistration). These predictions are directly quantified into constraint values. In another embodiment, a graph neural network can be used to simulate the complete production process chain from printing and cutting to sewing, modeling each process and its parameters as a time sequence diagram, predicting and quantifying the cumulative effect of errors in the process, thereby providing more systematic constraints.

[0051] Step S440: Automatic detection and correction are performed using a consistency correction neural network.

[0052] Specifically, the consistency correction neural network receives optimized design schemes for all sizes (typically automatically projected onto the coordinate systems of each individual fabric piece by the piece mapper). This consistency correction neural network performs batch processing, and its core function is to automatically detect and correct various potential defects. These include, but are not limited to: color shifts due to differences in color management processes; uneven margins of the same pattern on different fabric pieces; visual character breaks that may occur when text is located precisely at the seam line of a fabric piece; and misalignment caused by inaccurate alignment of patterns across fabric pieces at the seam. Upon detecting a defect, the consistency correction neural network automatically applies trained correction algorithms, such as calibrating color values, fine-tuning margins, intelligently deforming and avoiding text across seams, and realigning patterns on both sides of the seam boundary, thereby performing continuity correction for patterns across fabric pieces. Simultaneously, the network compares the corrected schemes for all sizes to complete a full-size batch consistency check, ensuring a high degree of visual uniformity from the smallest to the largest size.

[0053] In some specific implementations, a piece mapper automatically projects the optimized full-body design onto the coordinate systems of each piece corresponding to each size, preserving the alignment at seams and the integrity of continuous patterns across pieces during the projection process. A consistency correction neural network automatically detects and corrects the optimized design schemes for all sizes. The piece mapper can be implemented based on computational geometry. It utilizes a predefined digital model of the piece template and a seam relationship table to establish a transformation relationship for each size from the unified coordinate system of the "whole body" to the local coordinate system of each individual piece (front, back, sleeves, etc.). Its function is to automatically and accurately "project" the optimized full-body design onto each piece, ensuring that continuous patterns across multiple pieces remain mathematically continuous and precisely aligned at the seam boundaries during the projection process.

[0054] Step S450: Output the piece production file. This outputs a piece production file that has passed all checks and corresponds one-to-one with each piece (e.g., front, back, sleeve) for each size. These files are typically in vector format (e.g., SVG, PDF), which can directly drive digital cutting, printing, and other production equipment, and include necessary process instructions.

[0055] Taking football uniforms as an example, a specific workflow is as follows: Figure 1 The system shown receives initial design drafts of a new away jersey (including the team crest on the chest, advertising stripes, and fixed text on the back) along with a multi-source rule set including UEFA Champions League design specifications, a football club's bylaws, and FIFA general rules. A rule-reasoning neural network 110 processes these rules and discovers that the club requires a back number height of 30cm, while UEFA stipulates a minimum of 28cm. Based on a pre-defined priority of "event operator specifications take precedence," the network decides to adopt 28cm and generates a unified design constraint vector that includes this constraint.

[0056] After the manufacturability constraint neural network 130 analysis and design, based on the selected "elastic knitted fabric" and "silicone printing" processes, it predicts that the pattern may shrink by a maximum of 0.2 mm after high-temperature pressing, and quantifies this as a process constraint.

[0057] The Geometric and Visual Optimization Neural Network 120 optimizes six sizes from XS to XXL. Based on the initial design, size parameters, unified constraints, and a 0.2mm shrinkage constraint, it intelligently deforms the team logo, stripes, and text. Its internal multi-objective function ensures the back number is clearly readable at a simulated 30-meter distance (optimizing readability), retains the core features of the team logo after deformation (maintaining brand consistency), keeps all elements at least 3mm away from the edge of the fabric piece (ensuring a safe distance), and pre-compensates for 0.2mm of manufacturing shrinkage. The final result is a full-body optimization scheme for all six sizes.

[0058] The panel mapper 160 accurately maps the full-body design for each size to the corresponding independent pattern files for the front, back, left sleeve, and right sleeve, ensuring that the side advertising stripe extending from the shoulder to the sleeve is perfectly aligned with the coordinates of the front and back pattern pieces at the armhole seam after mapping.

[0059] The intelligent-level neural network 150 performs post-processing on the generated scheme, enhancing the internal details of the small team logo in the XS code and slightly tightening the middle deformation of the wide advertising stripes in the XXL code to eliminate the visual "stretching" feeling and achieve true visual equivalence.

[0060] The consistency correction neural network 140 batch processes all pattern files for all six sizes. It automatically detects a deviation of ΔE=1.8 between the club abbreviation color on the back and neck of size M and the standard Pantone color chart, and corrects it; at the same time, it finds a potential misalignment risk of 0.1mm in the under-rib stripes of size L at the seam of the front and back side panels, and automatically completes the alignment operation.

[0061] Ultimately, the system outputs a set of layered vector production files containing 6 sizes, each size containing multiple cut pieces, which are error-free and ready for direct delivery to production.

[0062] This disclosure also provides a computing device including one or more processors and a memory storing computer-executable instructions that, when executed by the processor, implement the above-described method. Furthermore, a non-volatile computer-readable storage medium storing corresponding computer-executable instructions is also provided.

[0063] Exemplary embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to embodiments of this disclosure.

[0064] Exemplary embodiments of this disclosure also provide a computer program product, including a computer program, wherein, when executed by a processor of a computer, the computer program is used to cause the computer to perform a method according to an embodiment of this disclosure.

[0065] refer to Figure 5The present invention describes a structural block diagram of a computing device 500 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The computing device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The computing device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0066] like Figure 5 As shown, the computing device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the computing device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0067] Multiple components in the computing device 500 are connected to the I / O interface 505, including: an input unit 506, an output unit 507, a storage unit 508, and a communication unit 509. The input unit 506 can be any type of device capable of inputting information to the computing device 500. The input unit 506 can receive input numerical or character information and generate key signal inputs related to user settings and / or function control of the computing device. The output unit 507 can be any type of device capable of presenting information and may include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. The storage unit 508 may include, but is not limited to, a hard disk and an optical disk. The communication unit 509 allows the computing device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network interface cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0068] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model methods, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, the sports uniform pattern intelligent cascading method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on the computing device 500 via ROM 502 and / or communication unit 509. In some embodiments, the computing unit 501 can be configured to implement the sports uniform pattern intelligent cascading method by any other suitable means (e.g., by means of firmware).

[0069] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0070] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0071] As used in this disclosure, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0072] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0073] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0074] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0075] The above are merely preferred embodiments of this disclosure and are not intended to limit this disclosure in any way. Although this disclosure has been disclosed above with reference to preferred embodiments, it is not intended to limit this disclosure. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the technical solution of this disclosure. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of this disclosure without departing from the content of the technical solution of this disclosure shall still fall within the scope of the technical solution of this disclosure.

Claims

1. A method for intelligently printing patterns on sports uniforms, characterized in that, Performed by a computing device, the method includes: Receive input, which includes the initial design of the sports uniform and a multi-source rule set; The multi-source rule set is processed by a rule-based reasoning neural network. When multiple rules conflict, the network automatically makes decisions based on rule priority and the current design context, and outputs unified design constraints after conflict resolution. By using a geometric and visual optimization neural network, based on the initial design, the unified design constraints, and the manufacturability process constraints, the proportions, positions, and deformations of the inherent patterns and texts of the sports uniform are visually scaled, deformed, and optically compensated to maintain visual consistency under non-uniform size changes, generating multiple optimized design schemes covering multiple different sizes. The consistency correction neural network automatically detects and corrects the multiple optimized design schemes. The correction includes correcting color shift, uneven margins, broken characters, or misalignment at the seams across fabric pieces, and performs pattern continuity correction across fabric pieces and batch consistency verification for all sizes. Output the verified production files corresponding to each size of the cut piece.

2. The method according to claim 1, characterized in that, The rule-based reasoning neural network is a model based on an attention mechanism or a graph neural network. It is trained to learn the implicit relationships and conflict resolution patterns among multi-source rules and adjust the decision weights according to the current design context.

3. The method according to claim 1 or 2, characterized in that, The manufacturability process constraints include a quantitative representation of process requirements, provided by a manufacturability constraint neural network, wherein the manufacturability constraint neural network is trained to transform the process requirements into constraints that the geometric and visual optimization neural network can handle. The process requirements include piece splicing relationships, seam direction, thermal transfer or screen printing process tolerances, color difference ranges, and overprint avoidance requirements.

4. The method according to claim 3, characterized in that, The manufacturability constraint neural network is a convolutional neural network or a graph neural network, which is trained to predict the deformation, displacement or color deviation of the pattern that may occur during production based on the input pattern features, fabric material and process parameters, and quantifies the prediction results into constraints.

5. The method according to claim 1, characterized in that, The joint optimization processing of the geometric and visual optimization neural network is achieved by minimizing a multi-objective function, which simultaneously optimizes the readability score of the pattern and text, the consistency of the brand logo, and the safe distance of the design element relative to the edge of the fabric piece.

6. The method according to claim 1, characterized in that, The geometric and visual optimization neural network is a conditional generative adversarial network, a differentiable rendering pipeline, or a diffusion model. It is trained to directly output the coordinates, deformation, and attribute parameters of the optimized pattern and text under given design input, target size parameters, and various constraints.

7. The method according to claim 1, characterized in that, The method further includes: performing visual enhancement processing on the multiple optimized design schemes through an intelligent cascading neural network to optimize the visual equivalent scaling and deformation effect. The intelligent cascading neural network is a combination of a super-resolution network and a spatial transformation network. It is trained to learn a nonlinear mapping from a baseline size design to a target size design, and to enhance the details of small-size patterns and normalize the proportions of large-size patterns.

8. The method according to claim 1, characterized in that, The method further includes: automatically projecting the optimized full-body design onto the coordinate system of each piece corresponding to each size through a piece mapper, and preserving the alignment relationship at the seams and the integrity of the continuous pattern across the pieces during the projection process.

9. The method according to any one of claims 1-8, characterized in that, Multiple neural networks are trained jointly or independently in an end-to-end or phased manner. The training data includes historical design cases, corresponding multi-size production documents, rule documents, and production quality inspection reports.

10. A computing device, characterized in that, include: One or more processors; as well as A memory storing computer-executable instructions that, when executed by one or more processors, implement the method as described in any one of claims 1-9.

Images