Computer-implemented method for generating 3D clothing models
By automatically generating 3D clothing models through pattern recognition and extraction algorithms, the problem of cumbersome 2D to 3D conversion in clothing design is solved, realizing the automation and efficient production of clothing design.
Patent Information
- Application Number
- CN202110441418.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-04-30
- Filing Date
- 2021-04-23
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-04-23
AI Technical Summary
In existing technologies, the process of converting 2D patterns into 3D garment models is cumbersome, requires a lot of manual intervention, and lacks machine-readable post-processing information, resulting in low efficiency.
By employing pattern recognition and extraction algorithms, the system automatically identifies the clothing type and pose indicators of 2D sample pieces, generates 3D clothing models, and performs virtual post-processing based on clothing post-processing instructions, thereby achieving automatic generation and batch processing of clothing models.
It reduces working time and manpower, improves the efficiency of clothing design, realizes the automated conversion from 2D samples to 3D clothing models, and supports rapid iteration and mass production of clothing designs.
Smart Images

Figure CN113593002B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a computer-implemented method for generating 3D garment models. Specifically, a first method for automatically generating a first 3D garment model representing a first garment to be made from a first garment panel is disclosed, a second method for virtually finishing a second 3D garment model without post-processing or with default post-processing representing a second garment to be made is disclosed, and a third method for automatically generating a plurality of third 3D garment models in a batch process is disclosed, each third 3D garment model representing a third garment to be made from a third garment panel is disclosed. Background Art
[0002] In the garment industry, a common process is to import a garment pattern (i.e., a collection of 2D drawings of panels of the garment) into a computer-aided design (CAD) program on a computer so that the garment can be manually assembled using a computer mouse. Each panel is dragged onto an avatar and then roughly rotated and moved to its correct position on the avatar, which can be done with a 3D or 2D projection of the avatar. After roughly orienting and placing the pieces around the avatar, the edges need to be manually selected where the individual panels are connected. These processing steps are very tedious and take a long time.
[0003] It is also the current state of the art to specify the technical definition of a garment in two or more documents. One document is the garment pattern mentioned above (usually a 2D CAD model in DXF format), which contains the pattern shape and is used to cut out different pieces from the cloth and then assemble the garment. Another document is a file or a set of files containing assembly instructions and other details in a format that is easy for humans to understand (such as an Excel spreadsheet or a PDF file). These kinds of documents that define garments are usually called "technical packages". Human-readable instructions usually contain information about what fabrics to use, what colors, what thread types and stitch types, what post-processing (finish), what trims (buttons, zippers, hooks, etc.) to process, etc. The size measurements for different sizes of garments to be produced in the diagram can also be included in the "technical package". In contrast to the garment pattern, which depends on the document format (and whether it can be obtained digitally at all), the assembly instructions are not actually machine-readable because they are for human readers.
[0004] The situation in the apparel industry today is that there are millions of “tech packs” (garment patterns and assembly instructions), but very few corresponding 3D virtual garments (sometimes referred to as “Garment Digital Twins” (GDTs). This is because the conversion from what the industry already has (patterns and tech packs) to 3D models is manual and cumbersome. GDTs are used synonymously with 3D garment models in this article.
[0005] Often, it is desirable in the industry to provide clothing designers with a three-dimensional model of a garment so that they can check or verify the design of the garment. Clothing design is a very tedious process with many iterative steps, usually starting with a three-dimensional sketch, followed by an estimate of the 2D pattern shape based on experience and knowledge. A first sample is made, checked by the designer, who then corrects the garment pattern again based on his experience. A second sample verifies the changes, and the process can continue like this until a garment is made that meets the designer's expectations.
[0006] Furthermore, a clothing designer may want to update a clothing design by modifying an existing GDT on the three-dimensional virtual graphical user interface. Therefore, it is desirable not to start the design process from scratch but to use a predecessor model (e.g., a previous season design) as a basis.
[0007] There is growing interest in 3D "digital twins of garments" in the industry because they can be used, for example, to illustrate garments on a website without the need for expensive and elaborate photo shoots. Once a GDT is in hand, it can be displayed in different poses, colors, sizes, and worn by different mannequins or hangers. Likewise, a GDT can be illustrated in a virtual fitting room, where a customer can provide a three-dimensional representation of themselves and virtually "try on" several different garments in different styles and sizes. Summary of the invention
[0008] Therefore, the present invention provides a computer-implemented method that helps to overcome this huge unmet need.The method according to the present invention allows to reduce working time and manpower.
[0009] The present invention relates to a first computer-implemented method for automatically generating a first 3D garment model representing a first garment to be made from a first garment piece, the computer-implemented method utilizing a program code stored on a machine-readable medium or embodied as an electromagnetic wave, the computer-implemented method comprising the following steps: (a) providing one or more first documents, the one or more first documents comprising a plurality of first 2D pattern pieces representing the first garment piece, (b) utilizing a pattern recognition algorithm to determine: (i) a garment type indicator of at least one of the first 2D pattern pieces, the garment type indicator indicating the garment type to which the corresponding first garment piece belongs and being one of a plurality of garment type indicators stored on the machine-readable medium, and (ii) a garment piece pose indicator of each of the first 2D pattern pieces, the garment piece pose indicator indicating an estimated position and an estimated orientation of the corresponding garment piece within the first garment, and (c) generating a first 3D garment model based on the first 2D pattern piece, the at least one garment type indicator and the garment piece pose indicator.
[0010] In an embodiment, the first computer-implemented method further comprises the step of adding a default post-processing to the first 3D garment model based on at least one of the at least one garment type indicator and the garment piece pose indicator.
[0011] In an embodiment, the first computer-implemented method further comprises the step of generating a first 2D piece model for each of the first 2D pieces, wherein generating the first 3D garment model is based on overlaying a reference object with the first 2D piece model.
[0012] In an embodiment, the first computer-implemented method further comprises the following steps: (a) positioning and orienting first 2D garment piece models based on the piece pose indicators, in particular also based on the proportions of the 2D pieces relative to each other, (b) shaping each of the first 2D garment piece models into a first 3D garment piece model based on the at least one garment type indicator and the piece pose indicator while, before or after positioning and orienting the 2D garment piece models so that the first 3D garment piece models adapt to the shape of the reference object, and (c) generating a first 3D garment model by virtually assembling the first 3D garment piece models along their respective edges.
[0013] In an embodiment of the first computer-implemented method, in order to determine the garment type indicator and the garment piece pose indicator, the pattern recognition algorithm is configured to analyze at least one of the following items: (a) the shape of the 2D pattern piece, (b) the proportion of the 2D pattern piece relative to other 2D pattern pieces, (c) text information associated with the 2D pattern piece in the corresponding first document, (d) whether the 2D pattern piece is at least partially symmetrical, and (e) the amount of 2D pattern pieces provided.
[0014] The present invention also relates to a second computer-implemented method for virtually post-processing a second 3D clothing model representing a second garment to be produced without post-processing or with default post-processing, the computer-implemented method utilizing program code stored on a machine-readable medium or specifically implemented as electromagnetic waves, the computer-implemented method comprising the following steps: (a) providing a second 3D clothing model, (b) providing one or more second documents, the one or more second documents comprising human-readable clothing post-processing instructions for post-processing the second garment, (c) utilizing an extraction algorithm to determine at least one clothing post-processing feature from the one or more second documents, the clothing post-processing feature corresponding to an additional clothing component or clothing characteristic and indicating the position and orientation of the corresponding additional clothing component or clothing characteristic within the second garment, and (d) virtually post-processing the second 3D clothing model based on the determined at least one clothing post-processing feature.
[0015] In an embodiment of the second computer-implemented method, the step of virtually post-processing the second 3D garment model includes modifying at least a portion of the second 3D garment model.
[0016] In an embodiment of the second computer-implemented method, the garment post-processing feature corresponds to a fabric type of at least a portion of the second 3D garment model, wherein the step of modifying at least a portion of the second 3D garment model includes: assigning at least one of a texture and a color to at least a portion of the second 3D garment model based on the garment post-processing feature.
[0017] In an embodiment of the second computer-implemented method, the garment post-processing feature corresponds to a seam of the second garment, wherein the step of modifying at least a portion of the second 3D garment model includes assigning at least one of a thread color, a thread type, and a stitch pattern to the seam of the second garment.
[0018] In an embodiment of the second computer-implemented method, the step of virtually post-processing the second 3D garment model comprises adding at least one garment post-processing model to the second 3D garment model.
[0019] In an embodiment of the second computer-implemented method, the garment post-processing feature corresponds to a seam of the second garment, wherein the at least one garment post-processing model represents a stitch to be placed at a specific location within the second garment.
[0020] In an embodiment of the second computer-implemented method, the garment post-processing feature corresponds to a binding of the second garment, wherein the at least one garment post-processing model represents a hem to be placed at an opening of the second garment.
[0021] In an embodiment of the second computer-implemented method, the garment post-processing feature corresponds to a closure of a garment, wherein the at least one garment post-processing model represents at least one of the following items: buttons, button hooks, snaps, zippers, Velcro, magnetic fasteners, hooks, loops for hooks or buttons, cord locks, and laces.
[0022] In an embodiment of the second computer-implemented method, in order to determine the at least one garment post-processing feature, the extraction algorithm is configured to analyze the one or more second documents with respect to at least one of the following items: text information, image data, arrows, size data, color swatches, color indicators, stitch types, and stitch designations.
[0023] The present invention also relates to a third computer-implemented method for automatically generating a plurality of third 3D garment models in a batch process, the computer-implemented method utilizing a program code stored on a machine-readable medium or specifically implemented as an electromagnetic wave, each third 3D garment model representing a third garment to be made from a third garment piece, the computer-implemented method comprising the following steps: (a) for each third 3D garment model to be generated: (i) providing one or more third documents, the one or more third documents comprising: (i1) a plurality of third 2D sample pieces representing the third garment piece, and (i2) a third 2D sample piece for generating the third garment piece; (b) generating, for the batched plurality of third 3D garment models, a virtual post-processing method comprising: (i) identifying a style of a third 2D pattern piece, (ii) generating a fourth 3D garment model based on the style of the identified third 2D pattern piece, (iii) extracting at least one garment post-processing feature from the human-readable garment post-processing description, (iv) generating a third 3D garment model by virtually post-processing each fourth 3D garment model in the fourth 3D garment models based on the extracted at least one garment post-processing feature; and (b) generating, for the batched plurality of third 3D garment models, a virtual post-processing method comprising at least one of the following items: The statistical data of: (i) for each of the third 3D clothing models to be generated, whether a fourth 3D clothing model can be generated, (ii) which of the fourth 3D clothing models can be generated, (iii) for each of the fourth 3D clothing models that can be generated, the confidence level that the fourth 3D clothing model is correctly generated, (iv) which of the fourth 3D clothing models cannot be generated, (v) which of the fourth 3D clothing models can be only partially generated and which of the fourth 3D clothing models can be generated to what extent, (vi) whether a third 3D clothing model can be generated from each of the fourth 3D clothing models, (vii) which of the third 3D clothing models can be generated, (viii) for each of the third 3D clothing models that can be generated, the confidence level that the third 3D clothing model is correctly generated, (ix) which of the third 3D clothing models cannot be generated, and (x) which of the third 3D clothing models can be only partially generated and to what extent.
[0024] In an embodiment of the third computer-implemented method, the step of identifying the style of the third 2D pattern pieces is based on determining the following items using a pattern recognition algorithm: (a) a garment type indicator of at least one of the third 2D pattern pieces, the garment type indicator indicating the type of garment to which the corresponding third garment piece belongs and being one of a plurality of garment type indicators stored on a machine-readable medium, and (b) a garment piece pose indicator of each of the third 2D pattern pieces, the garment piece pose indicator indicating an estimated position and an estimated orientation of the corresponding garment piece within the third garment, wherein the step of generating a fourth 3D garment model is also based on the garment type indicator and the garment piece pose indicator.
[0025] In an embodiment of the third computer-implemented method, the pattern recognition algorithm is trainable based on at least one of the generated reports and user feedback.
[0026] In an embodiment of the third computer-implemented method, the step of extracting at least one garment post-processing feature is based on determining the at least one garment post-processing feature from the one or more third documents using an extraction algorithm, the garment post-processing feature corresponding to an additional garment component or garment characteristic and indicating the position and orientation of the corresponding additional garment component or garment characteristic within the third garment.
[0027] In an embodiment of the third computer-implemented method, in order to determine the at least one garment post-processing feature, the extraction algorithm is configured to analyze the one or more third documents with respect to at least one of the following items: text information, image data, arrows, size data, color cards, color indicators, stitch types, and stitch names.
[0028] In an embodiment of the third computer-implemented method, the extraction algorithm is trainable based on at least one of the generated reports and user feedback.
[0029] In other words, the scope of certain embodiments of the first method, the second method, and the third method may be determined as follows:
[0030] The first approach starts by characterizing the pieces, which means identifying 1) what the piece is, and 2) where the piece is on the body (mannequin) and what rotation it is. Knowing what the piece is and where it is on the mannequin will simplify the next step, which is to automatically assemble the garment. When you know which pieces are close to which other pieces, there are many fewer potential assembly solutions. The first approach can be considered to describe a trained expert system used to characterize the pieces. However, an algorithmic approximation can also be used, which uses information such as text about the piece, the shape of the piece, the size, whether it is symmetrical, the number of discrete segments, etc. In any case, the different shapes belonging to the garment are identified, and it is determined what each piece is and where the piece is on the body. Based on this, the assembly of the garment into a 3D shape is automated. The 3D model can then be used with the help of a graphical user interface (GUI) to extend or modify the design.
[0031] The second method allows to find relevant text in the document (e.g., PDF) describing the assembly instructions of the garment, which is related to the garment post-processing (stitches, hems, necklines, waists, plackets, etc.). This information is extracted and related to the 3D digital twin of the garment that has not been post-processed so far or has only been post-processed by default. This means, for example, searching for key information in an Excel spreadsheet or PDF, which is then automatically extracted and applied to the 3D digital version of the garment in the following form: garment opening (collar, sleeves, waist, etc.), color of different samples used to produce the garment, fabric type, trim, etc. In addition, images, arrows, size charts, color cards or indicators, stitch types or names can be extracted and converted into garment components and / or characteristics that can be applied to 2D garment pieces and / or 3D models to generate virtually post-processed garments.
[0032] The third method implements automated "Garment Digital Twin" (GDT) creation for each style in a single style based on a list of styles (e.g., DXF files) and corresponding "Technical Packages" (e.g., PDF files that describe the garments in human-readable form). The GDT is automatically created in a batch mode that indicates the success or failure of each conversion. The batch mode involves reading a series of documents that can be organized by garment style. Each set of documents that defines a garment is read and then converted into a 3D digital twin without any user interaction. Finally, a report is provided to the user of all converted garments and the status of whether the garment was successfully converted, as well as the confidence level that the correct conversion has been performed. This will enable companies active in the apparel industry to convert large quantities of garments overnight and then edit or correct areas where the garment conversion was unsuccessful. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Preferred embodiments of the present invention are described more fully hereinafter, by way of example only, with reference to the accompanying drawings, in which:
[0034] Figure 1 showing at least a portion of an exemplary first document or a third document;
[0035] Figure 2 An embodiment of pre-orienting and positioning 2D garment part models in their specific poses within a mannequin in preparation for virtual garment assembly is shown;
[0036] Figure 3 An embodiment of preforming a 2D garment part model into a 3D garment part model in preparation for virtual garment assembly is shown;
[0037] Figure 4 showing at least a portion of an exemplary second document or a third document;
[0038] Figure 5 It is shown that based at least in part on Figure 4 The document shown is an implementation method for post-processing a 3D clothing model that has not yet been post-processed. DETAILED DESCRIPTION
[0039] Figures 1 to 3 Various aspects used in a first computer-implemented method presented herein are shown. The purpose of the method is to automatically generate a first 3D garment model such that the first 3D garment model represents a first garment to be actually made from first garment parts.
[0040] Therefore, for the first method, it is proposed to provide one or more first documents 1, which include a plurality of first 2D pattern pieces 2 representing those first garment pieces. These documents can be computer-aided design (CAD) files, portable document format (PDF) files, image files in any format, or any other files containing computer-readable or human-readable 2D shape patterns. In the example shown, the 2D pattern document 1 contains representations of the front and back garment pieces (3 and 4), the sleeve pieces 5 and 6, and the collar piping (or neck tape) 7 of a T-shirt.
[0041] According to the invention, a pattern recognition algorithm is used to determine a garment type indicator of at least one of the first 2D pieces 2. The garment type indicator indicates the type of garment to which the respective first piece belongs. In this case, the pattern recognition algorithm detects that the document 1 belongs to a T-shirt, because a typically shaped piece and / or a typical set of typically shaped pieces is detected.
[0042] The determined garment type indicator is one of a plurality of garment type indicators stored on the machine-readable medium. For example, the garment type indicator may refer to the type of garment at a very basic level, such as a t-shirt, underpants, underwear, etc., but it may alternatively or additionally refer to the type of garment at a more detailed level, such as long-sleeved or short-sleeved, women's or men's, dress or denim, suit pants or 5-pocket, etc.
[0043] Moreover, the pattern recognition algorithm is used to determine the garment piece pose indicator of each first 2D sample in the first 2D sample. The garment piece pose indicator represents the estimated position and estimated orientation of the corresponding garment piece in the first garment. For example, the algorithm will identify a garment piece with the shape of a back sample 4, which is to be located at the back and the long straight line is at the bottom. The algorithm also identifies the front sample 3, which is to be located at the front, but is turned upside down so that the long straight line is at the bottom again and the collar part is at the top. Now, a first 3D garment model is generated based on the first 2D sample, the at least one garment type indicator and the garment piece pose indicator.
[0044] Figure 2 and Figure 3 Two examples of this positioning and orientation are shown. These two assembly methods can be used complementary or alternatively.
[0045] according to Figure 2 As shown, the first method may further include the following steps: generating first 2D garment piece models 8 of each of the first 2D sample pieces 2, the first 2D sample pieces being positioned and oriented according to the garment piece pose indicator, wherein, in order to generate the first 3D clothing model 10, the reference object 9 is covered with the first 2D garment piece model 8. In the example shown, the reference object 9 is an invisible or transparent human model having an average human body shape, in particular according to the clothing type indicator (female / male / children, etc.).
[0046] according to Figure 3 As shown, individual components of the first 3D garment model 12 can be represented by first 2D garment piece models and then preformed to form the first 3D garment piece model 11 before final assembly. As shown, for example, the sleeves are rolled to form a tube as shown and sewn together, and the front and back pieces are shaped following typical chest and back shapes. In this way, it is straightforward to assemble these first 3D garment piece models 11 to form the first 3D garment model 13, because less computational work is required to find the connecting lines (seam lines).
[0047] The first 3D garment model 13 is shown on a transparent mannequin 12, but the reference object may also be a hanger or any other support that presents the garment in a way that allows a designer to easily design the garment in a CAD environment. The reference object may be two-dimensional or three-dimensional.
[0048] In summary, the first method may further comprise the step of positioning and orienting the first 2D part model 8 based on the part pose indicator, in particular also based on the proportions of the first 2D pattern pieces 2 relative to each other. If the proportions are also taken into account, the parts can be positioned and oriented already very close to each other (because the general size of the garment can be estimated), making the rendering of the assembly more efficient.
[0049] The first method may include the following steps: shaping each of the first 2D part models 8 into a first 3D part model 11 to adapt it to the shape of the reference object 13 (in the corresponding area) based on the at least one garment type indicator and the part pose indicator while, before or after said positioning and orienting of the 2D part models 8. Thus, the first method may take into account information about the typical 3D shape of the corresponding part (which may be stored on a computer-readable medium) according to the corresponding garment type indicator (in this case: T-shirt).
[0050] Finally, the first 3D garment model 13 is generated by virtually assembling the first 3D garment part models 11 (which are the shaped first 2D garment part models 8) along their respective edges. In particular, the first 2D or 3D garment part models are virtually stitched together in a manner of a default margin from the edge, wherein a model of a default seam can be provided at a virtual stitch line.
[0051] Typically, the first method may add default post-processing to the first 3D garment model based on the corresponding garment type indicator and / or garment piece pose indicator. For example, in the case where it is recognized that the current garment is a T-shirt, the first 3D garment model 10 / 13 may be equipped with standard French seams, wherein in the case of jeans, the method may equip the first 3D garment model with standard lap seams.
[0052] In an embodiment of the first computer-implemented method, in order to determine a garment type indicator and a garment piece pose indicator, a pattern recognition algorithm is configured to analyze at least one of the following items: (a) the shape of the 2D pattern piece, (b) the proportion of the 2D pattern piece relative to other 2D pattern pieces, (c) text information associated with the 2D pattern piece in a corresponding first document, (d) whether the 2D pattern piece is at least partially symmetrical, and (e) the amount of 2D pattern pieces provided.
[0053] A particular advantage of the first method is that a first 3D garment model is fully automatically generated based solely on at least one first document and its depiction or description of a first 2D pattern piece representing a first garment part to be made of a first garment, without requiring human intervention.
[0054] The first method, in particular the pattern recognition algorithm, can be based on machine learning, trained using training data from previous generations of the first 3D garment model. In particular, such a machine learning algorithm can "learn" from user input intended to correct the automatic selection made by the computer.
[0055] Figure 4 and Figure 5 Various aspects used in the second computer-implemented method presented herein are shown. The purpose of the method is to virtually post-process a second 3D garment model, which represents a second garment to be actually made and has not been post-processed or has only been post-processed by default.
[0056] Therefore, for the second method, it is proposed to provide the second 3D garment model without or with only default post-processing, and also to provide one or more second documents comprising human-readable garment post-processing instructions for post-processing the second garment. Figure 4 An exemplary document 14 of this type is shown, which is often referred to among experts as a "tech pack". These one or more second documents describe how to post-process a certain garment. Thus, the document may refer to a plurality of garment post-processing features, such as the type of fabric to be used, the colors to be selected for specific components, the type of stitches and hems to be made at openings, the addition of certain elements, and further parameters and descriptions instructing the sewist on how to post-process the garment. The instructions in such documents are intended for humans, not machines. In Figure 4 In the example shown in , individual garment post-processing features are pointed to by arrows and described using text, parameters, references to 2D styles and / or samples.
[0057] The second method provides an extraction algorithm that is configured to determine at least one such garment post-processing feature from the one or more second documents. The garment post-processing feature corresponds to an additional garment component and / or garment property and indicates the position and orientation of the corresponding additional garment component or garment property within the second garment. Thus, the extraction algorithm specifically extracts all information required for post-processing the garment, as specified in the second document. In particular, the extraction algorithm is configured to interpret human-readable instructions and illustrations to establish the garment post-processing feature.
[0058] Then, the second 3D garment model is virtually post-processed based on the determined at least one garment post-processing feature. Specifically, the second method can retrieve stored elements (e.g., a 3D model of a zipper or a 2D model of a seam) and / or model appearance (texture, color, etc.) to perform virtual post-processing on the second 3D garment model 21 (see Figure 5 ) is modified to form a post-processed second 3D garment model 22. In particular, the second method provides an explanation of the garment post-processing feature based on the following training data: the training data is about how and where to modify the second 3D garment model 21 so as to obtain the post-processed second 3D garment model 22. Therefore, when the garment post-processing feature corresponds to the fabric type 17 of at least a portion of the second 3D garment model, the at least a portion of the second 3D garment model is modified by the following steps: based on the garment post-processing feature, at least one of a texture and a color is assigned to at least a portion of the second 3D garment model.
[0059] If the garment post-processing feature corresponds to a seam of the second garment (eg, see reference numeral 18), the at least a portion of the second 3D garment model is modified by assigning at least one of a thread color, a thread type, and a stitch pattern to the seam of the second garment.
[0060] The step of performing virtual post-processing on the second 3D clothing model 21 may also include, as mentioned above, adding at least one clothing post-processing model to the second 3D clothing model. Figure 4 As shown, see text 15 pointing to the collar, using the added pressure collar (see Figure 1 In other embodiments, the neckline may be a border for further shaping the upper edge of the garment with a collar, a cowl, darts or pleats.
[0061] If the garment post-processing feature corresponds to a seam of the second garment, then the at least one garment post-processing model may represent a stitch to be placed at a specific location within the second garment, see Figure 4 If the garment post-processing feature corresponds to a piping of the second garment (see reference numeral 18), the at least one garment post-processing model may represent a hem to be placed at the lower opening of the second garment. Figure 5 In the given example of , the bottom edge of the second 3D garment model 21 is turned down (or simply shortened for visualization) and virtually sewn.
[0062] In an embodiment of the second computer-implemented method, the garment post-processing feature may also correspond to a placket of a garment (this is not the case in the example shown), wherein the at least one garment post-processing model represents at least one of the following items: buttons, button hooks, snaps, zippers, Velcro, magnetic fasteners, hooks, loops for hooks or buttons, cord locks, and laces.
[0063] Specifically, in order to determine the at least one garment post-processing feature, the extraction algorithm can be configured to analyze the one or more second documents 14 with respect to at least one of the following items: text information 15 to 19, image data, arrows, size data 20, color cards 17, color indicators, stitch types, and stitch names. The algorithm can be trained to search the second document 14 for keywords, known patterns describing garment post-processing features, arrows (especially the alignment of arrows), numerical values (especially typical numerical ranges), etc.
[0064] Specifically, the second 3D garment model 21 may be an assembly of a plurality of second 3D pattern pieces, which represent second garment parts of the second garment to be made, i.e., just as it is established as a first 3D garment model by the first method described above. The first 3D garment model lacks post-processing or is provided with default post-processing because it is based on the first 2D pattern pieces contained in the first document, and the first document is usually silent about how to assemble the garment parts.
[0065] In a specific embodiment, the third method presented herein is a combination of the first method and the second method. However, more generally, the purpose of the third method is to automatically generate multiple third 3D garment models in a batch process, each of which represents a third garment to be made from a third garment piece. According to the third method, one or more third documents are provided for each third 3D garment model to be generated. The third document includes: multiple third 2D samples representing the third garment piece; and human-readable garment post-processing instructions for post-processing the third garment. Therefore, just like the assembly process in reality, garment patterns and corresponding instructions ("technical packages") are provided.
[0066] Then, the pattern of the third 2D sample piece is automatically identified, and a fourth 3D garment model is generated based on the pattern. At least one garment post-processing feature is extracted according to the human-readable garment post-processing description, and a third 3D garment model is generated by virtually post-processing each of the fourth 3D garment models based on the at least one garment post-processing feature. Even if the nomenclature implies otherwise, the fourth 3D garment model is also a "semi-finished product" and a predecessor of the third 3D garment model.
[0067] After batch processing of multiple third 3D clothing models, a report is generated, which indicates at least one of the following items: (i) Can all third documents be translated into fourth 3D clothing models? (ii) Specifically, which of the fourth 3D clothing models can be generated? (iii) What confidence level (likelihood of correctness) does each of the generated fourth 3D clothing models have? (iv) Specifically, which of the fourth 3D clothing models cannot be generated? (v) Which of the fourth 3D clothing models can be only partially generated and to what extent (for example, what parts are missing)? (vi) Can third 3D clothing models be generated from each of the fourth 3D clothing models in the fourth 3D clothing models? (vii) Which of the third 3D clothing models can be generated? (viii) What confidence level (likelihood of correctness) does each of the generated third 3D clothing models have? (ix) Which of the third 3D clothing models cannot be generated? And (x) which of the third 3D garment models can be only partially generated and to what extent (what parts or features are missing)?
[0068] The user can use the report to quickly identify unsuccessful or partially successful conversions in order to manually correct the corresponding model or give feedback to the computer (e.g., by confirming or rejecting the corresponding part of the report). The report and / or manual correction / user feedback can be used to further improve or train pattern recognition, extraction and / or virtual garment assembly.
[0069] Specifically, the step of identifying the style of the third 2D pattern piece can be based on determining the following items using a pattern recognition algorithm: a garment type indicator of at least one of the third 2D pattern pieces; and a garment piece pose indicator of each of the third 2D pattern pieces. The step of generating the fourth 3D garment model is also based on the garment type indicator and the garment piece pose indicator.
[0070] The garment type indicator indicates (or in other words: indicates) the type of garment to which the corresponding third garment part belongs. The garment type indicator may be selected from a plurality of garment type indicators stored on a machine-readable medium. The garment part pose indicator indicates (or in other words: indicates) an estimated or preliminary position and an estimated or preliminary orientation of the corresponding garment part within the third garment.
[0071] Moreover, the step of extracting the at least one garment post-processing feature can be based on: using an extraction algorithm, determining the at least one garment post-processing feature from the one or more third documents (in particular, from human-readable garment post-processing instructions for post-processing the third garment). The garment post-processing feature corresponds to an additional garment component or a garment property and represents (or in other words: indicates) the position and orientation of (a) the corresponding additional garment component within the third garment, or (b) the position and orientation of the corresponding garment property.
[0072] In order to determine the at least one garment post-processing feature, the extraction algorithm can be configured to analyze the one or more third documents (particularly human-readable garment post-processing instructions for post-processing the third garment) with respect to at least one of the following items: text information, image data, arrows, size data, color cards, color indicators, stitch types, and stitch names.
[0073] In particular, the one or more third documents corresponding to a single garment (for which a third 3D garment model is to be generated) may be a so-called "technical package" corresponding to a 2D pattern and assembly instructions. This may all be combined in one file or document, however, it may also be provided as two or more related files or documents.
[0074] Notice, Figure 4 The text in is dummy text (lorem ipsum) whose sole purpose is to indicate that in such a document there may be several paragraphs of text that are for assembly instructions and similar comments for human users, i.e., not machine-readable. Figure 4 An example of at least a portion of a second document and at least a portion of a third document is shown.
[0075] Although the invention has been illustrated above, partly with reference to some preferred embodiments, it must be understood that many modifications of these embodiments and combinations of different features can be made. These modifications all fall within the scope of the appended claims.
Claims
1. A computer-implemented method for automatically generating a plurality of third 3D garment models in a batch process, the computer-implemented method utilizing program code stored on a machine-readable medium or embodied as electromagnetic waves, each third 3D garment model representing a third garment to be made from a third garment piece, the computer-implemented method comprising the following steps: -For each third 3D clothing model to be generated: o Providing one or more third documents, wherein the one or more third documents include: ■ a plurality of third 2D pattern pieces representing the third garment piece, and ■ human-readable garment post-processing instructions for post-processing the third garment, o identifying the pattern of the third 2D sample piece, ○ generating a fourth 3D garment model based on the identified style of the third 2D pattern piece, ○ extracting at least one garment post-processing feature from said human-readable garment post-processing instructions, ○ generating a third 3D clothing model by virtually post-processing each of the fourth 3D clothing models based on the extracted at least one clothing post-processing feature; and - generating, for the batch-processed plurality of third 3D garment models, a report including statistical data on at least one of the following items: ○ For each of the third 3D clothing models to be generated, whether a fourth 3D clothing model can be generated, ○ which of the fourth 3D clothing models can be generated, o for each of the fourth 3D clothing models that can be generated, a confidence level that the fourth 3D clothing model is correctly generated, o which of the fourth 3D clothing models cannot be generated, o which of the fourth 3D clothing models can be only partially generated and to what extent, o whether a third 3D clothing model can be generated from each of the fourth 3D clothing models, ○ which of the third 3D clothing models can be generated, o for each of the third 3D clothing models that can be generated, a confidence level that the third 3D clothing model is correctly generated, o which of the third 3D clothing models cannot be generated, and o which of the third 3D garment models can be only partially generated and to what extent, Wherein, the step of extracting at least one garment post-processing feature is carried out based on determining the at least one garment post-processing feature from the one or more third documents using an extraction algorithm, and the garment post-processing feature corresponds to an additional garment component or garment characteristic and indicates the position and orientation of the corresponding additional garment component or garment characteristic within the third garment.
2. The computer-implemented method of claim 1 , wherein: The step of identifying the pattern of the third 2D sample is performed based on determining the following items using a pattern recognition algorithm: a garment type indicator of at least one of the third 2D pieces, the garment type indicator indicating the type of garment to which the corresponding third garment piece belongs and being one of a plurality of garment type indicators stored on the machine-readable medium, and - a piece pose indicator for each of the third 2D pieces, the piece pose indicator representing an estimated position and an estimated orientation of the corresponding piece within the third garment, The step of generating the fourth 3D clothing model is also performed based on the clothing type indicator and the garment piece position indicator.
3. The computer-implemented method of claim 2, wherein: The pattern recognition algorithm is trainable based on at least one of the generated reports and user feedback.
4. The computer-implemented method of claim 1 , wherein: In order to determine the at least one garment post-processing feature, the extraction algorithm is configured to analyze the one or more third documents with respect to at least one of the following: - text messages, - image data, -arrow, - Dimensional data, - Color cards, - color indicator, -Stitch type, and -Stitch name.
5. The computer-implemented method of claim 1 , wherein: The extraction algorithm is trainable based on at least one of the generated reports and user feedback.
