Multi-View 2D Image Generation With 3D Scene Latent Blending

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing solutions for generating 2D images from 3D scenes do not adequately account for the 3D environment, leading to inaccurate, non-immersive, and inconsistent images across different viewpoints.

Innovation Solution

A computer-implemented method using a machine-learning model with a scene encoder and generative image model to generate multiple 2D images from a 3D scene layout and viewpoint, incorporating perspective and occlusion, and employing a weighted latent vector combination for consistency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If generative deep learning models are used to generate 2D images from 3D scenes, then image generation capability is provided, but the images fail to accurately represent the 3D environment, perspective, occlusion, and lighting factors

Engineering Contradiction:
Improveimage generation capabilityVSAvoidaccuracy of 2D image representation
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent transitions from 2D image generation to 3D scene understanding by incorporating depth information, spatial relationships, and 3D structural knowledge into the generative model. The model processes 3D scene representations (point clouds, depth maps, or 3D meshes) alongside 2D images to generate accurate multi-view images that preserve 3D geometry and perspective relationships.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent implements feedback mechanisms where generated 2D images are evaluated against 3D scene constraints and previously generated images. The model uses consistency checks across multiple viewpoints and compares generated content with 3D structural information to iteratively improve accuracy, ensuring that generated images faithfully represent the 3D environment.

Inventive Principle:
Principle #23Feedback

2Productivity

If current generative solutions are used, then single 2D image generation is possible, but generating a plurality of visually and functionally consistent images across different viewpoints is not achieved

Engineering Contradiction:
Improvesingle image generation speedVSAvoidvisual consistency across multiple viewpoints
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent merges multiple generation processes into a unified framework that simultaneously generates and coordinates multiple viewpoint images. The system integrates scene graph reasoning, 3D geometry constraints, and multi-view consistency mechanisms into a single model that produces coherent image sets rather than independent single images.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent performs preliminary 3D scene analysis and structure extraction before generating multiple 2D images. By pre-processing to understand the 3D layout, object relationships, and camera configurations, the system establishes a consistent representation that guides subsequent multi-view image generation, ensuring visual and functional consistency across all generated images.

Inventive Principle:
Principle #10Preliminary action

3Ease of manufacture

If existing image generation methods are applied, then basic 2D rendering is achieved, but perspective, occlusion, and lighting factors are not properly accounted for

Engineering Contradiction:
Improvebasic 2D rendering capabilityVSAvoidaccuracy of perspective and occlusion representation
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent segments the image generation task into separate modules that handle different aspects: 3D scene understanding, perspective projection, occlusion reasoning, and lighting simulation. Each module processes specific information (geometry, camera parameters, object relationships, material properties) and outputs refined representations that are combined to produce final accurate images.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent dynamically adjusts multiple parameters including camera position, field of view, object visibility, lighting conditions, and material properties based on the 3D scene representation. The model modifies these parameters across different generation calls to accurately represent varying perspectives, occlusion states, and lighting scenarios while maintaining consistency with the underlying 3D structure.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4675564A1Generating a plurality of 2d images of 3D scene
Publication Date: 2026.01.07 DASSAULT SYSTEMES SA
  • EP4675564A1 patent drawingFigure 1
  • EP4675564A1 patent drawingFigure 2
  • EP4675564A1 patent drawingFigure 3

AI summary

The disclosure notably relates to a computer-implemented method for generating a plurality of 20 images of a 3D scene. The method comprises obtaining arrangement data and a machine-learning model configured for generating a 2D image. The method comprises generating a plurality of first 2D images of the 3D scene each having a respective viewpoint. The method comprises, for each generated first 2D image, computing a first latent vector. The method comprises, for each generated first 2D image, computing a second latent vector as a weighted combination of the computed first latent vectors. The method comprises generating a plurality of second 2D images of the 3D scene by applying, for each given viewpoint of the first 2D images, the model to the computed second latent vector. Such a generating method forms an improved solution for generating a plurality of 20 images of a given 3D scene.