Deep Relightable Appearance Model for Dynamic Face Rendering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for generating relightable three-dimensional computer models of human faces are limited to single lighting conditions, making them unsuitable for dynamic renderings under novel expressions and lighting conditions, which hinders their adoption in applications like game and film production where consistency between character and environment is desirable.

Innovation Solution

A learning-based method using a Deep Relightable Appearance Model (DRAM) that leverages neural networks to generate expression-dependent and view-dependent textures, allowing for real-time rendering under novel viewpoints and lighting conditions, including challenging natural illumination and near-field lighting, through a teacher-student network framework and conditional variational auto-encoders.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional 3D rendering models are used with single lighting condition, then rendering speed is fast, but realism and adaptability to different lighting environments deteriorate

Engineering Contradiction:
Improveadaptability to different lighting conditionsVSAvoidmodel complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The appearance model is segmented into multiple lighting-specific sub-models, each trained on data from a particular lighting condition. This allows the system to maintain simplicity within each sub-model while achieving overall adaptability across diverse lighting environments through selective application of appropriate sub-models.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system creates a universal appearance model framework that can handle multiple lighting conditions through a collection of specialized sub-models. Each sub-model serves a specific lighting condition, but together they provide universal adaptability across all lighting scenarios, making the system multi-functional.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Productivity

If learning-based relighting approaches are applied on 2D images or static scenes, then processing intensity is reduced, but applicability to dynamic renderings under novel expressions and lighting conditions deteriorates

Engineering Contradiction:
Improvereal-time rendering capabilityVSAvoidgeneralization to novel expressions and lighting
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The appearance model is designed to be dynamic rather than static, incorporating temporal information and expression variations. The model adapts to novel expressions and lighting conditions in real-time, enabling dynamic renderings while maintaining generalization capability through continuous adaptation mechanisms.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system performs preliminary training on diverse lighting conditions and expression data to create pre-adapted sub-models. This preliminary action on training data enables the model to generalize better to novel conditions during real-time operation, reducing the need for intensive processing during actual rendering.

Inventive Principle:
Principle #10Preliminary action

3Manufacturing precision

If intensive processing is used to achieve realism, then rendering quality is high, but real-time application capability deteriorates

Engineering Contradiction:
Improverendering qualityVSAvoidreal-time rendering speed
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The system applies partial processing by using simplified rendering pipelines for common lighting conditions while reserving intensive processing only for edge cases or novel conditions that require higher quality. This selective approach maintains real-time performance for most scenarios while achieving high quality when needed.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system creates simplified copies or approximations of complex lighting effects that can be rendered in real-time. Instead of performing full physically-based rendering for all cases, it uses pre-computed approximations and learned representations that replicate realistic lighting behavior at lower computational cost.

Inventive Principle:
Principle #26Copying

Data Source

PatentUS11715248B2Deep relightable appearance models for animatable face avatars
Publication Date: 2023.08.01 META PLATFORMS TECHNOLOGIES LLC
  • US11715248B2 patent drawing
  • US11715248B2 patent drawing
  • US11715248B2 patent drawing

AI summary

A method for providing a relightable avatar of a subject to a virtual reality application is provided. The method includes retrieving multiple images including multiple views of a subject and generating an expression-dependent texture map and a view-dependent texture map for the subject, based on the images. The method also includes generating, based on the expression-dependent texture map and the view-dependent texture map, a view of the subject illuminated by a light source selected from an environment in an immersive reality application, and providing the view of the subject to an immersive reality application running in a client device. A non-transitory, computer-readable medium storing instructions and a system that executes the instructions to perform the above method are also provided.