Device and method for a context-based background generation process, vehicles, server or cloud, and method for training a generative machine learning model

Abstract

The invention relates to a device for a context-based background generation process. The device includes a processing circuit configured to receive input data indicative of the context of the vehicle and / or at least a portion of the vehicle occupants. Furthermore, the processing circuit is designed to generate a respective context-based background for at least one screen of the vehicle by means of a trained generative machine learning model which receives the input data as input.

Description

[0001] 24-2690 1

[0002] DEVICE AND METHOD FOR CONTEXT-DEPENDENT

[0003] BACKGROUND DESIGN, VEHICLES, SERVER OR CEOUD AND METHODS FOR TRAINING A GENERAL MACHINE LEARNING MODEEES

[0004] Description

[0005] The present invention relates to the contextualized generation of backgrounds for screens in vehicles. In particular, embodiments of the present invention relate to a device and a method for context-dependent background generation, vehicles, a server or cloud with the device, a method for training a generative machine learning model, a non-volatile machine-readable medium, and a program.

[0006] In vehicles, background images are used to display information on the screens. For example, different background images can be displayed for different driving modes. These background images are manually created by design agencies and are the same for all customers in every situation.

[0007] There is therefore a need to enable the adaptation of the display of visual content in the vehicle with reduced manual effort. This need is addressed by the device and method for context-dependent background generation, the vehicles, the server or cloud with the device, the method for training a generative machine learning model, the non-volatile machine-readable medium, and the program according to the independent claims. Further aspects and enhancements are described in the dependent claims, the following description, and the figures.

[0008] According to a first aspect, the present invention relates to a device for context-dependent background generation. The device comprises a processing circuit configured to receive input data indicating the context of the vehicle and / or at least some of its occupants. Furthermore, the processing circuit is configured to generate a respective context-dependent background for at least one screen of the vehicle by means of a trained, generative machine learning model that receives the input data as input. 24-2690 2

[0009] According to a second aspect, the present invention relates to a vehicle. The vehicle comprises at least one screen and a device for context-dependent background generation according to the first aspect. The processing circuit of the device is further configured to control the at least one screen and output the respective context-dependent background.

[0010] According to a third aspect, the present invention relates to another vehicle. The vehicle comprises at least one screen and an interface. The interface is configured to send first data, displaying a context of the vehicle and / or at least some of its occupants, to a server or cloud, and to receive second data, displaying a context-dependent background for the at least one screen of the vehicle, from the server or cloud in response to the sending of the first data. The vehicle further comprises a control circuit configured to control the at least one screen and display the respective context-dependent background.

[0011] According to a fourth aspect, the present invention relates to a server or a cloud. The server or cloud comprises a background generation device according to the first aspect and an interface. The interface is configured to receive input data, at least partially, from the vehicle and to output data to the vehicle, which displays the respective context-dependent background for the at least one screen of the vehicle.

[0012] According to a fifth aspect, the present invention relates to a method for context-dependent background generation. The method comprises receiving input data that indicates a context of the vehicle and / or at least some of its occupants. Furthermore, the method comprises generating a respective context-dependent background for at least one screen of the vehicle using a trained, generative machine learning model that receives the input data as input.

[0013] According to a sixth aspect, the present invention relates to a method for training a generative machine learning model to generate a context-dependent background for at least one screen of a vehicle as output based on input data that indicates a context of the vehicle and / or at least some of its occupants. The method comprises fine-tuning a pre-trained generative machine learning model based on training data that specifies predetermined visual and stylistic guidelines in order to train the model to adhere to the predetermined visual and stylistic guidelines when generating the respective context-dependent background.Furthermore, the method includes the insertion of 24-2690 3 control layers into the pre-trained generative machine learning model, which enable the specification of positions of one or more objects in the respective foreground for at least one screen and control the pre-trained generative machine learning model to take the positions of the one or more objects into account when generating the respective context-dependent background.

[0014] According to a seventh aspect, the present invention relates to a non-volatile machine-readable medium on which a program is stored containing program code for carrying out one of the methods according to the fifth and sixth aspects when the program is executed on a processor or a programmable hardware component.

[0015] According to an eighth aspect, the present invention relates to a program with program code for carrying out one of the methods according to the fifth and sixth aspects when the program is executed on a processor or a programmable hardware component.

[0016] The present invention enables automatic adjustment of the screen background based on the current context of the vehicle and / or its occupants. This allows for dynamic and context-dependent adaptation of the visual display on the screen. User-friendliness is thus improved, as the vehicle occupants receive an appealing and relevant visual environment on the screens. Since the generative machine learning model is capable of autonomously generating a variety of visual content for the background, the need for time-consuming, manual design of the screen background is reduced. The costs for developing and maintaining screen backgrounds can be significantly reduced, as the dynamic generation of content by the generative machine learning model is automated.The generation of screen backgrounds using the generative machine learning model takes place within a few seconds, so that the development times for generating new screen backgrounds can also be significantly reduced.

[0017] Examples of implementation are explained in more detail below with reference to the accompanying figures. These show:

[0018] Fig. 1 shows a schematic representation of a vehicle and a server or cloud with a device for context-dependent background generation; 24-2690 4

[0019] Fig. 2 is a schematic representation of a vehicle with a device for context-dependent background generation;

[0020] Fig. 3 shows a schematic representation of a method for context-dependent background generation;

[0021] Fig. 4 shows a schematic representation of a method for training a generative machine learning model.

[0022] Several embodiments are now described in more detail with reference to the accompanying drawings, in which some of these embodiments are illustrated. For the sake of clarity, the thickness dimensions of lines, layers, and / or regions may be exaggerated in the figures.

[0023] Figure 1 illustrates a schematic representation of a vehicle 120 that is communicatively coupled to a server or a cloud (computing cloud) 140 (e.g., wirelessly via a cellular network and / or a local wireless network). A server is generally a computer or computing system that provides and manages resources, data, applications, or services for other computers, devices, or users in a network. A cloud is generally a distributed computing system (e.g., consisting of multiple servers) that provides and manages IT resources, such as computing power, storage, and software applications, as a service over a network, particularly the internet.

[0024] In Fig. 1, the vehicle 120 is depicted as a passenger car. However, the present invention is not limited to this. Rather, the vehicle can be any land vehicle powered by an engine (such as an internal combustion engine, electric motor, or hybrid engine) that is not bound to rails. For example, in addition to a passenger car, the vehicle 120 can be a tractor, a truck, a motorized vehicle (e.g., a motorcycle or a scooter), a bus, or a self-propelled work machine.

[0025] The vehicle 120 comprises at least one screen 130. According to embodiments of the present invention, the vehicle can thus comprise exactly one screen or a plurality of screens. For the sake of clarity, a single screen is shown in Fig. 1. The at least one screen 130 can, for example, comprise or be a screen of an infotainment system of the vehicle 120. Accordingly, the screen can be installed in the center console of the vehicle, serve as a central interface for the infotainment system, and, for example, display navigation information. 24-2690 5

[0026] Media playback, vehicle status, and other functions can be used. Alternatively or additionally, the at least one screen 130 can be or include a screen that functions as a digital dashboard or digital instrument cluster and displays information such as speed, engine speed, fuel level, warning messages, and navigation instructions. Accordingly, the screen can then be located directly behind the steering wheel in the driver's field of vision. Furthermore, alternatively or additionally, the at least one screen 130 can be or include a screen for rear-seat passengers or vehicle occupants in the rear seats. Accordingly, the screen can, for example, be integrated into the back of one of the front seats or mounted elsewhere in the rear area of ​​the vehicle 120 to provide rear-seat occupants with access to entertainment, navigation, or communication functions.The at least one screen 130 can therefore be any screen arranged in the vehicle 120 for displaying information, for interaction with the vehicle occupant, or for supporting navigation and safety. The at least one screen 130 can use any suitable technology for displaying visual content and may, for example, be a liquid crystal display (LCD) screen, a light-emitting diode (LED) screen, an organic light-emitting diode (OLED) screen, a thin-film transistor (TFT) screen, an active matrix organic light-emitting diode (AMOLED) screen, a microLED screen, or a combination of several of these. According to exemplary embodiments, the at least one screen 130 can be a touchscreen.

[0027] The server or cloud 140 comprises a device 100 for context-dependent background generation. The device comprises a processing circuit 110. For example, the processing circuit 110 may be formed by or comprise a single dedicated processor, a single shared processor, or a plurality of single processors, some or all of which may be used jointly, a microcontroller, an application-specific integrated circuit (ASIC), an integrated circuit (IC), a system-on-a-chip (SoC), a programmable logic element, or a field-programmable gate array (FPGA) with a microprocessor on which software for screen control runs according to the principles described herein. The processor may, for example, be a computer processor (CPU).The processor may be a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a neuromorphic processor, and / or a tensor processor. The processor 110 may also be connected to a memory, such as a read-only memory (ROM) for storing software, a random access memory (RAM), and / or a non-volatile memory. For example, the device 100 may include or be coupled to a memory 24-2690 6 configured to store instructions which, when executed by the processor 110, cause the processor 110 to perform the steps and procedures described herein for context-dependent background generation.

[0028] The processing circuit 110 is configured to receive input data 101. The input data 101 indicates the context of the vehicle 120 and / or at least some of its occupants. The context of the vehicle 120 and / or at least some of its occupants is the entirety of the current conditions, circumstances, and information relating to the vehicle and / or at least some of the vehicle occupants. The input data 101 can thus, for example, display vehicle-related information such as the current position of the vehicle 120, a planned destination of the vehicle 120 (e.g.,a navigation destination stored in the vehicle's navigation system 120), (current) weather conditions at the vehicle's current position, weather conditions at the vehicle's planned destination (current or at a planned or estimated time of arrival at the destination), the current time of day at the vehicle's current position, or an estimated time of day at the vehicle's planned destination. Alternatively or additionally, the input data 101 can, for example, specify occupant-related information indicating characteristics of at least some of the vehicle occupants, such as the number of occupants, the (e.g., estimated) age of at least some of the occupants (e.g., adult, child, senior), the gender of at least some of the occupants, the identity of at least some of the occupants, or activities of at least some of the occupants (e.g., driving the vehicle 120, sleeping, playing, listening to music, working).Furthermore, alternatively or additionally, the input data 101 can display other relevant circumstances, such as the current traffic situation or events near the vehicle's current position or destination. It should be noted that the present invention is not limited to the aforementioned examples. The input data 101 can display more, less, or different contextual information.

[0029] The input data 101 is received at least partially by the vehicle 120 via an interface 150 of the server or cloud 140 (e.g., wirelessly via a cellular network and / or a local wireless network). The vehicle 120 is configured to collect context information within the vehicle 120, indicating the context of the vehicle 120 and / or at least some of its occupants, and to send (transmit) corresponding data to the server or cloud 140 via a corresponding interface 124 in the vehicle. The context information can originate from various sources within the vehicle 120. For example, at least some of the context information can originate from a navigation system 121 of the vehicle 120. The navigation system 121 can, for example, provide information about the current position of the vehicle 120.

[0030] Vehicle 120 provides one or more planned destinations (e.g., a final destination and planned intermediate stops), a planned route, the current time of day at the vehicle's current position, and an estimated time of day for arrival at the vehicle's planned destination. Alternatively or additionally, at least some of the contextual information can originate from at least one sensor 122 of the vehicle 120. For example, a positioning sensor of the vehicle 120 (which uses, for example, a global navigation satellite system such as GPS, GLONASS, Galileo, or Beidou) can provide information about the current position of the vehicle 120. Temperature and light sensors of the vehicle 120 can, for example, provide information about the weather conditions and time of day at the vehicle's current position. Occupant monitoring sensors of the vehicle 120 can, for example, provide information about the number, positions, and characteristics (e.g.,The input data 101 (age, gender, activities) of the occupants of the vehicle 120 can be provided. Alternatively or additionally, at least some of the contextual information can originate from an interface 123 of the vehicle 120 to the internet (which can be identical to the interface 124). In this way, vehicle-external data sources, such as online weather services, can be used to provide information about the weather conditions at the current position of the vehicle and / or at the destination. According to exemplary embodiments, the input data 101 can be received exclusively by the vehicle 120. Alternatively, the interface 150 can also receive some of the input data 101 from sources other than the vehicle 120. For example, some of the input data 101 can be received from vehicle-external data sources, such as the aforementioned online weather services. It should be noted that the present invention is not limited to the aforementioned examples.The input data 101 can be received from more, fewer, or other sources. Likewise, contextual information in the vehicle 120 can be collected from more, fewer, or other sources.

[0031] The processing circuit 110 is further configured to generate a context-dependent background for the at least one screen 130 of the vehicle 120 using a trained, generative machine learning model, which receives the input data 101 as input. The respective context-dependent background for the at least one screen 130 of the vehicle 120 is output by the trained, generative machine learning model in response to the input data 101.

[0032] A background or background image for a screen generally forms the rearmost or lowest layer of visual elements displayed on the screen and serves as a visual backdrop for visual elements displayed in one or more layers in front of, above, or higher than it. The visual elements displayed in these layers can be diverse and, for example, 24-2690 8

[0033] User interface elements (such as buttons, widgets, menus, icons, or text information) or information displays are presented at a lower or higher level in front of / above the background to ensure their functionality and legibility. A background generated according to the proposed invention is therefore designed to support and highlight the higher-level elements, such as user interface elements or information displays, rather than obscuring or impairing them.

[0034] The generative machine learning model is a data structure and / or a set of rules representing a statistical model that uses the processing circuit 110 to generate new data instances in the form of backgrounds similar to a given dataset (e.g., predefined training backgrounds). Unlike traditional models that perform tasks such as classification or regression based on input data, generative machine learning models learn the underlying patterns and distributions in training data, enabling them to generate new data points similar to the original data in the training dataset.

[0035] The generative machine learning model is trained by a generative machine learning algorithm. The term "generative machine learning algorithm" refers to a set of instructions used to create or train a generative machine learning model. To enable the generative machine learning model to generate new backgrounds, it can be trained with training data such as predefined training backgrounds, images, or text as input. By training the generative machine learning model with (e.g., a large amount of) training data, the model "learns" the probability distribution of the data. The generative model can then generate new data instances in the form of backgrounds by extracting values ​​from this learned distribution. The training data can, in particular, conform to predefined visual and stylistic guidelines.These guidelines ensure that the trained generative machine learning model is trained, or becomes trained, to generate the respective context-dependent background according to predetermined visual and stylistic guidelines. These predetermined visual and stylistic guidelines can also be considered style guides. For example, the predetermined visual and stylistic guidelines might be defined by the vehicle manufacturer, ensuring that the generated backgrounds conform to the manufacturer's desired brand image and guarantee a consistent visual identity. The predetermined visual and stylistic guidelines might include, for example, the color palette, image composition, design elements, typography, imagery, and other visual characteristics for the background.The generative machine learning model trained in this way enables the automatic creation of backgrounds that reflect the visual identity of the vehicle's brand, without the need to manually design each background. This significantly reduces the effort and cost of background creation, as the generative machine learning model can generate new, context-dependent backgrounds at any time, while still maintaining brand consistency.

[0036] The generative machine learning model can encompass specific architectures tailored to the task of generating new backgrounds. For example, the generative machine learning model can be a Generative Adversarial Network (GAN). A GAN consists of two neural networks: a generator and a discriminator. The generator is trained to produce data samples that resemble the training data, while the discriminator is trained to distinguish between genuine data samples from the training set and fake data samples generated by the generator. Both networks are trained concurrently in a process called adversarial training. As training progresses, the generator learns to produce increasingly realistic data, which the discriminator finds increasingly difficult to distinguish from the real data.

[0037] Alternatively, the generative machine learning model can be a Variational Autoencoder (VAE). A VAE is a type of neural network that learns a compressed representation (latent space) of the input training data (e.g., predefined backgrounds and / or predetermined visual and stylistic guidelines). The VAE consists of two parts: an encoder, which maps input data into a latent space, and a decoder, which reconstructs the input data from the latent space representation. During training, the VAE learns to encode the input training data into a latent space that captures the essential features of the data, thus enabling the generation of new data samples or backgrounds by sampling from this latent space.

[0038] The generative machine learning model can also be implemented in a multi-stage process where different types of machine learning models work together to achieve a final result. For example, the generative machine learning model can use a combination of a text generation machine learning model and an image generation machine learning model. In one implementation example, a natural language processing machine learning model, such as Claude or GPT-X (where X represents the version number), is used to analyze a given context and generate a corresponding text description. This text description is then fed into another generative machine learning model, such as Stable Diffusion or DALL-E, which is designed to generate images based on text input.Another example is the use of a language model such as T5 (Text-to-Text Transfer Transformer) to generate image descriptions, which are then used by image generation models such as Midjoumey or VQ-VAE 24-2690 10.

[0039] (Vector Quantized Variational Autoencoder) are used to create appropriate backgrounds. The combination of these models enables the dynamic creation of the respective background for at least one screen (130), which is contextually relevant and visually coherent.

[0040] The generative machine learning model can be trained using supervised learning or semi-supervised learning, as illustrated in the examples. In semi-supervised learning, the generative machine learning model is trained with a combination of labeled and unlabeled data, where some training samples have corresponding output values ​​and others do not. In supervised learning, all training samples have corresponding output values. By specifying both training samples and desired output results, the generative machine learning model "learns" to generate data or backgrounds that exactly match the desired outputs for inputs similar to the samples seen during training. For example, different contexts or...Contextual information is used as training samples and predetermined backgrounds as associated output values ​​for training.

[0041] Aside from supervised learning, the generative machine learning model can also be trained using unsupervised learning. In unsupervised learning, only input data is provided, and the generative machine learning model learns the underlying structure and distribution of the input data without requiring labeled output values. For example, the generative machine learning model can learn to generate backgrounds by learning from a large dataset of selected backgrounds, even if they are not labeled.

[0042] In some examples, the generative machine learning model can also be trained using reinforcement learning. Reinforcement learning involves training one or more software agents that perform actions in an environment to maximize cumulative rewards.

[0043] Furthermore, additional techniques can be applied to the generative machine learning algorithms. For example, feature learning can be used. The generative machine learning model can be trained, at least partially, using feature learning, whereby the generative machine learning model learns useful representations of the input data that capture significant patterns relevant to the generation task. 24-2690 11

[0044] Feature learning can include techniques such as principal component analysis or cluster analysis as preprocessing steps before generating new data.

[0045] In some examples, the generative machine learning model can be a hybrid model that combines different generative approaches to leverage their respective strengths. For example, a hybrid model can integrate elements of both GANs and VAEs to improve the quality of the generated data or backgrounds by balancing the advantages of GAN adversarial training with the latent spatial representation learned by VAEs.

[0046] In some examples, the generative machine learning model can be a combination of the above examples, leveraging the strengths of several generative architectures to achieve better performance in generating new backgrounds.

[0047] The context-dependent background generated for at least one screen 130 of the vehicle 120 using the trained, generative machine learning model is designed to be tailored to the specific context displayed by the input data 101. Each background serves as a visual design element on the respective screen and is intended to provide an appealing and relevant display for the occupants. As illustrated in the examples, each background is designed to conform to specific, predefined visual and stylistic guidelines, such as those established by the vehicle manufacturer.

[0048] The background is designed to provide a visually appealing and relevant display for the occupants, relating to the current context of the vehicle and / or at least some of its occupants. This creates a dynamic and immersive experience for the occupants. Because the background is context-dependent, it promotes greater interactivity and enhances the user experience by displaying content that is relevant and useful to the specific driving situation and the needs of the occupants. The vehicle experience can thus be personalized.

[0049] Interface 150 is configured to output data 102, which displays the respective context-dependent background for at least one screen 130 of the vehicle 120, to the vehicle 120 (e.g., wirelessly via a cellular network and / or a local wireless network). Interface 124 of the vehicle 102 receives the output data 102 accordingly in response to the transmission of the data displaying the collected context information to the server or cloud 140. 24-2690 12

[0050] A control circuit 125 of the vehicle 120 is configured to control at least one screen 130 of the vehicle 120, to display the respective context-dependent background, or to use it for the visual display of information. For example, the control circuit 125 can generate a control signal or control data 103 for the at least one screen 130 of the vehicle 120 and output or transmit this to the at least one screen 130 of the vehicle 120. Based on the control signal or control data 103, the at least one screen 130 of the vehicle 120 displays the respective context-dependent background. In other words, the control circuit 125 sends a control signal or corresponding data 103 to the at least one screen 130 of the vehicle 120, which causes the respective context-dependent background generated for this purpose to appear on the at least one screen 130 of the vehicle 120.

[0051] The device 100, or the vehicle 120, thus enables automatic adjustment of the background of the vehicle 120's at least one screen 130 based on the current context of the vehicle 120 and / or its occupants. This improves usability, as the occupants of the vehicle 120 receive an appealing and relevant visual environment on the vehicle 120's at least one screen 130. Since the generative machine learning model is capable of autonomously generating a variety of visual content for the background, the need for complex, manual design of the screen background is reduced. The costs for developing and maintaining screen backgrounds can be significantly reduced, as the dynamic generation of content by the generative machine learning model is automated.The generation of screen backgrounds using the generative machine learning model takes place within a few seconds, so that the development times for generating new screen backgrounds can also be significantly reduced.

[0052] According to exemplary embodiments, the processing circuit 110 can be configured to continuously receive the input data 101 and dynamically (i.e., not statically or with a delay, but in real time or near real time) generate the respective context-dependent background using the trained, generative machine learning model. For example, the processing circuit 110 can continuously and in real time process data streams that reflect the current context of the vehicle 120 and / or the occupants and continuously generate new backgrounds or adapt existing backgrounds to ensure that the background displayed on the respective screens always reflects the current context. The continuous and dynamic adaptation of the backgrounds leads to an improved user experience. The occupants receive visually appealing content that is always tailored to their current situation and environment.This creates a more immersive and engaging experience that adapts flexibly to changing circumstances.

[0053] For example, if a family with two children sets off from Munich to Italy on holiday in vehicle 120 and a location in Tuscany is entered as the destination into the navigation device of vehicle 120, the destination entered into the navigation device, the estimated time of arrival at the destination by the navigation device, and characteristics of the family detected by vehicle sensors (e.g., two adults and two children detected in vehicle 120) are provided as context information, which can be supplemented with further context information from the internet, such as the weather at the destination at the current time or the estimated time of arrival.According to the present invention, a background is generated using the trained, generative machine learning model. This background may, for example, depict a Tuscan landscape with the current weather or the weather at the estimated time of arrival, and a family consisting of two adults and two children engaged in an activity (e.g., hiking). The background is generated, for example, after the destination is entered into the navigation device of the vehicle 120 and displayed on the at least one screen 130 of the vehicle 120.

[0054] If, for example, the family decides to make a stop at a wildlife park or zoo along their route and enters this as a destination into the vehicle's navigation system, the background for at least one screen (130) of the vehicle (120) will be dynamically updated based on the updated context using the trained, generative machine learning model. For example, a family at the wildlife park or zoo can then be displayed along with the corresponding animals and the prevailing weather conditions.

[0055] If the journey to Tuscany continues after the stop at the wildlife park or zoo, the background for at least one screen (130) of the vehicle (120) is dynamically updated again using the trained, generative machine learning model, displaying, for example, a Tuscan landscape with updated time of day and weather. If the sun sets shortly before arrival at the destination, the background design is also adjusted based on the updated context.

[0056] In the embodiment shown in Fig. 1, the device 100 is a component of the server or cloud 140. However, the present invention is not limited to this. The device 100 can also be a component of the vehicle. This is illustrated in Fig. 2. Fig. 2 schematically shows a vehicle 220 that includes the device 100. Since no data exchange with an external server or cloud is necessary for background generation, the vehicle does not include the interface 124 24-2690 14, unlike the vehicle 120. The functionality of the control circuit 125 in the vehicle 120 is integrated into the processing circuit 110 of the device 100 in the vehicle 220 (however, the present invention is not limited to this). Otherwise, the processing circuit 110 operates as described above in connection with Fig. 1. Since the processing circuit 110 in the embodiment shown in Fig.2 is integrated into the vehicle 220, the context information contained in the input data 101 can be received directly from various sources in the vehicle such as the navigation system 121, the sensor 122 or the interface 123 to the Internet.

[0057] Background generation within the vehicle can enable reduced data exchange and lower latencies compared to background generation outside the vehicle.

[0058] To further illustrate the context-dependent background generation described above, Fig. 3 schematically shows a flowchart of a method 300 according to the invention for context-dependent background generation. The method 300 comprises receiving 302 input data that indicates a context of the vehicle and / or at least some of its occupants. Furthermore, the method 300 comprises generating 304 a respective context-dependent background for at least one screen of the vehicle using a trained, generative machine learning model that receives the input data as input.

[0059] Analogous to device 100, method 300 enables the automatic adjustment of the background of at least one of the vehicle's screens based on the current context of the vehicle and / or its occupants. The vehicle's occupants can thus be provided with an appealing and relevant visual environment on the vehicle's at least one screen. The costs for developing and maintaining screen backgrounds can be significantly reduced, as the dynamic generation of content is performed automatically by the generative machine learning model. Likewise, the development time for screen backgrounds can be significantly reduced.

[0060] Further details and aspects of Method 300 are described in more detail above. Method 300 can therefore additionally include one or more of the aspects described above. For example, the input data can be received continuously, and the respective context-dependent background can be dynamically generated using the trained, generative machine learning model. The generation of the respective context-dependent background based on the input data can take place in the vehicle as described above, and Method 300 can accordingly also include controlling at least one screen of the vehicle to display the respective context-dependent background. Alternatively, the generation of the respective context-dependent background based on the input data can take place on a server or in the cloud.The input data can be received at least partially by the vehicle at the server or cloud – analogous to the explanations above. Method 300 can then further include outputting data to the vehicle, which displays the respective context-dependent background for at least one of the vehicle's screens.

[0061] Various methods for training the generative machine learning model have been described previously. Below, with reference to Fig. 4, another training method is described in more detail. Fig. 4 shows a flowchart of a method 400 for training a generative machine learning model to generate a context-dependent background for at least one screen of a vehicle as output, based on input data that indicates a context of the vehicle and / or at least some of its occupants.

[0062] Procedure 300 comprises fine-tuning 302 of a pre-trained generative machine learning model based on training data that specifies predetermined visual and stylistic guidelines, in order to train the model to adhere to these guidelines when generating the respective context-dependent background. The pre-trained generative machine learning model is a model that has been pre-trained on a large amount of general data before being further adapted according to Procedure 300 for the specific application to generate the respective context-dependent background. The pre-trained generative machine learning model is further trained by fine-tuning using the training data that specifies predetermined visual and stylistic guidelines. The training data can consist of backgrounds or...background images consist of specific visual features, colors, layouts or styles that conform to predetermined visual and stylistic guidelines (e.g. brand-specific style guides, color schemes, layouts according to the specifications of a vehicle manufacturer).

[0063] During the fine-tuning process, the parameters (e.g., weights and biases) of the pre-trained generative machine learning model are adjusted. This adjustment can be achieved, for example, through backpropagation. Backpropagation is an error correction method that optimizes the model to minimize errors (differences between the generated output and the training data). Through this optimization, the pre-trained generative machine learning model adapts its internal parameters to better meet the visual and stylistic requirements of the training data. This allows the pre-trained generative machine learning model to learn how to generate context-dependent backgrounds according to the desired style guidelines. The pre-trained generative machine learning model might be, for example, a stable diffusion model. However, the process is not limited to this specific model.Further models are mentioned as examples in connection with the description of Fig. 1 above. For fine-tuning, the Dreambooth technique can be used, for example, to further train the pre-trained generative machine learning model with the training data and adapt it to the specific visual and stylistic requirements. Dreambooth makes it possible to train the generative machine learning model with small datasets so that it adheres precisely to the desired visual and stylistic guidelines. However, other techniques can also be used for fine-tuning, as shown in the examples.

[0064] Furthermore, the method 400 includes the insertion 404 of control layers into the pretrained generative machine learning model. These control layers allow the specification of positions of one or more objects in the respective foreground for at least one screen and control the pretrained generative machine learning model to take these positions into account when generating the respective context-dependent background. The inserted control layers serve to inform the generative machine learning model of the positions of objects (such as widgets, symbols, or other visual elements) in the foreground of the screen. The insertion of these control layers makes it possible to control the generative machine learning model in such a way that it takes these specific specifications into account when creating a background. This ensures that important objects or symbols that must always remain visible (e.g.,Vehicle controls, widgets, or navigation instructions) are correctly positioned and not obscured or affected by the generated background. Control layers use control signals and reference backgrounds or information to learn how specific objects should be positioned in the output of the generative machine learning model. This can be achieved by training the pre-trained generative machine learning model on a dataset of training data containing backgrounds with appropriate markers or annotations specifying the desired positions and characteristics of the objects. For example, the ControlNet technique can be used to insert the control layers. ControlNet enables precise control over the position and placement of objects, symbols, widgets, or other elements within the generated background.However, other techniques can also be used for inserting the control layers, as shown in the examples. 24-2690 17.

[0065] According to method 400, a trained, generative machine learning model can be obtained for generating backgrounds for screens in vehicles that meet certain visual and functional requirements.

[0066] Reference numeral list 0 Device for context-dependent background generation 1 Input data 2 Data that displays a respective context-dependent background 3 Control data or control signal 0 Processing circuit 0 Vehicle 1 Navigation system 2 Sensor 3 Internet interface 4 Interface 5 Control circuit 0 Screen 0 Server or cloud 0 Interface 0 Vehicle 0 Method for context-dependent background generation 2 Receiving input data 4 Generating a respective context-dependent background 0 Method for training a generative machine learning model 2 Fine-tuning a pre-trained generative machine learning model 4 Inserting control layers into the pre-trained generative machine learning model

Claims

24-2690 19 Patent claims 1. Device (100) for context-dependent background generation, wherein the device (100) comprises a processing circuit (110) which is configured as follows: to receive input data indicating a context of a vehicle (120, 220) and / or at least some of its occupants; and to generate a respective context-dependent background for at least one screen (130) of the vehicle (120, 220) using a trained, generative machine learning model that receives the input data as input.

2. Device (100) according to claim 1, wherein the processing circuit (110) is configured to continuously receive the input data and dynamically generate the respective context-dependent background using the trained, generative machine learning model.

3. Device (100) according to claim 1 or claim 2, wherein the input data specifies one or more of the following contextual information: a current position of the vehicle (120, 220), a planned destination of the vehicle (120, 220), weather conditions at the current position of the vehicle (120, 220), weather conditions at the planned destination of the vehicle (120, 220), current time of day, estimated time of day at arrival at the planned destination of the vehicle (120, 220), characteristics of at least some of the vehicle occupants.

4. Device (100) according to any one of claims 1 to 3, wherein the trained generative machine learning model is trained to generate the respective context-dependent background according to predetermined visual and stylistic guidelines.

5. Vehicle (220) comprising: at least one screen (130); and a device (100) for context-dependent background generation according to one of claims 1 to 4, wherein the processing circuit (110) of the device (100) is further configured to control the at least one screen (130) and to output the respective context-dependent background.

6. Vehicle (220) according to Annex 5, wherein the processing circuit (110) of the device (100) is configured to receive input data from one or more of the following: 24-2690 20 a navigation system (121) of the vehicle (220), at least one sensor (122) of the vehicle (220) and an interface (123) of the vehicle (220) to the Internet.

7. Vehicle (120) comprising: at least one screen (130); an interface (124) configured to send first data (101), indicating a context of the vehicle (120) and / or at least some of its occupants, to a server or cloud (140); and to receive second data (102), indicating a context-dependent background for the at least one screen (130) of the vehicle (120), in response to the sending of the first data (101) from the server or cloud (140); and a control circuit (125) configured to control the at least one screen (130) to output the respective context-dependent background.

8. Server or cloud (140) comprising: a device (100) for context-dependent background generation according to any one of claims 1 to 4; and an interface (150) configured to receive at least part of the input data (101) from the vehicle (120); and Output data (102), which displays the respective context-dependent background for at least one screen (130) of the vehicle (120), to the vehicle (120).

9. Method (300) for context-dependent background generation, comprising: Receiving (302) input data indicating a context of the vehicle and / or at least some of its occupants; and Generating (304) a respective context-dependent background for at least one screen of the vehicle using a trained, generative machine learning model which receives the input data as input. 24-2690 21 10. Method (300) according to claim 8, wherein the input data is continuously received and the respective context-dependent background is dynamically generated by means of the trained, generative machine learning model.

11. Method (300) according to claim 9 or claim 10, wherein the generation (304) of the respective context-dependent background is based on the input data in a server or a cloud, wherein the input data is at least partially received by the vehicle at the server or cloud, and wherein the method (200) further comprises: Outputting data to the vehicle that displays the respective context-dependent background for at least one of the vehicle's screens, wherein the generation (304) of the respective context-dependent background is based on the input data in the vehicle, and wherein the method (200) further comprises: Initiating the display of at least one screen in the vehicle to output the respective context-dependent background.

12. Method (300) according to claim 9 or claim 10, wherein the generation (304) of the respective context-dependent background is based on the input data in the vehicle, and wherein the method (200) further comprises: Initiating the display of at least one screen in the vehicle to output the respective context-dependent background.

13. Method (400) for training a generative machine learning model to generate a respective context-dependent background for at least one screen of a vehicle as output based on input data indicating a context of the vehicle and / or at least some of its occupants, the method (400) comprising: Fine-tuning (402) of a pre-trained, generative machine learning model based on training data that specifies predetermined visual and stylistic guidelines in order to train the model to adhere to the predetermined visual and stylistic guidelines when generating the respective context-dependent background; and Insertion (404) of control layers into the pretrained generative machine learning model, which enable the specification of positions of one or more objects in the respective foreground for at least one screen and control the pretrained generative machine learning model to take the positions of the one or more objects into account when generating the respective context-dependent background. 24-2690 22 14. Non-volatile, machine-readable medium on which a program is stored, comprising program code for performing one of the methods according to any one of claims 9 to 14, when the program is executed on a processor or a programmable hardware component.

15. Program comprising program code for performing one of the methods according to any one of claims 9 to 14, when the program is executed on a processor or a programmable hardware component.

Images