← ArchivePaper2024
Multi-Resolution Real-Time Deep Pose-Space Deformation
Deep network generates multi-resolution skeleton-driven soft-body shapes at sub-millisecond speeds for real-time game applications.
Abstract
We present a hard-real-time multi-resolution mesh shape deformation technique for skeleton-driven soft-body characters. Producing mesh deformations at multiple levels of detail is very important in many applications in computer graphics. Our work targets applications where the multi-resolution shapes must be generated at fast speeds ("hard-real-time", e.g., a few milliseconds at most and preferably under 1 millisecond), as commonly needed in computer games, virtual reality and Metaverse applications. We assume that the character mesh is driven by a skeleton, and that high-quality character shapes are available in a set of training poses originating from a high-quality (slow) rig such as volumetric FEM simulation. Our method combines multi-resolution analysis, mesh partition of unity, and neural networks, to learn the pre-skinning shape deformations in an arbitrary character pose. Combined with linear blend skinning, this makes it possible to reconstruct the training shapes, as well as interpolate and extrapolate them. Crucially, we simultaneously achieve this at hard real-time rates and at multiple mesh resolution levels. Our technique makes it possible to trade deformation quality for memory and computation speed, to accommodate the strict requirements of modern real-time systems. Furthermore, we propose memory layout and code improvements to boost computation speeds.
How to read this
- Category
- Method: a real-time neural pose-space deformation technique
- Contributions
- A multi-resolution mesh deformation method generating skeleton-driven soft-body shapes at hard-real-time (sub-millisecond) rates
- Combines multi-resolution analysis, mesh partition of unity, and neural networks to learn pre-skinning deformations
- Paired with linear blend skinning to reconstruct, interpolate, and extrapolate training poses across resolution levels
- Context
- Extends pose-space deformation and learned deformation approximation (Lewis et al.'s PSD and Bailey et al.'s FDDA) to multiple resolutions at real-time speeds.Builds on: Fast and Deep Deformation Approximations · Pose Space Deformation: A Unified Approach to Shape Interpolation and Skeleton-Driven Deformation
- Correctness
- Assumes high-quality training shapes from a slow rig (e.g. volumetric FEM) are available and that learned deformations generalize across poses; the real-time claim is the central result, so a reader should watch extrapolation quality and dependence on training-pose coverage.
- Clarity
- Accessible to graphics readers; a first pass conveys the goal and pipeline, a second pass for the multi-resolution and partition-of-unity formulation.
- How to read it
- Focus on how multi-resolution analysis combines with the neural pre-skinning model; a second pass pays off for the timing/quality tradeoff and generalization limits.
Builds on
Built upon by
Nothing yet.
Related work
- NeuroSkinning: Automatic Skin Binding for Production Characters with Deep Graph Networks 2019 / SIGGRAPH
- Fast and Deep Deformation Approximations 2018 / SIGGRAPH
- Implementing a Machine Learning Deformer for CG Crowds: Our Journey 2024 / DigiPro
- Data-Driven Physics for Human Soft Tissue Animation 2017 / SIGGRAPH
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →