Decorative panel

Abstract

The invention provides, amongst other aspects, a method for generating instructions for embossing a decorative panel, the method comprising: 1.a) obtaining a digital image comprising a texture, the digital image relating to the decorative panel, 1.b) determining a depth map based on the digital image, and 1.c) generating instructions based on the depth map for embossing the decorative panel.

Description

DECORATIVE PANELCross-reference to related application

[0001] This application claims priority to The Netherlands Patent Application No. 2036553 filed December 15, 2023, the disclosure of which is hereby incorporated by reference in its entirety.Field of the invention

[0002] The present invention relates to the technical domain of generating instructions for embossing of a decorative panel.Background art

[0003] Since recently, the variety in printed motifs has been improved by using digital printing techniques, such as inkjet printing, wherein a large image file, such as a scanned wood pattern-based image file is used to realize the printed motifs. The large image file is typically sufficiently large to cover 8 to 10 panels of regular size. This means that when 8 to 10 panels are packed in a package all panels may carry a unique printed motif. However, since a panel covering is typically composed of more than 10 panels, a repetition of printed motifs will still occur.

[0004] When reproducing natural materials, such as wood and stones, it is necessary to reproduce the superficial structure thereof, to obtain a material more like the original one also in the touch. Embossing is usually performed on the superficial layer and may be obtained by various methods, traditionally by means of pressing with molds, rollers, or belts on which the structure to be impressed is reproduced.

[0005] Modern scanners used to capture the image of materials, for example METIS DRS 2000, allow for the capture of the superficial structure that may be advantageously used for embossing in register.

[0006] Given the widespread use of embossing, it should have simplicity and costeffectiveness features.

[0007] It is therefore desirable to find new methods for generating instructions for embossing a decorative panel, which allow for quick, easy, cost-effective, and applicable embossing of surfaces of different materials.

[0008] The present invention aims at addressing issues, such as the issues mentioned above.Summary of the invention

[0009] According to a first aspect, the present invention provides a method for generating instructions for embossing at least one (physical) decorative panel, preferably a, typically rigid, floor panel or wall panel, preferably for composing adecorative panel covering composed of a plurality of said panels (in abutting and / or adjacent and / or interconnected state), the method comprising: a) obtaining at least one digital image comprising a texture, preferably wherein the digital image is related to at least one decorative panel, b) determining, preferably by a computer, at least one (digital) depth map, preferably at least partially based on at least one digital image, and c) generating instructions, preferably digital instructions, in particular computer- readable instructions, based on the depth map for embossing the at least one (physical) decorative panel and / or for generating at least one relief structure , in particular for applying an embossing structure in a decorative surface layer of said decorative panel.

[0010] It is also imaginable that the instructions generated during step c) are divided into a plurality of smaller instructions or sub-instructions, wherein said smaller or subinstructions are smaller or sub-instructions for embossing of different decorative panels. Each of said decorative panels may be provided with a smaller or subembossing, preferably wherein adjacent and / or side-by-side oriented decorative panels together for a single relief structure, in particular an embossing structure.

[0011] The method of the invention may advantageously provide a method for generating improved instructions for embossing of decorative panels. By determining a depth map of an image of the decorative panel to be embossed, the method ensures that the embossing can be unique to the decorative panel.

[0012] In particularly advantageous embodiments, the digital image has been generated and / or modified using a first generative model. Such particularly advantageous embodiments may provide an improved method for manufacturing decorative panels, particularly non-repetitive decorative panels, wherein a unique depth map is determined for each decorative panel for improved embossing thereof.

[0013] In particularly advantageous embodiments, the depth map is generated using the first generative model or alternatively a second generative model, based on the digital image. Such embodiments may advantageously provide an improved method for realizing more variety in decorative patterns or designs for decorative panels, while ensuring that the embossing of each of said decorative patterns is unique.

[0014] According to a second aspect, the present invention provides a device comprising means for carrying out the method according to the invention.

[0015] According to a third aspect, the present invention provides a computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to the invention. The computer program product ma comprise at least one readable medium in which computer-readable program code portions are saved, which program code portions comprise instructions for carrying out said method.

[0016] According to a fourth aspect, the present invention provides a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to the invention.

[0017] According to a fifth aspect, the present invention provides a system comprising the device according to the present invention, and embossing means for providing an embossing structure onto the physical decorative panel based on the generated embossing instructions.

[0018] Preferred embodiments and their advantages are provided in the description and the dependent claims.

[0019] According to a sixth aspect, the present invention provides a decorative panel obtained by carrying out the method according to the invention, preferably a set of decorative panels obtained by carrying out the method according to the invention, more preferably wherein each panel comprises at at least one pair of opposite edges coupling profiles allowing interlocking of adjacent panels.

[0020] According to a seventh aspect, the present invention provides a decorative covering, in particular a floor covering, wall covering, ceiling covering, or furniture covering, composed a set of, preferably interconnected and / or adjacent and / or abutting decorative panels according to the invention. In the context of this disclosure, said decorative covering composed of a plurality of decorative panels, preferably rigid decorative panels and / or preferably interconnected decorative panels (acting a prefabricated, modular components) is designed to be cover a, preferably entirely, subfloor, wall surface, ceiling surface and / or furniture surface, in a continuous manner. In case of a decorative floor covering, the decorative floor covering may or may not be affixed to a subfloor. In case of a decorative wall covering or ceiling covering or ceiling covering, the decorative covering is typically affixed, for example glued, to a structural subsurface, such as a wall surface, ceiling surface or furniture surface, respectively. The decorative panels serve both functional and aesthetic purposes, such as providing insulation, soundproofing, and enhancing visual appeal. The decorative covering is preferably a continuous decorative covering without gaps in between the decorative panels.

[0021] Brief description of the drawings

[0022] The present invention will be discussed in more detail below, with reference to the attached drawings.

[0023] Fig. 1A shows a first example of a method according to the invention.

[0024] Fig. 1 B shows a second example of a method according to the invention.

[0025] Fig. 2A and Fig. 2B show a third example of an embossed decorative panel according to the invention.

[0026] Fig. 3 shows a fourth example of a method according to the invention.

[0027] Fig. 4 shows a fifth example a decorative panel according to the invention.

[0028] Fig. 5 shows a sixth example of a method for producing a decorative panel according to the invention.Description of embodiments

[0029] The following descriptions depict only example embodiments and are not considered limiting in scope. Any reference herein to the disclosure is not intended to restrict or limit the disclosure to exact features of any one or more of the exemplary embodiments disclosed in the present specification.

[0030] Furthermore, the terms first, second, third and the like in the description and in the claims are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order. The terms are interchangeable under appropriate circumstances and the embodiments of the invention can operate in other sequences than described or illustrated herein.

[0031] Furthermore, the various embodiments, although referred to as “preferred” are to be construed as exemplary manners in which the invention may be implemented rather than as limiting the scope of the invention.

[0032] The term “comprising”, used in the claims, should not be interpreted as being restricted to the elements or steps listed thereafter; it does not exclude other elements or steps. It needs to be interpreted as specifying the presence of the stated features, integers, steps or components as referred to, but does not preclude the presence or addition of one or more other features, integers, steps or components, or groups thereof. Thus, the scope of the expression “a device comprising A and B” should not be limited to devices consisting only of components A and B, rather with respect to the present invention, the only enumerated components of the device are A and B, and further the claim should be interpreted as including equivalents of those components.

[0033] In this document, the term “generated design” refers to any design generated within a generative model, and the term “output design” refers to any digital design generated and output by the generative model. As will be understood, the output design relates to a generated design which has been accepted for output. This output design differs from the input image. The generated design preferably comprises and / or relates to a relief structure (embossing or embossing pattern) for one or more decorative panels. This may be considered as a 3D design. Additionally or alternatively, the generated design preferably comprises and / or is related to a decorimage for one or more decorative panels, which may lead to a planar (2D) design and / or to a spatial (3D) design. The generated design preferably comprises a relief structure image or relief structure file, and / or a decorative output image (decor image) or decorative output file, which serve(s) as basis to further finalize one or more decorative panels. However, it is also imaginable that the generated design comprises a plurality of decor image parts (image fractions) which together define a complete decor image. This may be advantageous, for example, in case the decor (image) should be construed via a plurality of decorative layers, each decorative layer being provided with at least one decor image part, for example in order to realize a depth effect and / or other special effect. This latter would lead to a three-dimensional (3D) design, in particular a 3D decor (image). As indicated above, it is additionally or alternatively imaginable that the output design comprises at least one design component other than a decor image, such as for example a relief structure (embossing structure) and / or relief image (embossing image) and / or one or more bevels and / or one or more grouts (artificial grout lines). Respectively, the term “generated depth map” refers to any depth map generated within a generative model, and the term “output depth map” refers to any depth map generated and output by the generative model. As will be understood, the output depth map relates to a generated depth map which has been accepted for output. This generated depth map may be based, but may alternatively differ from the input image. It is also imaginable that the generated depth map is mapped onto a single base panel or decorative panel. However, it is also conceivable that said depth map is mapped onto a plurality of base panels or decorative panels. It is additionally or alternatively imaginable that the depth map comprises at least one design component other than a relief structure reflecting a material texture or characteristic, such as for example one or more bevels and / or one or more grouts (artificial grout lines).

[0034] In this document, the term “texture” refers to the surface characteristics that can be perceived through touch and / or sight. For example, wood texture, or also known as wood grain, refers to the unique pattern or appearance of the fibers in wood. It is determined by the arrangement, size, and appearance of the wood cells. The grain is a result of the growth rings, knots, and other natural characteristics of the wood. In this document, the texture may merely relate to a decor, in particular a decorative image, which typically represents a 2D design (visual texture). This decor may e.g. be formed by a print or printed layer. Hence, the texture may be planar and flat. Optionally, said visual texture may have and / or create a 3D effect (depth effect). Additionally or alternatively the texture may also be a 3D design (tactile texture), such as a surface provided with one or more grooves and / or one or more recesses and / orone or more cavities. Optionally, the texture may encompass both options, wherein the texture comprises at least one decor image (2D texture or 2D design) and at least one, preferably at least partially corelated, relief structure (3D texture or 3D design) typically positioned on top of said decor image, for example to imitate the look and feel of a real-wood structure, optionally additionally provided with one or more bevels and / or one or more grout lines (and / or one or more other depth patterns).

[0035] In this document, the term “style” relates to an image filter and / or properties of an image and / or a relief related style, such as an embossing related style. Thus, a style transfer is a computer vision technology which relates to combining the content of a first image with the style of preferably a second image and / or transforms the first image (input) image based upon at least one obtained style, in particular at least one user-selected style. The relief related style may be of various nature, and relates to at least one relief related aspect, preferably at least one relief related creation or at least one relief related transformation of the texture, in particular 3D texture, of the digital image used during step 1.a). Such a relief related aspect may e.g. relate to the width, and / or length, and / or depth of cavities, such as grooves or other recesses, to be applied as embossing during step 1 .c), and / or to the sharpness or smoothness of edges defining cavities to be applied as embossing during step 1.c) and / or the location and / or size of one or more grout lines and / or one or more bevels, and / or one of more alternative user-selected, optionally user created, depth patterns. Optionally, the style to be applied to the digital input may be entirely user-defined. It is imaginable that a plurality of predefined styles is presented to a user from which said user may select at least one style, and optionally subsequently modify said selected at least one style. Examples of how a style is transferred when using a generative model will be described herein.

[0036] In this document, the term “depth map” refers to a two-dimensional (2D) image that contains information about the distance of surfaces in a scene from a viewpoint. It is often used in computer vision, computer graphics. In the depth map, each pixel corresponds to a point in the digital image or design, and the pixel value represents the distance from the viewpoint to the corresponding point. For example, the higher the texture, the brighter or lighter its representation in the depth map. Conversely, deeper textures appear darker. The terms depth map and relief image may be used interchangeably in this document. To this end, it may be advantageous in case the digital image is in grayscale mode and / or is converted into a grayscale mode. The depth map is preferably at least partially and more preferably entirely created by a computer. Human intervention may entirely be excluded here.

[0037] In this document, the term “decorative panel” refers to a panel having a decorative appearance, typically at its upper surface. Although the decorative panel may be a monolithic decorative panel (single layer decorative panel), the decorative panel may also be a laminated decorative panel comprising a laminate of a plurality of layers. The decorative panel may be a finished decorative panel or a partially finished decorative panel. The decorative panel optionally referred to in step 1.a) may be considered as a source decorative panel, which source decorative panel may either be a monolithic panel, such as a wooden board (wooden pane) and / or a stone tile (or stone slab), such as a marble tile (or marble slab), and / or a ceramic tile (or ceramic slab), etcetera. This source decorative panel may serve as basis for determining the depth map according to step 1.b) and to subsequently, based upon said depth map, generate instructions to realize an embossing (relief structure) in another decorative panel according to the 1.c). The decorative panel referred to in step 1.c) may be considered as a target decorative panel. Instead of using a decorative panel as source for a digital input image used during step 1 .a), the digital (input) image may also be unrelated to decorative panels and may e.g. be and / or be based upon at least one photo image. The decorative panel to be embossed, as referred to in step 1.c) may be considered as partially finished decorative panel and may be considered as finished once the embossing has been applied to said decorative panel, although further processing steps to finish the decorative panel, such as cutting and / or profiling the panel, are imaginable.

[0038] In this document, the term “embossing” refers as verb to the application of an embossing structure, also referred to as relief structure, and which may abbreviated as embossing (as noun). This embossing realizes a 3D pattern in a surface, typically, an upper surface of the decorative panel, and preferably in a decorative surface layer of said decorative panel. Such an embossing is typically applied to provide the decorative panel a more attractive and / or more realistic appearance, but may also be applied to provide the decorative panel with additional functionality, such as, for example, an improved anti-slip resistance and / or improved water drainage capacity. At least a part of the embossing may be designed and applied in register (in line) with a decorative image of the (target) decorative panel onto which the embossing is applied. This is for example attractive in case a wood pattern has to be imitated, wherein the decorative panel may comprise a printed decor of a wood pattern on top of which an embossing is applied or has to be applied, wherein said embossing comprises a plurality of impressions or cavities and / or grooves, which are in register with the wood nerves and wood pores of the printed wood pattern. The location and depth of the impressions, cavities, and / or grooves, being a function of the woodnerves and wood pores of the printed pattern gives an improved and more realistic look-and-feel effect to the eventual decorative panels. The embossing applied or to be applied may additionally or alternatively also comprises other 3D aspects, such as one or more bevels, and / or one or more grout lines and / or another pattern of cavities. Said decorative surface layer is typically affixed directly or indirectly onto a base panel of the decorative panel. Said decorative surface layer preferably comprises a plurality of sublayers, including a decorative print layer and at least one translucent or transparent protective layer covering said decorative print layer, wherein at least one protective layer, such as a wear layer, preferably comprises a relief structure which preferably is at least partially aligned with the decorative print layer. The decorative print layer preferably comprises a carrier film, such as a paper film or polymer film, onto which a decorative image is printed. The decorative surface layer may have a thickness less than 0.75 millimetre, or even less than 0.5 millimetre. It is also imaginable that the decorative print layer is printed directly onto the substrate and / or onto another layer, such as a primer layer and / or a, preferably white, basecoat layer already applied to the base panel.

[0039] In embodiments, step 1.a) comprises scanning or capturing an image of the decorative panel to be embossed. Thus, the method determines a depth map based on the scanned or captured digital image and generations instructions for embossing based on said depth map. This allows the method to provide improved embossing by ensuring that the embossing instructions are based on a depth map determined particularly for the decorative panel which is to be embossed. This can reduce the error in the embossing by ensuring that the embossing is relevant to the decorative panel.

[0040] In embodiments, the digital image comprises a pair of images of the decorative panel. The pair of images may have different viewpoints of the decorative panel. Preferable, step 1.a) comprises scanning or capturing a pair of images of the decorative panel. For example, the pair of images are captured simultaneously from slightly different viewpoints, mimicking the way human eyes perceive depth. Thus, a depth map can be more reliably determined based on the pair of images.

[0041] In embodiments, step 1.b) comprises applying a stereo matching algorithm to determine the depth map. The stereo matching algorithm may comprise at least one of: a block matching algorithm, a semi-global matching (SGM) algorithm, a graph cut algorithm, and a convolutional neural network (CNN). The block matching algorithm calculates differences (e.g., absolute or sum of squared differences) between corresponding pixels in the pair of images and selects the disparity that minimizes the sum of these differences. SGM considers the matching costs of pixels alongmultiple paths, including both horizontal and vertical directions. Graph cut algorithms, such as Graph Cuts Stereo (GCS) and Graph Cuts Belief Propagation (GC-BP), formulate stereo matching as an energy minimization problem. CNNs like DispNet, PSMNet (Pyramid Stereo Matching Network), and GC-Net use deep learning architectures to learn complex features and relationships in stereo image pairs.

[0042] In embodiments, step 1.b) comprises applying a CNN, such as U-Net or ResNet, to the digital image, preferably consisting of one digital image.

[0043] In particularly advantageous embodiments, step 1.a) comprises:2. a) obtaining at least one digital input image, preferably comprising a texture, 2.b) obtaining at least one style, and2.c) applying, by using at least one first generative model being associated with a corresponding style transfer, said at least one style to said at least one digital input image to generate at least one (decorative) output design, in particular at least one (decorative) digital output design and / or (decorative) digital output image, as the digital image for the at least one decorative panel, particularly the digital image obtained in step 2. a).

[0044] In particularly advantageous embodiments, step 2.b) comprises applying, using the first generative model, said style to said digital input image to generate the depth map for the decorative panel. Alternatively, step 2.b) comprises applying, using a second generative model being associated with a corresponding style transfer, said style to said digital input image to generate the depth map for the decorative panel.

[0045] In embodiments, the digital input image comprises an input depth map or an input relief image. Alternatively, the digital input image or the input relief image may be obtained separately from the digital input image. Thus, step 2.b) comprises applying the first generative model and / or the second generative model, said style to said input depth map to provide a generated depth map.

[0046] In embodiments, the first generative model and / or the second generative model are / is trained on a training dataset comprising scanned and / or captured images of decorative panels. The training dataset may comprise historically generated designs or digital images of decorative panels. The training dataset may comprise depth maps historically determined based on the respective images of decorative panels in the training dataset. The training dataset may comprise historically generated depth maps (and / or output depth maps) of the respective historically generated designs or digital images (and / or output designs). This can improve the training of the first and / or second generative model and to decrease the rate of generated depth maps and / or digital images which are discarded. For example, a generative model is trained on previously generated set of first digital images based on a first digital input imageand can improve the generating of a second digital image based on the first digital input image and / or a second digital input image. In another example, a generative model is trained on previously generated set of first depth maps of a first set of digital images based on a first digital input image and can improve the generating of a second depth map based on the first digital input image and / or a second digital input image. Thus, the set of second digital image and / or depth map can be more effectively generated.

[0047] In embodiments, the at least one style relates to at least one image related style chosen from the group of: shade, color temperature, brightness, contrast, exposure, brilliance, highlights, shadows, black point, saturation, vibrancy, tint, sharpness, definition, noise reduction, filter, etc.

[0048] In embodiments, the style transfer relates to a neural style transfer (NST) or a fast NST.

[0049] In embodiments, the style relates to an image filter chosen from the group of: mean filter, median filter, Gaussian smoothing, conservative smoothing, Crimmins speckle removal, frequency filters, Laplacian filter, Laplacian of Gaussian filter, and unsharp filter. Mean filtering is a simple, intuitive and easy to implement method of smoothing images, i.e. reducing the amount of intensity variation between one pixel and the next. It is often used to reduce noise in images. The median filter is normally used to reduce noise in an image, somewhat like the mean filter, although the image details are preserved in an improved manner by using the median filter. The Gaussian smoothing operator is a 2-D convolution operator that is used to make image blurrier and to remove detail and noise. In this sense it is similar to the mean filter, but it uses a different kernel that represents the shape of a Gaussian ('bellshaped') hump. Conservative smoothing is a noise reduction technique that derives its name from the fact that it employs a simple, fast filtering algorithm that sacrifices noise suppression power in order to preserve the high spatial frequency detail (e.g. sharp edges) in an image. It is explicitly designed to remove noise spikes, being isolated pixels of exceptionally low or high pixel intensity (e.g. salt and pepper noise) and is, therefore, less effective at removing additive noise (e.g. Gaussian noise) from an image. Crimmins Speckle Removal reduces speckle from an image using the Crimmins complementary hulling algorithm. The algorithm has been specifically designed to reduce the intensity of salt and pepper noise in an image. Increased iterations of the algorithm yield increased levels of noise removal, but also introduce a significant amount of blurring of high frequency details. Frequency filters process an image in the frequency domain. The image is Fourier transformed, multiplied with the filter function and then re-transformed into the spatial domain. Attenuating highfrequencies results in a smoother image in the spatial domain, attenuating low frequencies enhances the edges. All frequency filters can also be implemented in the spatial domain and, if there exists a simple kernel for the desired filter effect, it is computationally less expensive to perform the filtering in the spatial domain. Frequency filtering is more appropriate if no straightforward kernel can be found in the spatial domain, and may also be more efficient. The Laplacian filter is a 2-D isotropic measure of the 2nd spatial derivative of an image. The Laplacian of an image highlights regions of rapid intensity change and is therefore often used for edge detection (see zero crossing edge detectors). The Laplacian is often applied to an image that has first been smoothed with something approximating a Gaussian smoothing filter in order to reduce its sensitivity to noise, and hence the two variants will be described together here. The unsharp filter is a simple sharpening operator which derives its name from the fact that it enhances edges (and other high frequency components in an image) via a procedure which subtracts an unsharp, or smoothed, version of an image from the original image. The unsharp filtering technique is commonly used in the photographic and printing industries for crispening edges.

[0050] In embodiments, the at least one style relates to at least one relief related style chosen from the group of: relief structure, relief structure texture, relief structure type, relief structure layout, and relief structure size. Said relief structure, also referred to as embossing structure, may match the visuals of the respective output design at least partially, preferably entirely. The relief related style may lead to generating an additional digital file or instructions or an output relief image, e.g., an output depth map as described herein. Additionally or alternatively, (at least a part of) the relief structure may define one or more bevels and / or one or more grouts, more preferably, the number and / or the location and / or the size and / or the colour and / or the surface texture (roughness), of one or more bevels and / or one or more grouts to be realized in the decorative panel(s) to be formed. Optionally, these grout and / or bevel related options and / or another style and / or style options and / or embossing options may be partially or entirely customizable by a user by using a, preferably web-based, user interface directly or indirectly related to a computer at a merchant or manufacturer.

[0051] In embodiments, the style is obtained in the form of respective style vector representation.

[0052] In embodiments, the first generative model and / or the second generative model each comprises at least one likelihood based generative model. Such models may directly learn the distribution’s probability density (or mass) function via (approximate) maximum likelihood. Examples of such models include autoencoders,autoregressive models, energy-based models, and normalizing flow models. Such models will be explained herein.

[0053] In embodiments, the first generative model and / or the second generative model comprise at least one implicit generative model. Such models implicitly represent the probability distribution. Examples of such models include generative adversarial networks (GANs) and approximate Bayesian computation (ABC).

[0054] In embodiments, the first generative model and / or the second generative model each comprises at least one score-based generative models. Such models have connections to normalizing flow models, therefore allowing exact likelihood computation and representation learning. Examples of such models include latent score-based generative models and sliced score matching.

[0055] In embodiments, the first generative model and / or the second generative model each comprises at least one autoencoder (AE). Preferably, the at least one AE comprises at least one variational AE (VAE) and / or at least one adversarial AE (AAE). In embodiments, the first generative model and / or the second generative model comprise at least one traditional AE and / or at least one VAE and / or at least one AAE. For example, the first generative model and / or the second generative model comprise at least one traditional AE, at least one VAE and at least one AAE. AEs comprise an encoder and a decoder and are effective in feature learning and dimensionality reduction, capturing essential features in data. VAEs provide a probabilistic framework for generative modeling, offering a clear interpretation of uncertainty in generated samples and are effective in generative diverse designs and / or depth maps. AAEs incorporate adversarial training principles, introducing a discriminator to guide the encoding process, which are more robust to noise, resulting in more realistic and sharper-looking generated samples.

[0056] Said at least one AE may each be associated with a corresponding style transfer. One way may be to manipulate values in a latent space learned by the AE, thereby influencing the style of the reconstructed output design and / or depth map. For example, if an AE is trained on images of wood patterns, changing certain dimensions in the latent space could alter textural features, influencing the style of the generated wood-pattern design and / or depth map. Another way may be to introduce conditional information to the autoencoder's input, thereby allowing for the generation of output design and / or depth map with a specific style. For instance, the AE is conditioned on a particular style label during training, the AE could learn to generate the output design and / or depth map conditioned on the specified style.

[0057] In embodiments, the first generative model and / or the second generative model each comprises at least one autoregressive model. Preferably, the at least oneautoregressive model comprises at least one recurrent neural network (RNN), e.g., PixeIRNN, and / or at least one convolutional neural network (CNN), e.g., PixeICNN, both of which are autoregressive models that use particular neural networks to generate images pixel by pixel.

[0058] In embodiments, the first generative model and / or the second generative model each comprises at least one traditional autoregressive model and / or at least one RNN and / or at least one CNN. In embodiments, the first generative model and / or the second generative model are / is from the group of: at least one AE and at least one autoregressive model. Preferably, the first generative model and / or the second generative model are / is from the group of: at least one AE, at least one VAE, at least one AAE, at least one autoregressive model, at least one RNN, and at least one CNN.

[0059] Said at least one autoregressive model may each be associated with a corresponding style transfer. One way may be to condition autoregressive models on specific styles during the generation process. This involves providing additional information (conditioning variables) that guides the generation of the output design and / or depth map in a style-specific manner. For example, autoregressive models generating wood-pattern output design and / or depth map can be conditioned on a particular style. In another example, style information may be embedded directly into the input or hidden states of the autoregressive model. This embedding can influence the generation process and result in output designs and / or depth maps with different stylistic characteristics. For example, style-related features can be introduced into the initial hidden state or modifying the conditioning of subsequent steps.

[0060] In embodiments, the first generative model and / or the second generative model each comprises at least one normalizing flow model (NFM). Preferably, the at least one NFM comprises at least one real normalizing flow (Real NVP) and / or at least one generative latent optimization (Glow) and / or at least one planar flow and / or at least one radial flow and / or at least one masked autoregressive flow (MAF) and / or invertible autoregressive flow (IAF). NFMs typically learn a probability distribution over data by transforming a simple base distribution through a series of invertible transformations. NFMs have been shown to capture intricate dependencies in data.

[0061] In embodiments, the first generative model and / or the second generative model each comprises at least one traditional NFM and / or at least one Real NVP and / or at least one Glow and / or at least one planar flow and / or at least one radial flow and / or at least one MAF and / or IAF.

[0062] Said at least one NFM may each be associated with a corresponding style transfer. One way may be to calculate the style transfer loss by comparing the stylefeatures of the generated output design and / or depth map with those of the style reference image (e.g., the digital input image and / or the style).

[0063] In embodiments, the first generative model and / or the second generative model each comprises at least one energy-based model (EBM). Preferably, the at least one EBM comprises at least one Hopfield network and / or at least one (restricted) Boltzmann machine and / or at least one Markov random field and / or at least one Gaussian mixture model (GMM). EBMs typically define a probability distribution over data by associating an energy function with each data point.

[0064] In embodiments, the first generative model and / or the second generative model each comprises at least one traditional EBM and / or at least one Hopfield network and / or at least one (restricted) Boltzmann machine and / or at least one Markov random field and / or at least one GMM. The at least one EBM may comprise or may be combined with at least one score-based model and / or at least one NFM model and / or at least one autoregressive model, such as, Real NVP and PixeICNN.

[0065] Said at least one EBM may each be associated with a corresponding style transfer. One way may be to incorporate a style loss term into the energy function of the EBM, which measures the difference between the style features of the generated image and a reference image (e.g., the at least one digital input image and / or the at least one style).

[0066] In embodiments, the first generative model and / or the second generative model each comprises at least one score-based generative model (SGM). Preferably, the at least one traditional SGM comprises at least one latent SGM (LSGM) and / or at least one latent score-based generative adversarial network (SB-GAN) and / or at least one sliced score matching (SSM) algorithm and / or at least one amortized Stein variational gradient descent (ASVGD) algorithm. SGMs typically do not require explicit modeling of the probability density function, which can be advantageous when the true data distribution is complex or unknown. Furthermore, SGMs can potentially model a wide range of data distributions, including high-dimensional and multimodal distributions. This may be advantageous for wood patterns of different wood species. LSGMs extend the scored-based approach by incorporating latent space variables, which allows for modelling of more complex and hierarchical data structures.

[0067] In embodiments, the first generative model and / or the second generative model each comprises at least one traditional SGM and / or at least one LSGM and / or at least one SB-GAN and / or at least one SSM and / or at least one ASVGD. In embodiments, the first generative model and / or the second generative model are / is from the group of: at least one AE, at least one autoregressive model, at least oneEBM, at least one NFM, and at least one SGM. Preferably, the first generative model and / or the second generative model are / is from the group of: at least one traditional AE, at least one VAE, at least one AAE, at least one traditional autoregressive model, at least one RNN, and at least one CNN, at least one traditional SGM, at least one LSGM, at least one SB-GAN, at least one SSM, at least one ASVGD.

[0068] Said at least one SGM may each be associated with a corresponding style transfer. One way may be to train the SGM to capture the gradient or score of the data distribution for different stylistic features. This may involve training the model on a dataset with samples representing various styles. Another way may be to condition the model on specific style information during the generation process. For example, by modifying the input or hidden states of the model to incorporate information about the desired style. Another way may be by manipulating the learned score or gradient information to influence the stylistic characteristics of the generated samples. For example, by modifying the score in specific directions that correspond to different styles. Another way may be to generate samples using the modified score information, either through Langevin dynamics or other sampling methods associated with score-based models. Another way may be to evaluate the generated output design and / or depth map for their adherence to the desired style. For example, by fine-tuning the model or adjust the conditioning mechanism based on the evaluation results.

[0069] In embodiments, the first generative model and / or the second generative model each comprises at least one generative adversarial network (GAN), wherein the first generative model and / or the second generative model comprises respectively at least one first generator and / or at least one second generator, wherein the first generative model and / or the second generative model comprises respectively at least one first discriminator and / or at least one second discriminator. Preferably, the at least one first generator comprises two or more first generators and / or the at least one second generator comprises two or more second generators. Preferably, the at least one first discriminator comprises two or more first discriminators and / or the at least one second discriminator comprises two or more second discriminators.

[0070] In embodiments, each of the at least one first generator and / or at least one second generator comprises at least one deep CNN (DCNN) generator and / or at least one earthmover’s distance-based generator (e.g., Wasserstein-based generator) and / or at least one U-shaped encoder-decoder network (UNet)-based generator and / or at least one visual geometry group (VGG)-based generator and / or at least one residual network (ResNet)-based generator and / or at least one SB generator and / or at least one progressive generator. In embodiments, the at leastone GAN comprises at least one DCNN-GAN and / or at least one Wasserstein-GAN and / or at least one UNet-GAN and / or at least one VGG-GAN and / or at least one ResNet-GAN and / or at least one SB-GAN and / or at least one Pro-GAN. The DCNN generator allows for the model to learn spatial hierarchies and patterns in the data. The Wasserstein-based generator allows for better stability in training of the model. The ll-Net based generator includes both encoder and decoder paths, which allows for effective image segmentation and style transfer. The VGG-based generator can capture rich features which allow for image enhancement and improved style transfer. The ResNet-based generator allows for the generation of more complex and detailed images. The SB generator allows to control both high-level and low- level features in the generated images. The progressive generator starts with lower resolution and gradually adds layers to allow for generating higher-resolution images while maintaining stability during training of the model.

[0071] In embodiments, the first generative model and / or the second generative model are / is from the group of: at least one AE, at least one autoregressive model, at least one EBM, at least one NFM, at least one SGM, and at least one GAN. Preferably, the first generative model and / or the second generative model are / is from the group of: at least one traditional AE, at least one VAE, at least one AAE, at least one traditional autoregressive model, at least one RNN, and at least one CNN, at least one traditional SGM, at least one LSGM, at least one SB-GAN, at least one SSM, at least one ASVGD, at least one DCNN-GAN and / or at least one W-GAN and / or at least one UNet-GAN and / or at least one VGG-GAN and / or at least one ResNet-GAN and / or at least one SB-GAN. In one example, the SB-GAN is an embedded combination of an SB and a GAN. In another example, the first generative model and / or the second generative model comprise a first and second models, the first model being a traditional SB model and the second model being a traditional GAN.

[0072] Said at least one GAN may each be associated with a corresponding style transfer. One way may be to introducing a style encoder (e.g., a separate NN) that can encode the style of an input image or a reference image. Another way may be to introduce a style loss term in the overall loss function, which measures the difference between the style of the generated image and the style of a reference image.

[0073] In embodiments, the at least one first discriminator and / or the at least one second discriminator each comprises at least one patch-wise-based discriminator and / or at least one DCNN discriminator and / or at least one distance-based discriminator. The patch-wise discriminator divides an image into non-overlapping patches and applies convolutional layers to each patch, allowing to capture details and textureseffectively. The DCNN discriminator can extract hierarchical features from images and effectively capture spatial dependencies in the images. The distance-based discriminator may provide a measure of dissimilarity between real and generated images or designs, which can be more suitable for tasks where the emphasis is on the quality and diversity of the generated output designs. Additionally or alternatively, the distance-based discriminator may provide a measure of dissimilarity between real and generated depth maps, which can be more suitable for tasks where the emphasis is on the quality and diversity of the generated depth maps.

[0074] In embodiments, the at least one distance-based discriminator comprises at least one earth mover’s distance discriminator and / or at least one total variation distance discriminator. The earth mover’s distance discriminator is similar to the Wasserstein discriminator which quantifies dissimilarity between two probability distributions, which allows for improved design and / or depth map quality discrimination. The total variation distance discriminator uses statistical distance measures, such as also for measuring between probability distributions.

[0075] In embodiments, each of the at least one first discriminator and / or at least one second discriminator comprises at least one least squares loss function and / or at least one hinge loss and / or at least one Jensen-Shannon divergence loss function. The least square loss function discriminator output real-valued score and the generator aims to minimize the squared difference between these scores and the target values. The hinge loss function penalizes the model more severely for incorrectly classified designs and / or depth maps, leading to more stable training. The Jensen-Shannon divergence loss function uses a symmetric measure, unlike the non-symmetric Kullback-Leibler (KL) divergence in the Wasserstein discriminator.

[0076] In embodiments, each of the at least one first discriminator and / or at least one second discriminator comprises at least one multi-scale discriminator and / or at least one feature-matching discriminator and / or at least one ranking discriminator and / or at least one Gaussian mixture model (GMM) discriminator. The multi-scale discriminator may be preferably used in the context of Pro-GANs and involve using discriminators at multiple resolutions or scales, wherein each discriminator is responsible for assessing the realism of the generated design and / or depth map at a particular scale or resolution, thereby enabling to produce more realistic and high- resolution design and / or depth map. The feature-matching discriminator involves training the discriminator to provide additional feedback on the features and / or representations extracted from the input data, which stabilize the training and thereby allowing to produce more diverse and realistic designs. The ranking discriminator involves ranking generated designs in terms of their realism and assessing therelative quality of the designs, which allows to produce improved design quality and diversity. The GMM discriminator evaluates generated designs based on their likelihood under the GMM model, which allows to produce output designs from different modes of the data distribution.

[0077] In embodiments, step 2.c) comprises detecting that a similarity between the generated design and the digital input image is within a predetermined similarity range. This is particularly advantageous for generative models that are not GAN. The generated designs may be discriminated by comparing them to the digital input image. Additionally or alternatively, step 2.c) comprises detecting that a similarity between the generated depth map and the input depth map comprised in the digital input image is within a predetermined similarity range. This is particularly advantageous for generative models that are not GAN. The generated depth map may be discriminated by comparing to the input depth map in the digital input image. Additionally or alternatively, step 2.c) comprises detecting that a similarity between the generated depth map and the input depth map is within a predetermined similarity range. This is particularly advantageous for generative models that are not GAN. The generated depth map may be discriminated by comparing to the input depth map.

[0078] In embodiments, step 2.c) comprises detecting that a similarity between any of the generated designs is within the predetermined similarity range. This is particularly advantageous to any generative model, including GAN that are not (or even are) trained on historically generated and / or output designs. Since a set of decorative panels may include many panels, there is a risk that any of these panels are either too similar and / or too dissimilar, which is not wanted. Thus, each of the generated designs may be discriminated by comparing them to another of the generated designs. This ensures that the generated designs are not too similar to and / or dissimilar from each other, while being accepted as realistic designs. Additionally or alternatively, step 2.c) comprises detecting that a similarity between any of the generated depth maps is within the predetermined similarity range. This is particularly advantageous to any generative model, including GAN that are not (or even are) trained on historically generated and / or output depth maps. Since a set of decorative panels may include many panels, there is a risk that any of these panels are either too similar and / or too dissimilar, which is not wanted. Thus, each of the generated depth maps may be discriminated by comparing them to another of the generated depth maps. This ensures that the generated depth maps are not too similar to and / or dissimilar from each other, while being accepted as realistic depth maps.

[0079] In embodiments, step 1.b) and / or step 2.c) comprises detecting that a similarity between the generated depth map and the relief-related style is within a predetermined range.

[0080] In embodiments, step 1.b) and / or step 2.c) comprises applying, using the first generative model and / or the second generative model, said relief-related style to said digital input image or to said input depth map. Preferably, step 1.b) and / or step 2.c) comprises detecting that a similarity between the generated depth map and the input depth map is within a predetermined range.

[0081] In embodiments, step 1.b) and / or step 2.c) comprises applying, using the first generative model and / or the second generative model, said style to said digital input image to generate a plurality of depth maps for a plurality of decorative panels. Preferably, step 1.b) and / or step 2.c) comprises detecting that a similarity between the generated depth maps is within a predetermined range.

[0082] In embodiments, the similarity is detected based on at least one of: a structural similarity index, mean squared error, peak signal-to-noise ratio, cosine similarity, Euclidean distance, correlation coefficient, KL divergence, and earth mover’s distance. In embodiments, the similarity metric is different from the metric used by the discriminator. This can ensure that a generated design and / or generated depth map that is accepted by the discriminator, but that is too dissimilar to any other generated design in the set of generated designs and / or to any other generated depth map in the set of generated depth maps, respectively, would be discarded. Therefore, the combination of the at least one discriminator and at least one similarity detector can ensure that realistic designs are generated and at the same time improve the generated set of designs and / or depth maps such that the designs and / or the depth maps have a similarity that is within a predetermined similarity range, and thereby, ensuring that a set of uniform designs and / or depth maps is provided. In a first example, the generated depth maps are provided to the at least one discriminator and to the at least one similarity detector in parallel. In a second example, the generated depth maps are provided first to the at least one discriminator, wherein the accepted generated depth maps (i.e., the realistic depth maps) are provided to the at least one similarity detector. In a third example, the generated depth maps are provided first to the at least one similarity detector, wherein the accepted depth maps (i.e., within the predetermined similarity range) are provided to the at least one discriminator.

[0083] In embodiments, the discrimination and the similarity detection are merged into the at least one discriminator and / or the at least one similarity detector. For example, the at least one similarity detector detects which of the generated depth maps arerealistic (i.e., above a predefined threshold or within a first predetermined similarity range) and which are within a second predetermined similarity range.

[0084] In embodiments, step 1.a) and / or step 2. a) comprises obtaining at least one digital image and / or at least one digital input image from an image database and / or a camera and / or a scanner. The digital input image may have at least one texture that is desired in decorative panels, such as a wood texture, a tile texture, concrete texture, marble texture, mosaic texture, artificial texture etc. This can allow the first generative model and / or the second generative model to generate a design and / or depth map based on the patterns of the digital input image.

[0085] In embodiments, the image database comprises a plurality of scanned woodpattern images and / or scanned tile-pattern images and / or scanned concrete-pattern images and / or captured wood-pattern images and / or captured tile-pattern images and / or captured concrete-pattern images. This can ensure that the at least one digital input image is selected from a set of images which are high-quality scans (i.e., high resolution) and / or relate to a single type of pattern, e.g. wood pattern. Furthermore, the database may include highly desired images, which can improve the selection of the digital input image more.

[0086] In embodiments, the database comprises any accepted generated design and / or any output design. In preferred embodiments, the database comprises historically generated designs of decorative panels particularly output designs. This can ensure that new designs can be more efficiently obtained by selecting said new designs from the database.

[0087] In embodiments, the image database comprises any accepted generated depth maps and / or any output depth maps, preferably with their respective digital images and / or generated output designs. The input depth map may be obtained from the image database. This can allow the first generative model and / or the second generative model to generate a design and / or depth map based on the patterns of the input depth map.

[0088] In embodiments, the first generative model and / or the second generative model are / is trained on a training dataset comprising scanned and / or captured images of decorative panels, such as a plurality of scanned wood-pattern images and / or scanned tile-pattern images and / or scanned concrete-pattern images and / or captured wood-pattern images and / or captured tile-pattern images and / or captured concrete-pattern images. This can ensure that the first generative model and / or the second generative model are / is trained on high-quality images and / or images with highly desired patterns. Thus, the first generative model and / or the second generative model can generate more realistic designs with highly desired patterns.Alternatively or additionally, the training dataset may comprise any accepted generated design and / or any output design.

[0089] In embodiments, the training dataset comprises historically generated designs of decorative panels. This can improve the training of the first generative model and / or the second generative model and to decrease the rate of discarded designs. For example, a generative model is trained on previously generated set of first designs based on a first digital input image and can improve the generating of a set of second designs based on the first digital input image and / or a second digital input image. Thus, the set of second designs can be more effectively generated.

[0090] In preferred embodiments, the training dataset comprises depth maps historically determined based on the respective images of decorative panels in the training dataset. The training dataset may comprise historically generated depth maps (and / or output depth maps) of the respective historically generated designs or digital images (and / or output designs). This can improve the training of the first and / or second generative model and to decrease the rate of generated depth maps and / or digital images which are discarded. For example, a generative model is trained on previously generated set of first depth maps of a first set of digital images based on a first digital input image and can improve the generating of a second depth map based on the first digital input image and / or a second digital input image. Thus, the set of second digital image and / or depth map can be more effectively generated.

[0091] In embodiments, a first generated output design (i.e. , a design which is accepted as realistic design) is provided as feedback to the respective of the first generative model and / or the second generative model, whereby the respective of the first generative model and / or the second generative model generates a second generated design which is different from the first generated output design. The similarity between the second generated design and the first generated output design may be within a predetermined similarity range. Such feedback may be combined with a similarity detection and / or a discriminator after the second design is generated.

[0092] In preferred embodiments, a first generated output depth map (i.e., a depth map which is accepted as realistic depth map) is provided as feedback to the respective of the first generative model and / or the second generative model, whereby the respective of the first generative model and / or the second generative model generates a second generated depth map which is different from the first generated output depth map. The similarity between the second generated depth map and the first generated output depth map may be within a predetermined similarity range. Such feedback may be combined with a similarity detection and / or a discriminator after the second depth map is generated.

[0093] In embodiments, the method comprising detecting or classifying that the generated design is a real-looking design by determining the similarity between the generated design and at least one of: any of the at least one input image, any image and / or historically generated design (and / or output design) in the database and any image and / or historically generated design (and / or output design) in the training dataset. Said detecting or classifying may be performed by the at least one discriminator and / or the at least one similarity detector, as described herein.

[0094] In preferred embodiments, the method comprising detecting or classifying that the generated depth map is a real-looking depth map by determining the similarity between the generated depth map and at least one of: the digital input image, the input depth map, any historically generated design (and / or output design) in the database, any historically generated depth map (and / or output depth map) in the database, any historically generated design (and / or output design) in the training dataset, and any historically generated depth map (and / or output depth map) in the training dataset. Said detecting or classifying may be performed by the at least one discriminator and / or the at least one similarity detector, as described herein.

[0095] In embodiments, the method comprises:2.d) printing, preferably digitally printing, and / or otherwise applying at least a part of the output design, preferably generated during step c), directly or indirectly, onto at least one base panel of a decorative panel (to be formed), and2.e) preferably repeating step 2.d) to form the set of decorative panels and / or dividing the base panel prepared during step 2.d) into the set of decorative panels.The repeating of step 2.d) may be performed in a particular order, such that pattern or image repetition is avoided. The dividing may be performed, e.g. by cutting or sawing.

[0096] Each base panel may be provided, in particular printed, with a single smaller design or may be provided, in particular printed, with a plurality of smaller designs, preferably oriented side-by-side and / or adjacent to each another. Each base panel comprises and / or may form a core of the decorative panel (to be formed). Said core may be a monolithic layer (single layer) or may be a multi-layer core. Said base panel may comprise one or more further layers attached to a bottom surface of the core, such as a backing layer, and / or to a top surface of the core, such as a primer layer and / or basecoat layer, preferably white basecoat layer, to facilitate and / or improve the application of the design generated during step c), and / or to improve the aesthetical appearance of the decorative design generated during step c) once applied onto said further layer, such as the primer layer and / or (white) basecoat layer of the base panel. The printed or otherwise applied design is considered as adecorative layer of the decorative panel. Such a decorative layer may be a single layer and / or may be composed out of a plurality of layers. In case of a multi-layer decorative layer, it is imaginable that at least a part of the design generated during step c) is present in a first decorative layer and at least a part of the design generated during step c) is present in at least one further decorative layer. For example, said first decorative layer may be formed by a decor image, preferably digitally printed, decor image, while another decorative layer is provided with another part of the design generated during step c), such as a relief structure, preferably aligned with said decor image. It is also imaginable that the decor image is spread over various decorative layer applied on top of each other, which may realize e.g. depth effects and / or another special effect, such as e.g. a glitter effect and / or a position-selective matt and / or glossy effect. This may lead to a more realistic imitation of natural materials, such as wood, stone, in particular marble or sliced stone, and / or other materials, such as mosaic. The decorative layer may be planar and / or may be provided with a relief structure, such as an embossing structure and / or one or more bevels and / or one or more grouts. It is imaginable that a bottom surface of the decorative layer is planar, while at least a part of an opposite top surface of the decorative layer is provided with a relief structure. It is imaginable that the eventual decorative panel comprises one or more layers applied on top of said decorative layer, such as one or more transparent and / or translucent wear layers and / or one or more transparent and / or translucent coating layers. One or more of these layers, applied on top of the decorative layer, may bear a part of the design generated during step c), such as, for example, a relief structure aligned with a decor image forming the decorative layer and / or position-selectively applied matt and glossy areas aligned with a decor image forming the decorative layer.

[0097] In preferred embodiments, a part of the respective output designs is printed onto the respective of a plurality of base panels of the decorative panels to be formed.

[0098] The step of printing may be repeated to form the set of decorative panels. If after performing step 2.d) once, the base panels form the set of decorative panels, then step 2.d) may be performed once.

[0099] In embodiments, the method comprises obtaining a panel installation pattern, wherein the output designs generated during step 2.c) are based upon the panel installation pattern. Thus, the output designs, preferably the decorative panels, can be aligned based on the panel installation pattern. For example, in case the designs represent wood grain patterns with wood nerves, the wood nerves (main) direction and the longitudinal axis of the panels may be aligned with each other. This mayobviously also be applied to alternative decorative patterns having longitudinal (oblong) lines of sight characterizing said patterns.

[0100] In embodiments, the method further comprises obtaining surface area information relating to a surface area to be covered by an assembly of decorative panels, wherein the output design generated is sized to fit at least said surface area, preferably sized to fit a larger surface area than the obtained surface area. For example, a user may indicate that a surface area to be covered by decorative panels, wherein at least one decorative panel is made by making use of the method according to the invention, measures A x B square meters. Based upon these dimensions, a computer may calculate the number of panels needed to cover said surface area, based upon a predefined or chosen size of the decorative panels and / or installation pattern of said decorative panels. Based upon this surface area, the number of output designs and / or the number of output designs needed to realize a decorative panel is / are preferably calculated (by means of a computer) in order to realize a decorative covering to cover said surface area. In case this number is 1 or smaller, the output design can be sized, in particular cropped. In case this number is larger than 1 a plurality of output designs will have to be used to manufacture the entire decorative panel covering. Preferably, mutually different output images are used in this case to prevent pattern repetition. It is imaginable that at least one decorative panels comprises and / or is based upon a plurality of output designs. It is also imaginable that each output design is printed on a large base panel, in particular a slab, which is divided, in particular cut, into a plurality of single decorative panels after applying the output design to said large base panel, in particular said slab.

[0101] In embodiments, the output designs and / or output depth maps are generated in a particular order and / or in a particular layout. In embodiments, output designs and / or output depth maps are provided for printing in a particular order and / or in a particular layout. The order and / or the layout may be determined based on the similarity / dissimilarity of the output designs and / or output depth maps and / or to the surface area. The order and / or the layout may be determined based on the panel installation pattern. For example, output designs having the highest similarity can be printed on decorative panels which will be installed farther apart than other decorative panels in the set of decorative panels.

[0102] In embodiments, a decorative panel is a physical decorative panel, preferably comprising a core and a decorative layer. The core is referred to herein as a base panel and may comprise an upper layer and a lower layer. The decorative layer may be, either directly or indirectly, affixed and / or printed onto the upper layer of the core.The decorative layer of every decorative panel according to the invention is unique by the generated output design of the at least one generative model.

[0103] In embodiments, step 2.d) comprises applying at least one protective layer, after printing at least a part of said at least one output image, onto the at least one base panel, preferably after providing the relief structure. The protective layer may be affixed covering the decorative layer.

[0104] In embodiments, the method comprises step 1.d) of providing or applying at least one relief structure, in particular an embossing structure, e.g. after printing at least a part of the output design, onto the base panel. For example, the at least one relief structure is provided onto, and / or formed in, the decorative layer. The relief structure is provided or applied based on the embossing instructions generated in step 1.c). This can improve the visual appearance of the decorative panel by providing nerves and / or pores, which gives an improved and more realistic look-and-feel effect to the decorative panels. The relief structure may be provided onto the protective layer. This can provide a compromise between providing further improved and even more realistic look-and-feel effect to the decorative panels while comprising the protection of the relief structure. Typically, the relief structure, in particular the embossing structure, may be applied into an wear-resistive layer (wear layer) and / or an abrasion resistive layer serving to protect the decorative layer. Optionally, said relief structure is covered by at least one coating, such as a lacquer layer.

[0105] In embodiments, the relief structure is provided on a carrier layer, preferably a primer layer, provided on the decorative panel. The carrier layer may be provided on the printed decorative panel, i.e. on the printed layer. The carrier layer may comprise a pattern of at least one primer, preferably two or more primers, such as a mat primer and a glossy primer.

[0106] In embodiments, the relief structure is at least partially formed by a printed structure and / or can be formed by making use of a printed structure, preferably using transparent and / or translucent ink. Preferably, at least a part of the relief structure is provided by printing an embossing liquid, e.g. by means of an inkjet printhead, preferably wherein transparent and / or translucent ink is used. This allows the underlying decor to remain visible. The relief structure can be composed of a single ink layer or of a plurality of ink layers. The relief structure can be formed, for example, by printing ink droplets at locations wherein elevations in the relief structure are desired. Additionally or alternative, the relief structure can also be formed, for example, by printing ink droplets containing a curing inhibitor onto a liquid base layer to be cured, at locations where recesses, such as grooves, indentations, or other cavities, in the relief structure are desired. During curing the curing, such as UVcuring and / or excimer curing, inhibitor inhibits (and even prevents) curing of the covered parts of said liquid layer, while uncovered parts of said liquid BASE layer will be become cured (hardened), after which the still liquid parts are removed, for example by mechanical brushing, resulting in the eventual relief structure (see also Figure 5).

[0107] In embodiments, the relief structure comprises a continuous printed layer, preferably comprising and / o defining a texture and / or indentations and / or protrusions, e.g., wood grain and tile grout, etc.

[0108] In embodiments, the relief structure is provided in two or more distinct steps, and this allows, for example, to obtain combined effects by using two embossing liquids.

[0109] In embodiments, the relief structure is at least partially composed of a printed base layer. Preferably, the base layer may be a printed base layer, more preferably a digitally printed base layer. However, the liquid base layer is often applied in liquid form by means of roller coating. This means that the base layer, initially in liquid state, is applied either directly or indirectly on top of the decorative layer and my serve to realize the relief structure as described above. The depth of the relief structure may vary across the relief structure. The relief structure may have a depth up to 200 micron, or even larger (if desired).

[0110] In embodiments, the method comprises providing at last one UV curable layer, after printing at least a part of the output designs, onto the respective of the base panels. The UV-curable layer(s) may be attached to the decorative panel on top of the decorative layer and may for example be formed by a wear layer and / or a protective top coating. Typically the UV-coating or layer is applied in liquid state and is cured afterwards by means of a UV light source, such as a mercury lamp, a LED light source, an excimer light source, and / or combinations thereof. Other curable layers of the decorative panel may be cured in the same way. Different UV light sources may lead to different gloss levels. For example, using a mercury light source leads to a more glossy effect than using an excimer light source which will lead to a more matt effect. It is imaginable that different gloss levels are created in a single layer and / or in different layers, wherein the gloss level created may at least partially be dependent on the at least one output design used to realize at least one decorative layer of the decorative panel.

[0111] In embodiments, the method comprises providing a backing layer which is, either directly or indirectly, affixed to the lower layer of the core or base panel. The core and / or the base panel may comprise coupling profiles at the panel edges for coupling, e.g., locking, of decorative panels for the covering of a floor, a wall, a ceiling or furniture. Preferably, the decorative panel comprises a first pair of opposite edges,which preferably form the long edges of the panel, as well as comprises a second pair of opposite edges, which preferably form the short edges of the decorative panel; wherein both pairs of opposite edges comprise coupling parts, which allow that a plurality of such decorative panels mutually can be coupled to each other. Preferably, these coupling parts, on both pairs of edges, form a first locking system, which, in a coupled condition of two of such decorative panels, effects a locking in the plane of the panels and perpendicular to the respective edges, as well as form a second locking system, which, in a coupled condition of two of such panels, effects a locking transverse to the plane of the decorative panels. These coupling parts on the first and / or second pair of opposite edges substantially are preferably realized in the material of the decorative panel itself. The coupling parts of at least the first pair of opposite edges may be configured such that two of such panels can be coupled to each other at these edges by means of a turning movement and to this aim these coupling parts on the first pair of opposite edges preferably comprises a tongue and a groove, as well as of locking parts, which, in the coupled condition, prevent the shifting apart of the tongue and groove / Preferably, on the first pair of opposite edges the groove is bordered by a lower lip and an upper lip, of which the lower lip extends laterally up to beyond the distal extremity of the upper lip. The coupling parts on the second pair of edges are preferably configured such that two of such floor panels can be coupled to each other at these edges by means of a downward movement of one decorative panel in respect to the other, more particularly such that a plurality of such decorative panels can be coupled to each other by means of the so-called "folddown" technique. Preferably, the first locking system of the second pair of edges is at least formed of an upward-directed lower hook-shaped part, which is situated on one of said two edges and which comprises a distal extremity, as well as a downward-directed upper hook-shaped part, which is situated on the opposite edge, wherein the lower hook-shaped part consists of or comprises a lip with an upward- directed locking element (, whereas the upper hook-shaped part consists of a lip with a downward-directed locking element. The second locking system of the second pair of edges is preferably at least formed by a locking part, which is situated on the distal extremity of the lower hook-shaped part, as well as a locking part, which cooperates therewith in the coupled condition of two of such floor panels, on the edge which comprises the upper hook-shaped part. The distal extremity of the upper hookshaped part and the side which, in coupled condition, is opposite thereto, may or may not be free from mutually vertically locking portions, or said distal extremity and said side have mutually vertically locking portions. The coupling parts on the first pair of edges are preferably configured such that during the turning into each other of twoof such decorative panels an elastic force has to be overcome during the turning movement, which force, during the turning into each other, first increases in order to subsequently decrease again. Preferably, the floor panels on the first pair of edges, in the final coupled condition of two of such decorative panels, are coupled in horizontal direction free from play, or, if such play indeed exists, this play is less than 0.25 mm. The aforementioned locking parts of the first locking system on the first pair of edges are preferably formed at least of a locking part on the lower side of the tongue and a locking part on the aforementioned lower lip, which, by means of contact surfaces, which can cooperate with each other, result in the respective locking. The coupling parts on the first pair of edges in the coupled condition may show one or both of the following characteristics: - in the direction of the distal extremity of the tongue, at the same height (H) as the aforementioned contact surfaces, a free space is present between the tongue and the opposite floor panel; - the cooperation of the aforementioned contact surfaces defines a tangent line which, in respect to the horizontal, forms an angle of at least 40 degrees.

[0112] In case the digital layout is digitally printed onto the base panel, for example the core, during manufacturing and in case the base panel is divided into a plurality of decorative panels, such that the front side of each decorative panel includes one of the printed decorative panel designs (meaning together they form the output design), it is preferred that a positioning code is present on the panel. In such an embodiment, the backside, for example the backing layer, of each decorative panel preferably includes the positioning code of the decorative panel design of the decorative layer printed on the front side of said decorative panel, preferably the core.

[0113] In embodiments, step 2.b) comprises obtaining the style from a style database. The style may include a particular texture that is desired in decorative panels, such as a wood texture, a tile texture, concrete texture, etc. For example, a first digital input image has a first wood texture, and a second digital input image has a second wood texture. This can allow the at least one generative model to generate designs based on the patterns of the digital input images. Other examples of the at least one style, such as relief-related style, which may be included in the style database, have been described earlier in this document. It is imaginable that at least a part of the output relief (structure) design is entirely unrelated to the output design or input image, which may, for example, be the case for defining the colour and / or the size and / or the location and / or the number of grouts or bevels to be applied to the decorative panels.

[0114] In embodiments, the style, preferably a relief related style, and / or the digital input image and / or the input depth map are obtained from or selected by a user, such asan end-user (e.g. a consumer) inclined to (contribute to the) design of a set of decorative panels to be purchased. For example, the user selects the style and / or the input depth map on a particular website online and / or on a computer at a merchant or manufacturer and / or selects at least one user-owned digital input image. In this latter case, the user-owner input image will normally have to be uploaded to a server, typically a web server of the merchant or manufacturer, for further processing according to the invention.

[0115] In embodiments, the device comprises means for carrying out the method according to the present invention. The device may comprise a central processing unit (CPU) configured to carry out the method according to the present invention by executing instructions; a memory module for storing program instructions and data; one or more input / output interfaces for communication with external devices (e.g., user device and / or sensor module); a display screen for rendering visual output to the user; and a storage medium for long-term data storage.

[0116] ln embodiments, a non-transient computer readable medium containing a computer executable software which when executed on a computer system performs the method as defined herein before by the embodiments of the present invention. A non-transient computer readable medium may include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a randomaccess memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), digital video disks (DVDs), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a non-transient computer readable medium may be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, device or module.

[0117] Preferred embodiments of the invention are presented in the non-limitative set of clauses presented below:Clause 1. A method for generating instructions for embossing at least one decorative panel, such as a decorative floor panel or wall panel, in particular for composing a decorative (floor or wall) covering composed of a plurality of such panels, the method comprising: a) obtaining at least one digital image comprising at least one texture, wherein the digital image is preferably related to a decorative panel, b) determining, preferably by a computer, at least one depth map based on at least one digital image, andc) generating instructions, preferably digital instructions, based on the depth map for embossing at least a part of at least one decorative panel, in particular for applying an embossing structure in a decorative surface layer of said decorative panel. Clause 2. The method according to clause 1, wherein step 1.a) comprises scanning or capturing at least one image of at least one decorative panel, preferably a decorative panel to be embossed.Clause 3. The method according to clause 1 or clause 2, wherein the digital image comprises a plurality of images, preferably of one or more decorative panels.Clause 4. The method according to clause 3, wherein step 1.b) comprises applying a stereo matching algorithm to determine the depth map.Clause 5. The method according to clause 4, wherein the stereo matching algorithm comprises a block matching algorithm.Clause 6. The method according to clause 4 or clause 5, wherein the stereo matching algorithm comprises a semi-global matching, SGM, algorithm.Clause 7. The method according to any one of clauses 4-6, wherein the stereo matching algorithm comprises a graph cut algorithm.Clause 8. The method according to any one of clauses 1-7, wherein the stereo matching algorithm comprises a convolutional neural network, CNN, and / or wherein step 1.b) comprises applying a CNN to the digital image to determine the depth map. Clause 9. The method according to any one of clauses 1-8, wherein step 1.a) comprises:- obtaining at least one digital input image,- obtaining at least one style, in particular at least one embossing style, and- applying, by using at least one first generative model being associated with a corresponding style transfer, said at least one style to said at least one digital input image to generate at least one design as the digital image for the decorative panel. Clause 10. The method according to clause 9, wherein step 1.b) comprises applying, using the first generative model, said style to said digital input image to generate the depth map for the decorative panel.Clause 11. The method according to clause 9, wherein step 1.b) comprises applying, using at least one second generative model being associated with a corresponding style transfer, said style to said digital input image to generate the depth map for the decorative panel.Clause 12. The method according to any one of clauses 9-11 , wherein at least one style comprises at least one relief-related style and / or at least one embossing related style and / or at least one bevel related style and / or at least one grout line related style.Clause 13. The method according to any one of clauses 9-12, wherein the digital input image comprises at least an input depth map.Clause 14. The method according to any one of clauses 9-13, wherein the first generative model and / or the second generative model are / is trained on a training dataset comprising scanned and / or captured images of decorative panels.Clause 15. The method according to clause 14, wherein the training dataset comprises historically generated digital images, preferably historically generated images, of decorative panels.Clause 16. The method according to clause 14 or clause 15, wherein the training dataset comprises depth maps historically determined based on the respective images of decorative panels.Clause 17. The method according to clause 15 or clause 16, wherein the training dataset comprises historically generated depth maps relating to the respective historically generated digital images.Clause 18. The method according to any one of clauses 9-17, wherein each of the first generative model and / or second generative model is an autoencoder, AE.Clause 19. The method according to any one of clauses 9-18, wherein each of the first generative model and / or second generative model is an autoregressive model.Clause 20. The method according to any one of clauses 9-19, wherein each of the first generative model and / or second generative model is a score-based generative model, SGM.Clause 21. The method according to any one of clauses 9-20, wherein the first generative model and / or second generative model is a first and / or second generative adversarial network, GAN, respectively, wherein the first GAN comprises at least one first generator and at least one first discriminator and the second GAN comprises at least one second generator and at least one second discriminator.Clause 22. The method according to clause 21 , wherein each of the at least one first generator and / or the at least one second generator comprise(s) two or more generators.Clause 23. The method according to clause 21 or clause 22, wherein each of the at least one first discriminator and / or the at least one second discriminator comprise(s) two or more discriminator.Clause 24. The method according to any one of clauses 21-23, wherein each of the at least one first generator and / or the at least one second generator comprises at least one deep CNN, DCNN, generator.Clause 25. The method according to any one of clauses 21-24, wherein each of the at least one first generator and / or the at least one second generator comprises at least one Wasserstein-based generator.Clause 26. The method according to any one of clauses 21-25, wherein each of the at least one first generator and / or the at least one second generator comprises at least one autoencoder-based generator or a II Net-based generator.Clause 27. The method according to any one of clauses 21-26, wherein each of the at least one first generator and / or the at least one second generator comprises at least one visual geometry group, VGG, -based generator.Clause 28. The method according to any one of clauses 21-27, wherein each of the at least one first generator and / or the at least one second generator comprises at least one residual network, ResNet, -based generator.Clause 29. The method according to any one of clauses 21-28. wherein each of the at least one first generator and / or the at least one second generator comprises at least one style-based generator.Clause 30. The method according to any one of clauses 21-29, wherein each of the at least one first generator and / or the at least one second generator comprises at least one progressive generator.Clause 31. The method according to any one of clauses 21-30, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one patch-wise-based discriminator.Clause 32. The method according to any one of clauses 21-31 , wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one DCNN discriminator.Clause 33. The method according to any one of clauses 21-32, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one distance-based discriminator.Clause 34. The method according to clause 33, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one of: at least one earth mover’s distance discriminator and at least one total variation distance discriminator.Clause 35. The method according to any one of clauses 21-34, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one of the following loss functions: least squares, hinge loss, and Jensen- Shannon divergence.Clause 36. The method according to any one of clauses 21-35, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one multi-scale discriminator.Clause 37. The method according to any one of clauses 21-36, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one feature-matching discriminator.Clause 38. The method according to any one of clauses 21-37, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one ranking discriminator.Clause 39. The method according to any one of clauses 21-38, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one Gaussian mixture model, GMM, discriminator.Clause 40. The method according to any one of clauses 12-39, wherein step 1.b) comprises detecting that a similarity between the generated depth map and the relief- related style is within a predetermined range.Clause 41. The method according to any one of clauses 12-40, wherein step 1.b) comprises applying, using the first generative model or the second generative model, said style to said digital input image to generate a plurality of depth maps for a plurality of decorative panels.Clause 42. The method according to clause 41, wherein step 1.b) comprises detecting that a similarity between the generated depth maps is within a predetermined range.Clause 43. The method according to any one of clauses 40-42, wherein the similarity is detected based on at least one of: a structural similarity index, mean squared error, peak signal-to-noise ratio, cosine similarity, Euclidean distance, correlation coefficient, Kullback-Leibler divergence, and earth mover’s distance. Clause 44. The method according to any one of clauses 1-43, wherein step 1.a) comprises obtaining at least one digital image from an image database and / or a camera and / or a scanner.Clause 45. The method according to any one of clauses 9-43, wherein at least one digital input image is obtained from the image database and / or a camera and / or a scanner.Clause 46. The method according to clause 44 or clause 45, wherein the image database comprises a plurality of scanned wood-pattern images and / or scanned tilepattern images and / or scanned concrete-pattern images.Clause 47. The method according to any one of clauses 44-46, wherein the image database comprises a plurality of images and their respective depth maps.Clause 48. The method according to any one of clauses 1-47, further comprising providing at least a part of an embossing structure onto at least a part of at least one decorative panel based on the embossing instructions generated in step 1.c).Clause 49. The method according to clause 48, wherein the embossing structure is an at least partially digitally printed embossing structure, preferably realized by:(i) applying a liquid base layer directly or indirectly onto a decorative layer of the panel(ii) position-selectively digitally printing ink droplets comprising at least one UV inhibitor position-selectively at locations of the liquid base layer which should become recessed portions of the embossing, based upon the instructions generated during step 1.c),(iii) subjecting the base layer and the ink droplets to a UV curing treatment, and(iv) mechanically removing uncured portions of the base layer and / or ink droplets to form the embossing at least partially.Clause 50. The method according to clause 48 or clause 49, wherein the embossing structure is provided in two or more distinct steps.Clause 51. A device comprising means for carrying out the method of any one of clauses 1-47.Clause 52. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of clauses 1-47.Clause 53. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of clauses 1-47.Clause 54. A system comprising:- the device according to clause 51 , and- embossing means for providing an embossing structure onto the physical decorative panel based on the generated embossing instructions.Clause 55. A decorative panel obtained by carrying out the method according to any of clauses 1-50.Clause 56. A set of decorative panels obtained by carrying out the method according to any of clauses 1-50.Clause 57. Set of decorative panels according to clause 57, wherein each panel comprises at at least one pair of opposite edges coupling profiles allowing interlocking of adjacent panels.Clause 58. A decorative covering, in particular a floor covering, wall covering, ceiling covering, or furniture covering, composed a set of, preferably interconnected decorative panels according to clause 56 or 57.Examples

[0118] Example embodiments of the invention will be described with reference to Fig. 1 and 2, which are not intended to limit the scope of the invention in any way.Example 1 : example of a method according to the invention

[0119] Fig. 1A illustrates example embodiments of a method according to the present invention. It relates to a method for generating for generating instructions for embossing a decorative panel, the method comprising:1.a) obtaining (20) a digital image comprising a texture, the digital image relating to the decorative panel,1.b) determining (22) a depth map based on the digital image, and1.c) generating instructions based on the depth map for embossing the decorative panel.1.d) providing (30) an embossing structure onto the decorative panel based on the embossing instructions generated in step 1.c), wherein the step 1.a) comprises:2. a) obtaining (1) a digital input image,2.b) obtaining (2) a style,2.c) applying (10), by using a first generative model being associated with a corresponding style transfer, said style to said digital input image to generate an output design as the digital image for the decorative panel, and2.d) printing (25) at least a part of the output design, directly or indirectly, onto the base panel, wherein the step 1.b) comprises applying (10), using the first generative model, said style to said digital input image to generate the depth map for the decorative panel.

[0120] The step(s) of applying (10) the first generative model may relate to any of the aforementioned generative models, which is similar to the generative models described in Example 4.

[0121] In the step 1.d) of providing (30) an embossing structure, the embossing structure is provided on the printed decorative panel resulting from step 2.d) based on the generated instructions resulting from step 2.c) which are based on the determined, particularly generated, depth map resulting from step 1.b).Example 2: example of a method according to the invention

[0122] Fig. 1B illustrates example embodiments of a method according to the present invention. It relates to a method for generating for generating instructions for embossing a decorative panel, the method comprising:1.a) obtaining (T) a digital image comprising a texture, the digital image relating to the decorative panel,1.b) determining (22) a depth map based on the digital image, and1.c) generating instructions based on the depth map for embossing the decorative panel.1.d) providing (30) an embossing structure onto the decorative panel based on the embossing instructions generated in step 1.c).

[0123] Step 1.a) of obtaining (T) a digital image may relate to obtaining a digital image from a database which may comprise at least one of: historically generated designs, historically generated output designs, scanned images of (decorative) panels, captured images of (decorative) panels, etc.

[0124] The method comprises applying (10) a CNN to the digital image for determining (22) the depth map.

[0125] In the step 1.d) of providing (30) an embossing structure, the embossing structure is provided on the printed decorative panel resulting from step 2.d) based on the generated instructions resulting from step 2.c) which are based on the determined, particularly generated, depth map resulting from step 1.b).Example 3: example of a decorative panel according to the invention

[0126] Fig. 2A shows a schematic representation of a side view of an example of a decorative panel (220) according to the present invention. The panel (220) comprises a core (200) provided with an upper side and a lower side, and a decorative top structure (201) affixed, directly or indirectly, on said upper side of the core (200). The decorative top structure (201) comprises a decorative print layer forming at least one decor image. The panel (220) also comprises a substantially transparent or translucent three-dimensional embossing structure (202) covering said print layer (201). In the shown embodiment the embossing structure (202) comprises a continuous printed base layer (204) provided with one or more internal grouts (203) with a concave shape, and an elevated pattern layer formed by a plurality of (discontinuous) elevations (205) printed on top of said continuous base layer (204). Two side edges of the panel (220), as also shown in Fig. 2B, are provided with an external grout (210), also referred to as peripheral grout line (210), wherein a lower rectangular part of the grout (210) is formed by the base layer (204), and wherein an upper part of the grout (210) is provided with a bevel (211) formed by one of the printed elevations (205). The elevations form part of a lacquer layer (205). A carrier layer (206), and in particular a primer layer (206) is enclosed between the top structure (201) and the embossing structure (202). In the shown embodiment, the primer layer (206) comprises a pattern of mat primer (206A) and glossy primer(206B). The indentations (203) are present where the primer layer (206) is provided with mat primer (206A). The structured elevations (205) cover the glossy primer (206B) of the primer layer (206). Due to the embossing structure (202) being substantially transparent, the differences within the primer layer (206) are visible. It is also conceivable that the primer layer (206) is attached onto the upper side of the core (200), and that the decorative top structure (201) is attached onto the primer layer (206).

[0127] Fig. 2B shows a top view of the panel (220) shown in figure 2a. It can be seen that due to a part of the base layer being provided with said plurality of indentations (203) and part of the base layer being free of indentations a visually observable pattern is obtained. This effect is further reinforced by the primer layer (206) comprising both mat and glossy primer (206A, 206B) in a pattern which is in line with the embossing structure (202).Example 4: example of a method according to the invention

[0128] Fig. 3 illustrates example embodiments of a method according to the present invention. It relates to a method for generating an output design for a decorative panel, the method comprising:2. a) obtaining (1) a digital input image comprising a texture,2.b) obtaining (2) a style, and2.c) applying (10), by using a generative model associated with a corresponding style transfer, said style to said digital input image to generate (20) at least one output design and / or at least one respective output depth map, wherein the generative model comprises a GAN comprising a generator and a discriminator.

[0129] In this example, the step of applying (10) comprises applying (11) the generator to generate (13) designs and / or depth maps and applying (12) the discriminator to detect or classify real-looking designs and / or depth maps. The detection or classification may be performed by determining the similarity between the generated designs and a reference image, e.g., the input image and / or any image from the database. The detection or classification may also be performed by determining the similarity between the generated output maps and a reference image, e.g., the depth map comprised in the input image and / or any image from the database and / or the relief-related style comprised in the style.

[0130] The step of applying (12) the discriminator comprises accepting (15) designs and / or depth maps which have been detected or classified as being real-looking designs and / or depth maps. The step of applying (12) the discriminator may further comprise rejecting (14) designs and / or depth maps which have been detected orclassified as not being real-looking designs and / or depth maps, and preferably discarding said rejected designs and / or depth maps.

[0131] In this example, the step of applying (10) further comprises storing (24) the accepted designs and / or depth maps in a training dataset. The training dataset may comprise scanned and / or captured images of decorative panels and / or historically generated output designs and / or depth maps of decorative panels. Thus, the training dataset may be used by the discriminator to detect or classify real-looking designs and / or depth maps by determining the similarity between the generated designs and / or depth maps (i.e. , in step 13) with any one of: the images and / or depth maps, and historically generated designs and / or depth maps in the training dataset.Example 5: example of a decorative panel according to the invention

[0132] Fig. 4 schematically shows a perspective view of a decorative panel (40) according to the invention. A decorative panel (40) comprises a core (41) which comprises an upper side and a lower side. A decorative layer (43) is, either directly or indirectly, affixed to the upper side of the core (41). The decorative layer (43) of every decorative panel (40) according to the invention and / or obtained by applying the method according to the invention is unique by the generated output design of the at least one generative model. On top of the decorative layer (43), at least one protective layer (44), such as a wear layer, is affixed which covers the decorative layer (43). In addition to the decorative layer (43), the protective layer (44) can comprise a relief structure (embossing structure) which is unique and defined by the generated output design of the at least one generative model.

[0133] As shown in Fig. 4, the decorative panel (40) is provided with a UV-coating (45) that is attached to on top of the protective layer (44). Further, a backing layer (46) is, either directly or indirectly, affixed to the lower layer of the core (41). The core (41) further comprises coupling profiles (42) at at least one pair, and preferably each pair, of opposite panel edges. The coupling profiles (42) enable the mutual locking of decorative panels (40) for forming a floor covering, wall covering, ceiling covering or furniture covering. Preferably, the coupling profiles are configured to interlock adjacent decorative panels both in a direction parallel to a plane defined by the panels and / or in a direction perpendicular to said plane defined by the panels. In case the digital layout is digitally printed onto the base panel, for example the core (41), during manufacturing and in case the base panel is divided into a plurality of decorative panels, such that the front side of each decorative panel includes one of the plurality of printed decorative panel designs (meaning together they form the output design), it is or may be preferred that a positioning code is present on the panel for installation purposes. In such an embodiment, the backside, for examplethe backing layer (46), of each decorative panel (40) preferably includes the positioning code (47) of the decorative panel design of the decorative layer (43) printed on the front side of said decorative panel (40), preferably the core (41). Prior to printing, saw losses and / or cutting waste, for example due to later division of the panels and / or profiling of panel edges, may be taken into account, to prevent loss of the output design due to later sawing, profiling, cutting actions. Pre-known (predefined) cutting sections or other waste sections may e.g. be kept free from the output design and may remain unprinted during the printing step.

[0134] (End of Example 5)

[0135] Example 6: Fig. 5 shows an example of a method for producing a decorative panel (50) and / or a decorative slab to be cut into a plurality of decorative panels (50) according to the invention. The production process shown in figure 5 makes use of the method for generating instructions for an embossing according to the invention. In this exemplary embodiment, a user (51) selects at least one digital input image (52) and at least one style, such as a decor related style and / or a relief structure related style, or style transfer (53), such as a decor related style transfer and / or a relief structure related style transfer, to be applied to said digital input image (52). This user selection is made by using a digital user interface (554), which may be a website of a merchant or manufacturer running on a (web)server. The input image (52) may be selected from a prestored collection of input images (52) and / or may be an input image (52) uploaded by said user to said (web)server. The input (52, 53) given by the user is entered into at least one generative model (55), in particular at least one pretrained generative model (55) running on a computer. Said computer may be the same computer as said (web)server and / or may be able to communicate with said (web)server. The generative model (554) is preferably programmed to, based upon said selected input image(s) and style(s), generate at least one output design for a decorative panel (50) to be produced. Additionally or alternatively, the generative model (54) is preferably programmed to, based upon said selected input image(s) and style(s), generate a depth map based upon the input image and / or chosen style(s). This depth map may serve to generate instructions, in particular computer-readable instructions, and preferably generated by a computer, for embossing the at least a part of at least one decorative panel, in particular for applying an embossing structure in a decorative surface layer of said decorative panel (50) to be produced. The instructions may be based on the input (52, 53) given by the user. To this end, the digital image (52) is preferably analyzed and a depth map based on said image (52) is acquired. The output design and / or depth map is commonly created and stored as one or more output design files. The output designcomprises an output decor design and / or an output relief structure design. In case the output design comprises both an output decor design and an output relief structure design, then the output relief structure is preferably at least partially aligned with said output decor design. This allows the realization of improved look-and-feel characteristics of the decorative panel (50) produced. The output decor design differs from the input image. The output decor design predominantly contributes to the visible characteristics of the decorative panel. The output relief structure design predominantly contributes to the tactile characteristics of the decorative panel. The output decor design is digitally sent to a decor printing station (III), wherein the output relief structure design is digitally sent to an embossing station (VIII). After making the selection(s) by the user (51), the printing line may be activated in order to produce to customized decorative panel(s) (50). To this end, a base panel (60) is provided, which may for example be an HDF or MDF board, a polymer comprising board, a mineral board, or a combinations thereof. Preferably, at least one layer of the base panel, such as a core layer, may comprise at least one polymer material, in particular at least one polymer matrix, preferably at least one thermoplastic, which polymer material is preferably at least partially composed of a polymer that is selected from the group consisting of: polypropylene (PP), polyurethane (Pll), thermoplastic polyurethane (TPU), polystyrene (PS), polyethylene (PE), polyvinyl chloride (PVC), at least one polyester, or mixtures thereof. In case at least one polyester is used in the core layer(s), this polyester is preferably a homopolyester, such as polyethylene terephthalate (PET), or polyethylene furanoate (PEF), and / or a copolyester, such as polyethylene furanoate terephthalate (PEFT) or polyethylene terephthalate glycol- modified PETG). PET is a homopolyester made from terephthalic acid and ethylene glycol. PEF is also a homopolyester, derived from furandicarboxylic acid and ethylene glycol. It is considered a more sustainable alternative to PET due to its biobased origin. PEFT is a copolyester made by copolymerizing furandicarboxylic acid, terephthalic acid, and ethylene glycol. PETG is a copolyester. It is modified by adding glycol, such as cyclohexanedimethanol (CHDM), which provides greater clarity and durability compared to standard PET. The use of one or more polyesters in the panel(s) according to the invention may be advantageous and preferred over PVC and various other thermoplastic polymers due to the relatively high temperature resistance of polyesters. Contrary to PVC and various other thermoplastic polymers, polyesters are suitable to withstand temperatures of above 200 degrees Celsius. This polymer based layer of the base panel (or even the base panel as such) may be an extruded (or co-extruded) layer. One or more further additives, such as wood fibers, glass fibers, etcetera and / or one or more additional layers, such as a reinforcementlayer (e.g. a glass fiber layer) may be incorporated in the board as well. Preferably, the base panel is rigid or semi-flexible, which typically results in a rigid or semiflexible decorative panel. A top surface of the base panel (60) is successively provided with a primer layer (I), a first base coat layer (II) and optionally a second base coat layer (II). Typically these basecoat layers are white layers to improve the colour authenticity of the decor to be printed in a subsequent step. Based upon the generated output decor design, a decorative layer (decor) (III) is digitally printed onto said base coat layer(s) (II). The printer used for this printing step is schematically shown and is in this illustrative embodiment preferably a single pass printer, which comprises four print heads fixed over the whole working width, wherein each print head is configured to print its own colour (Cyan, Magenta, Yellow, Black (CMYK)). The inks used are preferably water-based inks as these inks are environmental friendly, odorless, and have the capacity to dry quickly. After applying the decorative layer (III), a transparent primer layer (IV) is applied, followed successively by a wear layer (V) to protect the decorative layer, and a matt layer (VI) to reduce the gloss level to an acceptable level. Additionally, a relief structure is created on top of said matt layer (VI), preferably based upon said output relief structure design and / or based on the instructions for embossing. The formation of this relief structure concerns a multistep process. Firstly, a structure layer (VII), also referred to as base layer, is applied in liquid form, preferably by means of roller coating. Subsequently, ink droplets (VIII) are digitally printed onto the still liquid structure layer (VII), wherein said ink droplets comprise at least one UV inhibitor, and wherein said ink droplets (VIII) are printed position-selectively at locations which should become recessed portions of the relief structure. During a subsequent UV curing step the structure layer (VII) is cured (hardened) except for the portions which are covered by the UV inhibiting ink. In a subsequent step the still liquid portions of the structure layer (VII) are mechanically removed, preferably by means of one or more rotating brushes, in particular metal brushes. After formation of the relief structure (VII+VIII), a protective top coating (IX) is applied, typically initially in liquid form following by a UV curing step. In a last step (X) (or series of steps) the panel (50) or slab may be cut into a plurality of smaller decorative panels and / or may be subjected to an edge-profiling step to realize coupling profiles at at least one pair of opposite decorative panel edges to allow interlocking of decorative panels when installing a decorative covering consisting of those panels. The decorative panels may be used e.g. as floor panel, wall panel, ceiling panel, or furniture panel. For decorative floor panels, common lengths range from 1200-2400 mm for decorative planks (oblong, often rectangular, panels) and 600-800 mm for decorative tiles (square panels), with widths of 150-300 mm for planks and 300-600 mm for tiles, and thicknesses of 4-12 mm, depending on the material and thickness of the base panel. For decorative wall panels, lengths range from 1200-3000 mm for vertically oriented panels and 600- 2400 mm for horizontally oriented panels, with widths of 200-600 mm for planks (oblong, often rectangular panels) and 600-1200 mm for broader panels, and thicknesses of 4-10 mm for standard applications or up to 15 mm for acoustic or structural needs. The thickest layer of the decorative panels is preferably the base layer, which typically also provides rigidity and stiffness to the panels.

[0136]

[0137] The above-described inventive concepts are illustrated by several illustrative embodiments. It is conceivable that individual inventive concepts may be applied without, in so doing, also applying other details of the described example. It is not necessary to elaborate on examples of all conceivable combinations of the abovedescribed inventive concepts, as a person skilled in the art will understand numerous inventive concepts can be (re)combined in order to arrive at a specific application.

[0138] Although the present invention has been described above with reference to certain embodiments thereof, it will be evident that various modifications and changes may be made to these embodiments without departing from the broader scope of the present invention, as defined by the appended claims.

Claims

Claims1. A method for generating instructions for embossing at least one decorative panel, such as a floor panel or wall panel, in particular for composing a decorative covering, such as a decorative floor covering or decorative wall covering composed of a plurality of said decorative panels, the method comprising: a) obtaining at least one digital image comprising at least one texture, wherein the digital image is preferably related to a decorative panel, step 1.a) comprising:- obtaining at least one digital input image,- obtaining at least one style, in particular at least one embossing style, and- applying, by using at least one first generative model being associated with a corresponding style transfer, said at least one style to said at least one digital input image to generate at least one design as the digital image for the decorative panel; b) determining at least one depth map based on at least one digital image, and step 1.b) comprising:- applying, using either the first generative model or at least one second generative model being associated with a corresponding style transfer, said style to said digital input image to generate the depth map for the decorative panel; c) generating instructions, preferably digital instructions, based on the depth map for embossing at least a part of at least one decorative panel, in particular for applying an embossing structure in a decorative surface layer of said decorative panel.

2. The method according to claim 1 , wherein step 1.a) comprises scanning or capturing at least one image of at least one decorative panel, preferably a decorative panel to be embossed.

3. The method according to claim 1 or claim 2, wherein the digital image comprises a plurality of images, preferably of one or more decorative panels.

4. The method according to claim 3, wherein step 1.b) comprises applying a stereo matching algorithm to determine the depth map.

5. The method according to claim 4, wherein the stereo matching algorithm comprises a block matching algorithm.

6. The method according to claim 4 or claim 5, wherein the stereo matching algorithm comprises a semi-global matching, SGM, algorithm.

7. The method according to any one of claims 4-6, wherein the stereo matching algorithm comprises a graph cut algorithm.

8. The method according to any one of claims 1-7, wherein the stereo matching algorithm comprises a convolutional neural network, CNN, and / or wherein step 1.b) comprises applying a CNN to the digital image to determine the depth map.

9. The method according to any one of the preceding claims, wherein at least one style comprises at least one relief-related style and / or at least one embossing related style and / or at least one bevel related style and / or at least one grout line related style.

10. The method according to any one of the preceding claims, wherein the digital input image comprises at least an input depth map.

11. The method according to any one of any of the preceding claims, wherein the first generative model and / or the second generative model are / is trained on a training dataset comprising scanned and / or captured images of decorative panels.

12. The method according to claim 11 , wherein the training dataset comprises historically generated digital images, preferably historically generated images, of decorative panels.

13. The method according to claim 11 or claim 12, wherein the training dataset comprises depth maps historically determined based on the respective images of decorative panels.

14. The method according to claim 12 or claim 13, wherein the training dataset comprises historically generated depth maps relating to the respective historically generated digital images.

15. The method according to any one of preceding claims, wherein each of the first generative model and / or second generative model is an autoencoder, AE.

16. The method according to any one of preceding claims, wherein each of the first generative model and / or second generative model is an autoregressive model.

17. The method according to any one of preceding claims, wherein each of the first generative model and / or second generative model is a score-based generative model, SGM.

18. The method according to any one of the preceding claims, wherein the first generative model and / or second generative model is a first and / or second generative adversarial network, GAN, respectively, wherein the first GAN comprises at least one first generator and at least one first discriminator and the second GAN comprises at least one second generator and at least one second discriminator.

19. The method according to claim 18, wherein each of the at least one first generator and / or the at least one second generator comprise(s) two or more generators.

20. The method according to claim 18 or claim 19, wherein each of the at least one first discriminator and / or the at least one second discriminator comprise(s) two or more discriminator.

21. The method according to any one of claims 18-20, wherein each of the at least one first generator and / or the at least one second generator comprises at least one deep CNN, DCNN, generator.

22. The method according to any one of claims 18-21 , wherein each of the at least one first generator and / or the at least one second generator comprises at least one Wasserstein-based generator.

23. The method according to any one of claims 18-22, wherein each of the at least one first generator and / or the at least one second generator comprises at least one autoencoder-based generator or a II Net-based generator.

24. The method according to any one of claims 18-23, wherein each of the at least one first generator and / or the at least one second generator comprises at least one visual geometry group, VGG, -based generator.

25. The method according to any one of claims 18-24, wherein each of the at least one first generator and / or the at least one second generator comprises at least one residual network, ResNet, -based generator.

26. The method according to any one of claims 18-25, wherein each of the at least one first generator and / or the at least one second generator comprises at least one style-based generator.

27. The method according to any one of claims 18-26, wherein each of the at least one first generator and / or the at least one second generator comprises at least one progressive generator.

28. The method according to any one of claims 18-27, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one patch-wise-based discriminator.

29. The method according to any one of claims 18-28, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one DCNN discriminator.

30. The method according to any one of claims 18-29, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one distance-based discriminator.

31. The method according to claim 30, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one of: at least one earth mover’s distance discriminator and at least one total variation distance discriminator.

32. The method according to any one of claims 18-31 , wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at leastone of the following loss functions: least squares, hinge loss, and Jensen-Shannon divergence.

33. The method according to any one of claims 18-32, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one multi-scale discriminator.

34. The method according to any one of claims 18-33, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one feature-matching discriminator.

35. The method according to any one of claims 18-34, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one ranking discriminator.

36. The method according to any one of claims 18-35, wherein each of the at least one first discriminator and / or the at least one second discriminator comprises at least one Gaussian mixture model, GMM, discriminator.

37. The method according to any one of the preceding claims, wherein step 1.b) comprises detecting that a similarity between the generated depth map and the relief-related style is within a predetermined range.

38. The method according to any one of the preceding claims, wherein step 1.b) comprises applying, using the first generative model or the second generative model, said style to said digital input image to generate a plurality of depth maps for a plurality of decorative panels.

39. The method according to claim 38, wherein step 1.b) comprises detecting that a similarity between the generated depth maps is within a predetermined range.

40. The method according to any one of claims 37-39, wherein the similarity is detected based on at least one of: a structural similarity index, mean squared error, peak signal-to-noise ratio, cosine similarity, Euclidean distance, correlation coefficient, Kullback-Leibler divergence, and earth mover’s distance.

41. The method according to any one of claims 1-40, wherein step 1.a) comprises obtaining at least one digital image from an image database and / or a camera and / or a scanner.

42. The method according to any one of the preceding claims, wherein at least one digital input image is obtained from the image database and / or a camera and / or a scanner.

43. The method according to claim 41 or claim 42, wherein the image database comprises a plurality of scanned wood-pattern images and / or scanned tile-pattern images and / or scanned concrete-pattern images.

44. The method according to any one of claims 41-43, wherein the image database comprises a plurality of images and their respective depth maps.

45. The method according to any one of claims 1-44, further comprising providing at least a part of an embossing structure onto at least a part of at least one decorative panel based on the embossing instructions generated in step 1.c).

46. The method according to claim 45, wherein the embossing structure is an at least partially digitally printed embossing structure, preferably realized by:(i) applying a liquid base layer directly or indirectly onto a decorative layer of the panel,(ii) position-selectively digitally printing ink droplets comprising at least one UV inhibitor position-selectively at locations of the liquid base layer which should become recessed portions of the embossing, based upon the instructions generated during step 1.c),(iii) subjecting the base layer and the ink droplets to a UV curing treatment, and(iv) mechanically removing uncured portions of the base layer and / or ink droplets to form the embossing at least partially.

47. The method according to claim 45 or claim 46, wherein the embossing structure is provided in two or more distinct steps.

48. A device comprising means for carrying out the method of any one of claims 1-44.

49. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of claims 1-44.

50. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of claims 1-47.

51. A system comprising:- the device according to claim 48, and- embossing means for providing an embossing structure onto the physical decorative panel based on the generated embossing instructions.

52. A decorative pane obtained by carrying out the method according to any of claims 1-47, wherein said decorative panel is preferably a rigid decorative panel.

53. A set of decorative panels obtained by carrying out the method according to any of claims 1-47.

54. Set of decorative panels according to claim 53, wherein each panel comprises at at least one pair of opposite edges coupling profiles allowing interlocking of adjacent panels.

55. A decorative covering, in particular a floor covering, wall covering, ceiling covering, or furniture covering, composed a set of, preferably interconnected decorative panels according to claim 53 or 54.

Images