Design apparatus and method for providing 3D model for design development of object

Abstract

The invention relates to a design device (1) and a method for providing a 3D model (5) for the design development of an object. The method comprises: presetting (S2) an object type (18) on a trained artificial neural network (10), training the artificial neural network (10) with a plurality of base image data (13) of objects of the object type (18); creating (S3) an image data set (15) of a preset object type by means of a trained artificial neural network (10), the artificial neural network (10) providing image data (14) from a plurality of perspectives as the image data set (15); checking (S4) whether a design selection criterion exists in the created image data set (15), and if a design selection criterion (17) does not exist, recreating the image data (14) by means of a modification image attribute criterion (S5); creating (S6) a digital 3D model (5) from the image data set (15) by means of a reconstruction algorithm if a design selection criterion (16) is present; the created digital 3D model (5) is provided (S7) for design development of the object.

Description

Technical Field

[0001] The invention relates to a method for providing a 3D model for the design development of an object. Furthermore, the invention relates to a design device which is configured to carry out the method. Background Art

[0002] The drawing of 3D CAD geometries is an important stage in the design process for obtaining the correct proportions, technical boundary conditions and / or an overall feel for the product / object. To date, 3D geometries have been generated by designers based on 2D sketches or photographs with the aid of CAD (computer aided design) software. The drawing of this 3D geometry is refined by the engineering design team through multiple design reviews until a certain technical and design maturity is reached. Usually 2D sketches are drawn, but each of the 2D sketches often needs to be adjusted several times. Subsequently, a 3D model is drawn based on the 2D sketch with the aid of CAD, wherein a rendering is generated for the decision maker based on the 3D model, which often requires multiple iterations.

[0003] This approach has many disadvantages, especially when drawing 2D sketches, multiple sketches are required for each view of a 3D object, and when the design changes, each view must be adjusted individually, that is, each incremental adjustment of the design leads to a large number of adjustments in the sketch. In addition, when converting to a three-dimensional model based on the 2D sketch, a media interruption occurs, wherein the drawing of the 3D model mostly needs to be done by specialized professionals. Due to the incremental adjustment of the design, this also leads to complex manual adjustments of the 3D model and requires a lot of communication between the designer and the CAD professional, because the 2D sketch often does not contain enough information to correctly describe and then draw the 3D model. Summary of the invention

[0004] It is therefore an object of the present invention to simplify and / or accelerate the design and development of an object.

[0005] This object is achieved by the independent claim. Advantageous developments of the invention are disclosed in the dependent claims, the following description and the drawings.

[0006] The invention provides a method for providing a 3D model for the design and development of an object. The method comprises the following steps, which can preferably be performed by a computing device, in particular a processor and / or a microchip. The method comprises: presetting an object type for at least one trained artificial neural network or on at least one trained artificial neural network, wherein the at least one artificial neural network is trained by a plurality of basic image data of an object of this object type. That is, firstly an object type can be preset, by which the type of object to be designed is determined. For example, the object type can be preset for a computing device or a computer program configured to perform the method. By means of the object type, it can be defined for the program, for example, whether motor vehicle parts (such as, for example, body, rim, interior parts) and / or parts from other fields besides motor vehicle technology should be created as objects. For example, the object type can be preset for the computer program or the trained artificial neural network by means of a graphical user interface (GUI).

[0007] At least one artificial neural network may include, for example, a generative adversarial network (GAN). An artificial neural network (often also referred to as an artificial neural network) is understood to be a network composed of artificial neurons for machine learning and artificial intelligence. Various problems can be solved on a computer basis with the help of artificial neural networks. Finally, an artificial neural network is an algorithm according to which different data sources (such as, for example, image data) can be interpreted and information or samples can be extracted from them in order to apply the extracted information or samples to hitherto unknown data. Finally, data-driven predictions can be made thereby, so that, for example, artificial data can be created by the artificial neural network from the learned data. The artificial neural network can be trained by image data, in particular by basic image data of objects of a predetermined object type.

[0008] In other words, the basic image data can be training data for an artificial neural network, which training data originate from a plurality of objects of the same object type. For example, if a new design is to be developed for an object, the object can be a motor vehicle, wherein, for the design development of the motor vehicle, the object type that should be designed by the method can be preset, such as a vehicle body or a wheel rim of a motor vehicle. Preferably, the artificial neural network can then be trained by basic image data of different vehicle bodies or wheel rims, in particular of a plurality of different motor vehicles. The basic image data can thus comprise a collection of a plurality of images, wherein the individual images each comprise a different design of the object type. For example, it can be provided that a training process is carried out before the described method steps, within the scope of which the GAN is trained based on a data set of basic image data of real or artificial objects of the object type. Subsequently, by applying the GAN, new images based on the image properties of the basic image data can be generated therefrom by artificial intelligence.

[0009] As a further method step, an image data set of a predetermined object type is created by means of a trained artificial neural network, wherein the artificial neural network provides image data of the object type from a plurality of different viewing angles as the image data set, which image data are determined from the learned basic image data.

[0010] In other words, the trained artificial neural network can render new image data based on the learned basic image data, in particular for a plurality of different perspectives or views of an object of a preset object type. Here, an artificial neural network that has been trained with basic image data of the same object type can be selected for thereby creating artificial image data at corresponding different perspectives of the object, wherein the image data refers to 2D views from different perspectives.

[0011] In a next step, the created image data set can be checked for the presence of design selection criteria, wherein, if no design selection criteria are present, the image data of one or more perspectives are recreated by means of changing the image attribute criteria. For example, the design selection criteria can include whether the image attributes in the corresponding perspectives of the image data match or are consistent with each other. For example, if there is a color difference between two perspectives, this is an inconsistency, wherein, in this case, there is no design selection criterion. In the example of a vehicle body, for example, different vehicle styles may also be present, such as a luxury sedan in one perspective and a station wagon in another perspective, which is equivalent to an inconsistency and therefore does not meet the design selection criteria. Alternatively or additionally, the design selection criteria can also depend on the designer's design intuition, that is, aesthetic criteria. At this point, for example, the designer can manually approve or disapprove the image data.

[0012] If there is no design selection criterion, the image data of one or more perspectives can be recreated by the artificial neural network by changing the image attribute criterion. Here, the image attribute criterion may include rules, algorithms, data and / or information based on which the image data can be changed. In particular, by changing the image attribute criterion, a plurality of the created image data of a perspective can be recreated by interpolating and / or recombining the image attributes of the generated image data or parts of the image data. For example, when using a GAN, one or more latent vectors can be changed to change the image data of at least one perspective.

[0013] Preferably, when interpolating, interpolation is performed between at least two provided image data of the image data set. Preferably, a special design of GAN is used, more precisely, a style-based GAN is used. From a mathematical point of view, the first artificial neural network based on the style GAN is used to map random vectors in an intermediate latent space. For this purpose, for example, linear interpolation can be performed between two or more latent vectors of the image data set, which are respectively assigned to each of the at least two provided image data. At this time, another latent vector assigned to another image data set different from the at least two provided data images is calculated, wherein the other latent vector is assigned to the image data located between the two provided latent vectors of the image data set. From a mathematical point of view, when interpolating, a straight line is generated that runs through the image data set that can be understood as a type of image data space or potential image space, which connects the two provided image data. Each position on the straight line represents another image data arranged at the position in the image data set. Thus, when interpolating, more image data can be extracted from the image data set.

[0014] By means of interpolation, it can be achieved that a weighted average is generated between at least two provided image data. For example, 70% of the information (in particular design attributes) of one of the at least two provided image data can be used, while 30% of the information of the other of the provided image data can be used. The combination of the corresponding information then generates further image data that is different from the provided image data. Ultimately, any combination of the provided image data and / or a combination selected by the user by correspondingly changing the latent vector can be created in the form of new image data by interpolation.

[0015] The reconstruction is based on the recognition that the image data each describe different image properties, which are each described by a portion of the image data. For example, the provided image data can have up to 18 image properties. Image properties can be understood as different information levels or levels of image data. Image properties can be referred to as abstract features. For example, one or more image properties can at least partially describe and / or determine the number of spokes of a rim, the color design of the rim and / or the shape of the spokes of the rim.

[0016] For example, during the reconstruction, it can be selected that a certain first image attribute is to be adopted from the image data, while a second image attribute different from the first image attribute is to be adopted from other image data. In this way, for example, the color of the spokes can be adopted from the first image data and the number of individual spokes of the rim can be adopted from the second image data to create new image data. Thus, during the reconstruction, the respectively selected image attributes of the respectively provided image data can be combined with one another.

[0017] In other words, the basis for the reorganization is the image properties of the image data generated at different levels in the artificial neural network. In the case of a GAN, the image properties are created by a so-called generator of the GAN. The generator is an artificial neural network of the GAN, that is, after a training process with basic image data, it creates artificial image data, such as a rim, i.e., an image data set. Preferably, during the training process of the generator for a preset object type or preset object style (e.g. a preset rim style), it is always preferred from a mathematical point of view to use a plurality of latent vectors on all levels of the generator. In the example case of a training process (for the sake of clarity, only two of the latent vectors are described here), for example, a first latent vector can be assigned to levels 1 to k, wherein a second vector is assigned to levels k+1 to n, wherein n describes the maximum number of levels. In the case of using a trained artificial neural network, different latent vectors can be used at each level. Preferably, in the case of up to 18 levels of this type in the generator, a large number of possible, respectively different combinations of image properties are obtained. Preferably, during the reorganization, a plurality of provided image data that differ from one another are combined with one another, and the trained generator thus has a plurality of different latent vectors, in particular up to n different levels, wherein n describes the number of levels in the trained generator. Ultimately, this allows, for example, the creation of one or more new image data with a combination of image attributes that best corresponds to the personal ideas of the designer.

[0018] Preferably, the image properties adapted by changing the latent vector can be known, whereby the image properties of the image data can be regenerated or redesigned in a targeted manner, for example, shapes, proportions, colors and / or individual details in the corresponding image data can be changed in a targeted manner.

[0019] If design selection criteria are present, a digital 3D model can then be created from the image data set by a reconstruction algorithm, wherein the reconstruction algorithm calculates the digital 3D model from the image data from a plurality of different viewing angles. Preferably, the plurality of different viewing angles can be predefined in such a way that they enable the reconstruction algorithm to calculate the 3D model. The number of viewing angles required can depend on the object, wherein at least two viewing angles, preferably 5 to 10 viewing angles, are used that look at the object from different angles. In order to reconstruct the digital 3D model from the image data set comprising a 2D view of the object, known reconstruction techniques can be used, which can be based in particular on neural networks and / or photogrammetry and / or differentiable rendering.

[0020] Finally, the created digital 3D model can be provided as a design for the development of an object. In this case, the digital 3D model can be provided, for example, to a CAD program for further development of the object, for example a vehicle. Particularly preferably, the digital 3D model can be provided to a manufacturing device for manufacturing a real 3D model from the digital 3D model, wherein in this case, the manufacturing device can be a 3D printer, and / or the digital 3D model can be provided to a manufacturing device for manufacturing an object according to a template of the digital 3D model, for example a production line for a vehicle body.

[0021] The advantage obtained by the invention is that the design development can be completed in an accelerated and targeted manner. New designs can be quickly exported, thereby shortening the time cost for design development. In addition, rapid iteration and further development of existing designs are achieved, as well as targeted matching with existing brands or different design languages ​​of brands. In addition, there is no media interruption, that is, no conversion from drawings to digital object representations is required, for example. Ultimately, an abstract, but still targeted and, for example, preset by the designer, interpolation and / or reorganization of artificial basic image data can be achieved, thereby simply and quickly obtaining control over the design and 3D geometry and style, which is usually very complicated or impossible to achieve with existing CAD / CAS methods.

[0022] The invention also includes embodiments that provide additional advantages.

[0023] One embodiment provides that a plurality of artificial neural networks are provided, wherein the corresponding artificial neural network is trained for each preset viewing angle of the object with the aid of basic image data of an object of the object type from this viewing angle. In a corresponding manner, the corresponding artificial neural network for creating the image data set can then only create image data of the corresponding learned viewing angle. In other words, a plurality of trained neural networks can be trained for corresponding 2D views, wherein the basic image data for training are provided to the corresponding neural networks from the same viewing angle. Thus, by each of the trained neural networks, only a 2D view observed from this viewing angle is generated, wherein a plurality of viewing angles are provided by all the trained neural networks, which can then be used to reconstruct the 3D model. For example, the artificial neural network can be a generative model, in particular StyleGAN 2 or SWAGAN. The advantage resulting from this is that a preferred embodiment for creating image data from a plurality of viewing angles can be provided.

[0024] Another embodiment provides that at least one artificial neural network is trained with the aid of corresponding 2D views of 3D objects of the object type, wherein 2D views from a plurality of preset viewing angles of the corresponding 3D object are provided. In other words, the artificial neural network is trained with 3D objects, wherein the corresponding 3D object can be rotated into a plurality of preset viewing angles and the 2D views generated therefrom are then used for training. Here, for example, real or digital 3D objects or models can be used, which are rotated into different viewing angles, in particular in a digital manner. For example, the 3D objects used for training the artificial neural network can exist in a CAD tool, whereupon the 3D objects are rotated into a preset viewing angle by the CAD tool to provide 2D views for training. As a result, the artificial neural network can learn the 2D representation of the 3D object and then create new image data from the learned 2D views when creating the image data set. Alternatively, a perspective-dependent heterogeneous data set of the basic image data of the 3D object can be used to train at least one artificial neural network. Here, for example, Style-NERF can be used as an artificial neural network. This embodiment has the advantage that the neural network can already be trained using consistent basic image data of the 3D object and thus inconsistencies when creating new image data can be reduced.

[0025] Preferably, it is provided that a design selection criterion is present only if the image properties under a plurality of viewing angles of the image data are identical. In particular, it can be checked whether the image properties of the image data under a plurality of different viewing angles are identical or, for example, whether there are stylistic breaks, in particular incoherent shapes, between the different viewing angles. For example, in the case of a vehicle body, it can be checked whether a station wagon is shown in one viewing angle and a limousine in another viewing angle, wherein in this case no design selection criterion is present. Preferably, the design selection criterion can be checked automatically, in particular by means of a further neural network, which checks the generated image data for this purpose. Alternatively or additionally, the design selection criterion can be checked by a designer who can approve the image data set. This has the advantage that the design development of the object can be further simplified.

[0026] Another embodiment provides that as an artificial neural network, a generative adversarial network, GAN (Generative Adversarial Network) is used. Alternatively, a generative adversarial network can be described as a type of generative deep neural network. GAN is an algorithmic architecture that uses two artificial neural networks. These two artificial neural networks are ultimately used to generate a synthetic new data set, that is to say an artificial new data set, in particular an image data set.

[0027] The artificial neural network of the GAN is called the generator and creates an image data set. Another artificial neural network of the GAN is called the discriminator and evaluates the image data of the created image data set. Typically, from a mathematical point of view, the generator maps a vector comprising latent variables onto a desired result space. During training, the generator learns to create image data of the image data set according to a distribution preset for it. The discriminator is trained to distinguish the results of the generator from data from a preset distribution. Here, the preset distribution is the base image data. At the end of the training, the generator is designed to create image data that the discriminator cannot distinguish from image data of the base image data. As a result, the created distribution, that is to say the created image data set, should gradually approach the real distribution, that is to say, for example, an actual image of an object (such as, for example, a vehicle body).

[0028] Preferably, a special design of a GAN is used, more specifically, a style-based GAN, such as StyleGAN 1, StyleGAN 2 and / or SWAGAN (Style and Wavelet Based GAN, style and wavelet based GAN). From a mathematical point of view, a first artificial neural network, such as a style-based GAN, which is constructed as StyleGAN 1, is used to map random vectors in an intermediate latent space. For example, the data output of the first artificial neural network is fed to a second artificial neural network at different adaptive instance normalization layers (AdalN for adaptive normalization (Adaptive InstanceNormalisation)) and it is monitored that the image data created by the second neural network is configured as close to realistic image data of the object. Furthermore, alternatively or additionally, the data output of the first artificial neural network can be fed to the second artificial neural network at a similar functional layer. For example, in the case of StyleGAN 2, the similar functional layer is a so-called demodulation and / or modulation. The second neural network can be referred to as a synthesis network. Image data are created and then generated by the synthesis network step by step, wherein the image resolution starts from 4×4 pixels and is gradually refined with the help of artificial neural networks. At different levels, latent vectors are used as style vectors. Depending on the desired size of the generated image data, the number of levels is up to 18. For example, a portable network graphic (Png image file) is finally output. How a certain style vector or latent vector affects the generated image data also depends on the selected depth of the corresponding AdalN layer.

[0029] The image data created by the generator have image attributes, wherein when using a style-based GAN, the image attributes may be referred to as style. The image attributes assigned to the first level are preferably responsible for the global attributes of the mapped object. Preferably, the image attributes assigned to the lower levels determine the local attributes of the object, like for example the color. Depending on the selected image data resolution, the synthesis network can monitor up to 18 style vectors, that is to say 18 image attributes. The concatenated style vectors may also be referred to as DNA vectors.

[0030] A prerequisite for the successful training of a style-based GAN is basic image data for training. The basic image data preferably have a variety of object designs. Preferably, the basic image data can also be evaluated by a principal component analysis PCA (Principal Component Analysis) in Fourier space. With the help of T-distributed stochastic neighbor embedding T-SNE (T-SNE), the results of this evaluation are transferred to the two-dimensional image space. As a result, particularly similar image data can be identified at an early stage and classified.

[0031] Therefore, based on artificial intelligence methods, by training a GAN, in particular a style-based GAN, with a base image dataset suitable for the intended object, reliable, realistic new image data of the object are created and thus suitable image datasets are provided for the method.

[0032] Another embodiment provides that the image data of one or more viewing angles are recreated by interpolating and / or recombining the image data, by means of changing the image property criteria. In other words, at least two different image data can be used to create a new and changed image data set, so that the single provided image data can be changed. Preferably, by means of a trained artificial neural network, a plurality of image data can first be created for each viewing angle, wherein by changing the image property criteria, two or more image properties in the corresponding image data of the corresponding viewing angle can be reestablished by interpolation and / or recombining, thereby creating completely new image data. Here, the GAN function for interpolation and recombining has been described above. Thus, starting from the image data set created by means of at least one artificial neural network, combined image data can be generated as new image data by selecting at least two image data and interpolating and / or recombining them. Preferably, the combined image data describes the image of an object newly designed by the artificial neural network, that is to say the image of an object unknown to the artificial neural network at least so far. Alternatively, the new image data can be generated by randomly changing the latent vector in the latent space. The advantage obtained by this embodiment is that the image properties can be changed purposefully and thus fine adjustment of the design development can be achieved.

[0033] Preferably, provision is made for changing the image property criterion by means of adjusting the latent vector. In other words, the image property criterion is defined by the latent vector. Here, the latent vector can access different levels of the learned basic image data and thereby can access the image properties that the image data created for the image data set should have. In particular, it is thereby possible, for example, for the designer to change the image properties in one or more image data at different viewing angles in a targeted manner without having to laboriously create a completely new 2D sketch for this purpose. Preferably, the latent vector can be changed by means of a graphical user interface in which the relevance of the latent vector is provided. That is, it is possible to know via the graphical user interface which image property is adjusted by changing the corresponding latent vector. In particular, the designer can move the point described by the latent vector along the principal component in the latent space and thereby change the form of expression of a specific property of the object. The advantage obtained thereby is that the image data can be adjusted in a simple manner.

[0034] Preferably, it is provided that the image attributes that are adjusted by changing the image attribute criteria include: the dimension of the object, i.e., for example, the size or size of the object; the proportion of the object; the style direction of the object; the color of the object and / or the shape of individual details of the object. For example, these image attributes can be adapted by controlling different levels of the base image with the aid of the latent vectors described above.

[0035] Another embodiment provides that an object and / or a real 3D model is manufactured from the provided digital 3D model, in particular by means of a manufacturing device. That is, the manufacturing device for making the real object can be controlled by the data of the digital 3D model. Preferably, the manufacturing device can automatically produce the object with the help of the data or information of the digital 3D model, that is, without additional manual working steps. Here, the manufacturing device can, for example, be based on an additive manufacturing method, in particular including a 3D printer or a lithography device. Alternatively or additionally, the manufacturing device can have an injection molding machine or can be an injection molding machine. Thus, for example, a prototype based on the digital 3D model provided by the method can be manufactured. Ultimately, various types of objects, such as vehicle bodies or rims for motor vehicles, can be manufactured simply and cost-effectively. Thus, the method according to the present invention can assist the entire object manufacturing process from design development to the manufacture of a newly designed object.

[0036] Another aspect of the invention relates to a design device, which is designed to carry out a method according to any of the above-described embodiments. For example, the design device may include a computing device, in particular a computer, on which at least one trained artificial neural network can be run. Preferably, the artificial neural network can be controlled via a graphical user interface and has at least one screen in which the individual steps of the method can be monitored and / or the provided digital 3D model can be displayed. The design device may also include a manufacturing device to which the digital 3D model is transmitted and which can then manufacture the real 3D model. Here, the same advantages and variants as in the method are obtained.

[0037] For use cases or application scenarios that may occur in the method and are not explicitly described here, it may be provided that the method outputs an error message and / or requests user feedback and / or adjusts standard settings and / or predetermined initial states.

[0038] The present invention also includes a control device for designing equipment. The control device may have a data processing device or a processor device, which is configured to execute an embodiment of the method according to the present invention. For this purpose, the processor device may have at least one microprocessor and / or at least one microcontroller and / or at least one FPGA (Field Programmable Gate Array) and / or at least one DSP (Digital Signal Processor). In addition, the processor device may have a program code, which is configured to execute an embodiment of the method according to the present invention when executed by the processor device. The program code may be stored in a data memory of the processor device. Alternatively or additionally, a method that can be provided as a program code may be provided by a computing cloud (Cloud computing).

[0039] The invention also includes improvements of the design device according to the invention, which have the features already described in conjunction with the improvements of the method according to the invention. For this reason, the corresponding improvements of the design device according to the invention are not described in detail here.

[0040] Preferably, a model of a motor vehicle component, in particular for a passenger car, commercial vehicle, bus and / or motorcycle, can be provided by the method.

[0041] As another solution, the present invention also includes a computer-readable storage medium, which includes commands that, when executed by a computer or a computer group, cause the execution of an embodiment of the method according to the present invention. For example, the storage medium can be provided at least partially as a non-volatile data storage (such as flash memory and / or SSD-solidstate drive) and / or at least partially as a volatile data storage (such as RAM-randomaccess memory). A processor circuit with at least one microprocessor can be provided by the computer or computer group. The commands can be provided as binary code or as assembly code and / or as source code in a programming language (such as C language).

[0042] The present invention also includes combinations of features of the embodiments described. Therefore, the present invention also includes implementations which each have a combination of features of a plurality of the embodiments described, as long as these embodiments are not described as mutually exclusive. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The following describes an embodiment of the present invention.

[0044] Figure 1 A schematically illustrated design device according to an exemplary embodiment is shown;

[0045] Figure 2 A schematic method flow chart according to an exemplary embodiment is shown. DETAILED DESCRIPTION

[0046] The embodiments described below are preferred embodiments of the present invention. In the embodiments, the described parts of the embodiments are the individual features of the present invention that can be regarded as independent of each other, and these features also improve the present invention independently of each other. Therefore, the present disclosure should also include feature combinations that are different from the feature combinations of the illustrated embodiments. In addition, the embodiments may also be supplemented by other features in the features already described of the present invention.

[0047] In the figures, the same reference numerals respectively denote elements with the same function.

[0048] exist Figure 1, a design device 1 is depicted, which can be designed to carry out a method for providing a 3D model for the design development of an object. The design device 1 can include a computing device 2, in particular a computer, which has a processor device 3 and a storage medium 4. Thus, a program code for at least one artificial neural network and / or basic image data of an object can be stored on the storage medium 4, for example a hard disk of the computer, wherein the processor device 3 can be designed to execute the program code of the artificial neural network to generate a digital 3D model 5 of an object, wherein in this example, the object can be a motor vehicle. Alternatively, the processor device 3 can be provided, for example, in the cloud.

[0049] The computing device 2 may be provided with a display device 6 for displaying the digital 3D model 5, wherein the method for providing the digital 3D model 5 may be monitored and / or adjusted via the display device 6, preferably via a graphical user interface. In particular, the design apparatus 1 may include an input device 7, by means of which settings may be changed, in particular on a graphical user interface. For example, the input device may include a computer keyboard, a computer mouse and / or a touch input panel.

[0050] Preferably, the design device 1 additionally comprises a manufacturing device 8, which can be configured as an additive manufacturing device 8, such as a 3D printer or a photolithography system. In particular, the manufacturing device 8 can be configured to convert the digital 3D model 5 into a real 3D model 9, that is, in this embodiment, a real motor vehicle model.

[0051] The corresponding data can be transmitted by wire and / or wirelessly between the individual devices of the design device 1 , in particular between the computing device 2 , the display device 6 and / or the production device 8 .

[0052] exist Figure 2, a method for providing a 3D model 5 of an object according to an exemplary embodiment is shown. In this case, in a first method step S1, an artificial neural network 10 can be trained, wherein the artificial neural network 10 is preferably a generative adversarial network, i.e. a so-called GAN (generative adversarial network). Such a GAN 10 comprises two artificial neural networks 11, 12, wherein the first network is referred to as a generator 11 and the second network is referred to as a discriminator 12. The GAN 10 can alternatively be trained using a plurality of basic image data 13 of the object, wherein these basic image data can preferably be provided from different viewing angles of the object, and a separate GAN 10 is trained for each of these viewing angles. Suitable for this are, for example, the so-called StyleGAN 2, SWAGAN or StyleNERF. Alternatively, a 3D object can be provided as basic image data 13, which can be rotated into a plurality of different 2D views and thus provide a preset viewing angle for the corresponding 3D object. In this example, the basic image data can be a vehicle body. However, other object types can also be trained with the help of corresponding basic image data.

[0053] In method step S2, an object type 18 can be predefined for the trained GAN 10, for which the GAN 10 is to create images, wherein the GAN 10 has preferably been trained beforehand for the predefined object type 18 using basic image data 13 for the object type 18. Thus, in this example, it can be predefined that a vehicle body is to be provided as a design development as object type 18.

[0054] In step S3, the GAN 10 can create an image data set 15 of a preset object type 18, wherein the trained GAN 10 interpolates and / or reorganizes image features learned from the basic image data 13 or training data to generate new image data 14 of the object type 18, in particular, image data 14 for multiple different perspectives of the object.

[0055] In step S4, it can be checked for the generated image data set 15 whether a design selection criterion 16 is present or not, wherein the design selection criterion can be, for example, whether the image properties in the image data 14 are consistent. This means, for example, whether the color of the vehicle body is consistent and / or whether the shapes match each other in the corresponding viewing angles. For example, the designer can also check whether the style is as expected or whether the image data 14 needs to be changed. If a change is required and there is no design selection criterion 17, one or more viewing angles can be regenerated by the GAN 10 in step S5 by changing the image property criterion. In this case, in particular, one or more latent vectors can be adjusted, which can control different levels of the basic image data and thus in particular can change the size, proportions, style direction, color and / or shape of individual details. Preferably, for this purpose, the image properties of the image data (which may be present for each viewing angle) are interpolated and / or reorganized by changing the image property criterion in order to adapt the image properties in a targeted manner. Thus, for example, a designer can draw completely new image data 14 or simply change individual image properties of the image data 14 in one or more viewing angles to obtain a new image data set 15. In this case, the image data 14 of the image data set 15 provide 2D views from different viewing angles, which can be displayed and checked, for example, by means of a display device 6.

[0056] If the image data remain consistent and the designer is satisfied with the appearance of the image data 14 at different viewing angles, design selection criteria 16 may be present, and the digital 3D model 5 may then be calculated in step S6 using a reconstruction algorithm. The reconstruction algorithm may be, for example, a further neural network that draws the 3D model 5 from the views of the image data 14 and / or may be back-calculated into the 3D model 5 by photogrammetry and / or differentiable rendering methods.

[0057] Finally, in step S7 , the created digital 3D model 5 can be provided as a design development for an object, in this case a vehicle body, wherein preferably a manufacturing device 8 can produce a real 3D model 9 , for example a plastic model produced by a 3D printer.

[0058] In general, the examples show that 3D geometric figures can be generated based on GAN through the present invention.

Claims

1. A method for providing a 3D model (5) for design development of an object, the method comprising the following steps: - prescribing (S2) an object type (18) for at least one trained artificial neural network (10), wherein the at least one artificial neural network (10) is trained using a plurality of basic image data (13) of objects of the object type (18); - creating (S3) an image data set (15) of a predetermined object type by means of a trained artificial neural network (10), wherein the artificial neural network (10) provides image data (14) of the object type (18) as the image data set (15) from a plurality of different viewing angles, which image data are determined from the learned basic image data (13); - checking (S4) the created image data set (15) to determine whether a design selection criterion exists, wherein if no design selection criterion exists (17), recreating the image data (14) of one or more viewing angles by means of modifying the image property criterion (S5); - in the case where a design selection criterion (16) is present, creating (S6) a digital 3D model (5) from the image data set (15) by means of a reconstruction algorithm, wherein the reconstruction algorithm calculates the digital 3D model (5) from image data (14) from a plurality of different viewing angles; - The created digital 3D model (5) is provided (S7) for design development of the object.

2. The method according to claim 1, characterized in that A plurality of artificial neural networks (10) are provided, wherein a corresponding artificial neural network (10) is trained for each predetermined viewing angle at an object with the aid of basic image data (13) of an object of the object type from the viewing angle.

3. The method according to claim 1, characterized in that The at least one artificial neural network (10) is trained with the aid of corresponding 2D views of 3D objects (13) of the object type, wherein the 2D views are provided from a plurality of preset viewing angles towards the corresponding 3D object.

4. The method according to any one of the preceding claims, characterized in that At least in the case where the image properties in multiple views of the image data are consistent, there is a design selection criterion (16).

5. The method according to any one of the preceding claims, characterized in that As the artificial neural network (10), a generative adversarial network, namely, GAN, is used.

6. The method according to any one of the preceding claims, characterized in that Image data (14) of one or more viewing angles are recreated by interpolating and / or recombining the image data (14) with the aid of altering the image attribute criteria.

7. The method according to any one of the preceding claims, characterized in that The image attribute criteria are changed by adapting the latent vector.

8. The method according to any one of the preceding claims, characterized in that Image properties that are adapted by changing the image attribute criteria include: - the size of the object; - the proportions of the object; - the style direction of the object; - the color of the object; -The shape of individual details of an object.

9. The method according to any one of the preceding claims, characterized in that An object and / or a real 3D model (9) is produced from the provided digital 3D model (5), in particular by means of a production device (8).

10. A design device (1) designed to carry out the method according to any of the preceding claims.

Images