← ArchivePaper2020
NiLBS: Neural Inverse Linear Blend Skinning
Timothy Jeruzalski, David I. W. Levin, Alec Jacobson, Paul Lalonde, Mohammad Norouzi, Andrea Tagliasacchi
Uses a pose-conditioned neural network to invert LBS deformations, enabling efficient canonical-space queries such as signed distance lookups for deformed characters.
Abstract
In this technical report, we investigate efficient representations of articulated objects (e.g. human bodies), which is an important problem in computer vision and graphics. To deform articulated geometry, existing approaches represent objects as meshes and deform them using "skinning" techniques. The skinning operation allows a wide range of deformations to be achieved with a small number of control parameters. This paper introduces a method to invert the deformations undergone via traditional skinning techniques via a neural network parameterized by pose. The ability to invert these deformations allows values (e.g., distance function, signed distance function, occupancy) to be pre-computed at rest pose, and then efficiently queried when the character is deformed. We leave empirical evaluation of our approach to future work.
How to read this
- Category
- Method: neural inversion of linear blend skinning
- Contributions
- A pose-conditioned neural network that inverts traditional LBS deformations
- Enables values such as distance, signed distance, or occupancy to be precomputed at rest pose and queried efficiently when the character deforms
- Context
- Frames articulated-object representation against SMPL-style skinned body models, proposing a learned inverse of the standard skinning operation rather than a new forward deformer.Builds on: SMPL: A Skinned Multi-Person Linear Model
- Correctness
- Presented as a technical report that explicitly leaves empirical evaluation to future work, so the inversion is proposed and motivated rather than validated; treat reported behavior as a design sketch, not a benchmarked result.
- Clarity
- Short and readable as a report; the LBS-inversion idea comes across in one pass.
- How to read it
- A single first pass is enough to capture the canonical-space query idea; only revisit for the parameterization details, and note the absence of empirical results when comparing to later work.
Builds on
Built upon by
Nothing yet.
Related work
- NeuroSkinning: Automatic Skin Binding for Production Characters with Deep Graph Networks 2019 / SIGGRAPH
- A Neural Network Model for Efficient Musculoskeletal-Driven Skin Deformation 2024 / SIGGRAPH
- SCANimate: Weakly Supervised Learning of Skinned Clothed Avatar Networks 2021 / CVPR
- A Statistical Model of Human Pose and Body Shape 2009 / CGF
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 →