3D Scene Image Generation with Layout Control and Gaussian Splatting

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for generating synthetic digital images of three-dimensional scenes lack the ability to accurately control object positions and orientations, leading to inefficiencies in training and testing machine learning systems.

Innovation Solution

A computer-implemented method using text prompts to describe three-dimensional layouts and styles, combined with three-dimensional Gaussian Splatting and stable diffusion techniques, allows for precise object placement and rendering of complex scenes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing methods for generating synthetic digital images are used, then the generation process is simple, but the ability to control object positions and orientations is insufficient

Engineering Contradiction:

Improvecontrol precision of object positions and orientationsVSAvoidcomplexity of generation process

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The generation process is segmented into distinct stages: text prompt interpretation, layout generation with bounding boxes, scene assembly, Gaussian Splatting representation, and stable diffusion rendering. Each stage handles specific aspects of scene generation, allowing precise control at each step while maintaining overall process manageability

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The method performs preliminary actions by first generating a layout with bounding boxes that define object positions and orientations before actual scene assembly. This preliminary structuring enables precise control over final object placement without requiring complex adjustments during rendering

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If text prompts with detailed layout descriptions are provided, then object-level control is achieved, but the processing complexity increases

Engineering Contradiction:

Improveobject-level control capabilityVSAvoidprocessing complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The layout with bounding boxes serves as an intermediary representation between the text prompt and the final scene. This intermediate structure translates natural language descriptions into precise spatial constraints, enabling object-level control while simplifying the subsequent scene assembly process

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The method changes parameters by converting text descriptions into quantitative bounding box parameters (position coordinates, orientation angles, dimensions). This parameter transformation enables precise control over object properties while maintaining a systematic processing approach

Inventive Principle:

Principle #35Parameter changes

3Manufacturing precision

If three-dimensional Gaussian Splatting representation is used, then rendering accuracy is improved, but computational requirements increase

Engineering Contradiction:

Improverendering accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Manufacturing precisionVSUse of energy by moving object

Solution Approach 1:

The method dynamically adjusts the level of Gaussian Splatting detail based on scene requirements. By adapting the representation complexity to the specific rendering needs, the system achieves high accuracy where necessary while reducing computational overhead in less critical areas

Inventive Principle:

Principle #15Dynamics

4Manufacturing precision

If stable diffusion is applied for style determination, then image quality is enhanced, but generation time increases

Engineering Contradiction:

Improveimage qualityVSAvoidgeneration time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The stable diffusion process is applied in advance to generate style-consistent images before final scene assembly. This preliminary style determination allows the main rendering process to focus on structural accuracy, reducing overall generation time while maintaining high image quality

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20260024273A1Device and computer implemented method for generating a synthetic digital image of a three-dimensional scene

Publication Date: 2026.01.22 ROBERT BOSCH GMBH

AI summary

A device and a computer implemented method for generating a synthetic digital image of a three-dimensional scene, in particular for a dataset for training and/or testing of a machine learning system. The method includes providing at least one text prompt which includes a description of a three-dimensional layout of the scene, wherein the at least one text prompt comprises a description of a style of the scene, generating the layout depending on the description of the layout, assembling the scene depending on the layout, determining a three-dimensional Gaussian Splatting representation of the assembled scene depending on the assembled scene, rendering a digital image from the three-dimensional Gaussian Splatting representation, and determining the synthetic digital image with a stable diffusion depending on the digital image and the description of the style.