Static-Dynamic Latent Code Separation for Generative Rendering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing 3D rendering technologies struggle to efficiently separate static and dynamic factors in a scene representation, leading to suboptimal photorealistic rendering results, especially when dynamic factors change over time.

Innovation Solution

A method and device that deconstructs latent code into static and dynamic codes, using separate generative models to generate and compose rendered images, allowing for photorealistic rendering by incorporating static and dynamic scene information effectively.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If existing 3D rendering technologies are used, then rendering can be performed, but the separation of static and dynamic factors is inefficient and photorealistic rendering results are suboptimal

Engineering Contradiction:
Improvephotorealistic rendering resultsVSAvoidrendering efficiency
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The patent segments the latent code into separate static code and dynamic code components, allowing independent processing and optimization of each factor. This segmentation enables the generative model to efficiently separate and render static scene representations from dynamic factors, improving both photorealistic rendering quality and computational efficiency by avoiding redundant processing of static elements.

Inventive Principle:
Principle #1Segmentation

2Manufacturing precision

If extensive rendered databases are used to improve rendering quality, then photorealistic results improve, but the need for extensive databases increases complexity and resource requirements

Engineering Contradiction:
Improvephotorealistic rendering qualityVSAvoiddatabase requirements
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent creates a learned representation (latent code) that captures essential scene information through the generative model's encoding process. This learned representation serves as a compressed copy of scene data that preserves photorealistic rendering quality while significantly reducing the need for extensive original rendered databases, thereby lowering storage requirements and system complexity.

Inventive Principle:
Principle #26Copying

3Device complexity

If static and dynamic factors are not separated, then processing is simpler, but rendering accuracy for dynamic scenes deteriorates

Engineering Contradiction:
Improveprocessing complexityVSAvoidrendering accuracy
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The patent implements segmentation of the latent code into static and dynamic portions, enabling the system to process each factor appropriately. The static code captures time-invariant scene representations while dynamic code handles temporal variations, improving rendering accuracy for dynamic scenes without excessive processing complexity by leveraging the structured separation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces dynamic code that specifically models temporal variations and dynamic factors in the scene. This dynamic component allows the rendering system to adapt to changing conditions over time while maintaining a stable static representation, thereby improving rendering accuracy for dynamic scenes through controlled complexity in the dynamic factor processing.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentUS20250238899A1Rendering method and device and training method for rendering
Publication Date: 2025.07.24 SAMSUNG ELECTRONICS CO LTD
  • US20250238899A1 patent drawing
  • US20250238899A1 patent drawing
  • US20250238899A1 patent drawing

AI summary

A rendering method and device are provided. The rendering method includes receiving a scene representation and a camera view, deconstructing latent code by separating static code corresponding to a static factor and dynamic code corresponding to a dynamic factor in the scene information, generating a first rendered image by inputting the static code and the camera view into a generative model based on an artificial neural network, generating a second rendered image by inputting the dynamic code, the static code, and the camera view into the generative model, and generating an output image by composing the first rendered image and the second rendered image.