CoChat

The DOI matches its Crossref record; 58 of the 58 references it cites were found in a registry or an index; 1 of the 33 sources its claims cite were checked for support.

Robust and accurate skeletal rigging from mesh sequences

Le, Binh Huy, Deng, Zhigang · 2014 · ACM Transactions on Graphics

Abstract

We introduce an example-based rigging approach to automatically generate linear blend skinning models with skeletal structure. Based on a set of example poses, our approach can output its skeleton, joint positions, linear blend skinning weights, and corresponding bone transformations. The output can be directly used to set up skeleton-based animation in various 3D modeling and animation software as well as game engines. Specifically, we formulate the solving of a linear blend skinning model with a skeleton as an optimization with joint constraints and weight smoothness regularization, and solve it using an iterative rigging algorithm that (i) alternatively updates skinning weights, joint locations, and bone transformations, and (ii) automatically prunes redundant bones that can be generated by an over-estimated bone initialization. Due to the automatic redundant bone pruning, our approach is more robust than existing example-based rigging approaches. Furthermore, in terms of rigging accuracy, even with a single set of parameters, our approach can soundly outperform state of the art methods on various types of experimental datasets including humans, quadrupled animals, and highly deformable models.

Overview

CoChat Summary

## TL;DR

The paper presents an example-based rigging method that automatically produces a skeleton-based linear blend skinning (LBS) model from a sequence of example poses. It improves robustness by over-initializing bones and pruning redundancies, and improves accuracy by iteratively updating weights, joint locations, and bone transformations with soft joint constraints.

## Key findings

  • The method outputs the skeleton, joint positions, skinning weights, and bone transformations directly from example poses.
  • On 9 test datasets, the paper reports that using a single parameter set, the method outperformed three state-of-the-art methods in robustness and accuracy.
  • In Table 1, the reported RMSEs for the authors’ method are lower than the comparators across all listed datasets, e.g.:
  • cat-poses: 0.25 vs. 0.68 / 1.04 / 0.63
  • horse-poses: 0.21 vs. 0.54 / 1.24 / 0.75
  • hand: 0.18 vs. 0.23 / 0.42 / 0.18
  • cow: 1.52 vs. 5.41 / 5.61 / 5.58
  • The paper reports that the method can rig difficult models with few frames, e.g. 9 frames produced a 28-bone skeleton for the cat model.

## Methods

  • Example-based rigging pipeline with three main stages: initialization from motion-driven clustering, topology reconstruction via skinning decomposition and a minimum spanning tree, and iterative rigging with block coordinate descent.
  • Input: a temporally coherent set of example poses (mesh sequence) plus a rest pose; output: skeletal structure, joint locations, skinning weights, and bone transformations.
  • Main analysis: optimization of an LBS model with data fitting, weight smoothness regularization, and soft joint constraints; redundant bones are pruned during iterative updates.

## Limitations

  • The paper states that the method has low computational efficiency and is suitable only for offline applications.
  • The paper states that results depend on the quality of the input example poses; noise, limited motion range, and asymmetry can affect output.
  • The paper states that the LBS model has limited approximation power and cannot capture some complicated nonlinear deformations.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/2601097.2601161

Cites

58 of 58 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Only using a single set of parameters, our example-based method can accurately rig various models such as quadrupled animals (a), humans (b, c), and highly deformable models (d). Our method can even generate bone structures for challenging parts, such as the mouth and the two ears of the cat (a), the skirt (b), and the elastic cow model (d). Our method is robust: using only 9 frames, it can generate a skeleton with 28 bones for the cat model (a); note that even though the given example poses of the cat model have asymmetric poses, it still can generate an almost symmetric skeleton without imposing any symmetry constraints.
Figure 1 Only using a single set of parameters, our example-based method can accurately rig various models such as quadrupled animals (a), humans (b, c), and highly deformable models (d). Our method can even generate bone structures for challenging parts, such as the mouth and the two ears of the cat (a), the skirt (b), and the elastic cow model (d). Our method is robust: using only 9 frames, it can generate a skeleton with 28 bones for the cat model (a); note that even though the given example poses of the cat model have asymmetric poses, it still can generate an almost symmetric skeleton without imposing any symmetry constraints.From the PDF
Figure 2 Examples that show two common limitations of current rigging methods: (i) an over-estimated bone initialization may generate inaccurate and redundant bones (indicated by red arrows); (ii) an inaccurate estimation of bone transformations (indicated by red arrows with ellipse) causes noticeable deformation errors when the bones are connected by joints (see the change of the legs). The examples in this figure are generated using [Hasler et al. 2010].
Figure 2 Examples that show two common limitations of current rigging methods: (i) an over-estimated bone initialization may generate inaccurate and redundant bones (indicated by red arrows); (ii) an inaccurate estimation of bone transformations (indicated by red arrows with ellipse) causes noticeable deformation errors when the bones are connected by joints (see the change of the legs). The examples in this figure are generated using [Hasler et al. 2010].From the PDF
Figure 3 The pipeline of our skeletal rigging approach
Figure 3 The pipeline of our skeletal rigging approachFrom the PDF
Figure 4 An illustration of our cluster splitting strategy
Figure 4 An illustration of our cluster splitting strategyFrom the PDF
Figure 5 The effect of different weight smoothness regularizations. Compared to the graph Laplacian [Merry et al. 2006; Kim et al. 2010], our rigidness Laplacian regularization offers better smoothness while still preserving the global shape. Note that how the graph Laplacian over-smooths the fracture while it starts to change the global shape (indicated by red circles). Our rigidness Laplacian is robust as demonstrated by the similar results with different ω values.
Figure 5 The effect of different weight smoothness regularizations. Compared to the graph Laplacian [Merry et al. 2006; Kim et al. 2010], our rigidness Laplacian regularization offers better smoothness while still preserving the global shape. Note that how the graph Laplacian over-smooths the fracture while it starts to change the global shape (indicated by red circles). Our rigidness Laplacian is robust as demonstrated by the similar results with different ω values.From the PDF
Figure 6 (Left) Without joint constraints, the bone transformations generated by [Le and Deng 2012] do not always rotate around the joints. (Right) With the soft joint constraints, bones rotate more strictly around the joints. Since bone transformations are alternatively updated with joint locations, our model can converge to a local optimum with a good approximation of the input.
Figure 6 (Left) Without joint constraints, the bone transformations generated by [Le and Deng 2012] do not always rotate around the joints. (Right) With the soft joint constraints, bones rotate more strictly around the joints. Since bone transformations are alternatively updated with joint locations, our model can converge to a local optimum with a good approximation of the input.From the PDF
Figure 7 Redundant bones in the left panel are pruned to achieve the neat skeleton in the right panel. As illustrated in the two yellow boxes, a redundant bone j is identified by utilizing the weight regularization term to force its weights degenerate.
Figure 7 Redundant bones in the left panel are pruned to achieve the neat skeleton in the right panel. As illustrated in the two yellow boxes, a redundant bone j is identified by utilizing the weight regularization term to force its weights degenerate.From the PDF
Figure 8 The skeletons extracted from 3 datasets by our approach (from left to right): cat-poses , dance , and hand . The clustering results obtained at the initialization step are also illustrated via a color-coded scheme. Using the same set of parameters, our approach can robustly determine the optimal number of bones in the skeletons using skeleton pruning.
Figure 8 The skeletons extracted from 3 datasets by our approach (from left to right): cat-poses , dance , and hand . The clustering results obtained at the initialization step are also illustrated via a color-coded scheme. Using the same set of parameters, our approach can robustly determine the optimal number of bones in the skeletons using skeleton pruning.From the PDF
Figure 9 The extracted horse-gallop skeletons with different cluster initializations. Despite very different numbers of initialized clusters, our approach can output similar final skeletons with consistent structure on the trunk and legs; minor differences on the tail and feet are due to the high deformations on these regions.
Figure 9 The extracted horse-gallop skeletons with different cluster initializations. Despite very different numbers of initialized clusters, our approach can output similar final skeletons with consistent structure on the trunk and legs; minor differences on the tail and feet are due to the high deformations on these regions.From the PDF
Figure 10 Comparisons between our method and three state of the art approaches. The five test datasets shown in this figure are (in the clockwise direction starting from the top-left corner): cat-poses , lion-poses , scape , horse-gallop , and hand . For a fair comparison, we set the same number of bones for all the four methods. Only our method can generate sound outputs for all the 5 datasets. The issues in the results are indicated by red arrows.
Figure 10 Comparisons between our method and three state of the art approaches. The five test datasets shown in this figure are (in the clockwise direction starting from the top-left corner): cat-poses , lion-poses , scape , horse-gallop , and hand . For a fair comparison, we set the same number of bones for all the four methods. Only our method can generate sound outputs for all the 5 datasets. The issues in the results are indicated by red arrows.From the PDF
Figure 12 Detailed computation breakdown of our method on the samba model. The number inside a Skeleton-Reconstruction block denotes the number of bones at the current step. The number inside a Rigging-Iterations block denotes the number of iterations executed at the iterative rigging step. Each rigging iteration includes skinning weights update, joint positions update, bone transformations update, and skeleton pruning. The computational times of all the rigging iterations are approximately the same; one detailed breakdown is illustrated in the Last-Rigging-Iteration block.
Figure 12 Detailed computation breakdown of our method on the samba model. The number inside a Skeleton-Reconstruction block denotes the number of bones at the current step. The number inside a Rigging-Iterations block denotes the number of iterations executed at the iterative rigging step. Each rigging iteration includes skinning weights update, joint positions update, bone transformations update, and skeleton pruning. The computational times of all the rigging iterations are approximately the same; one detailed breakdown is illustrated in the Last-Rigging-Iteration block.From the PDF
Our Method
Our MethodFrom the PDF
Method III [de Aguiar et al. 2008a]
Method III [de Aguiar et al. 2008a]From the PDF
Figure 13 Despite the high deformation on the skirt part of the samba model, our approach is still able to generate a reasonable skeletal structure where the skirt is rigged by some bones originated from the hip. Meanwhile, the three previous approaches cannot extract similar skeleton patterns.
Figure 13 Despite the high deformation on the skirt part of the samba model, our approach is still able to generate a reasonable skeletal structure where the skirt is rigged by some bones originated from the hip. Meanwhile, the three previous approaches cannot extract similar skeleton patterns.From the PDF
Ground Truth
Ground TruthFrom the PDF
Figure 15 Our approach fails to remove 4 redundant bones in the upper arms and upper legs of the scape model, since the LBS model need them for a better approximation of muscle bulging on these parts (indicated by red arrows). To the end, our approach keeps the 4 bones to capture the non-linear deformation effect. The gray models are the ground-truth.
Figure 15 Our approach fails to remove 4 redundant bones in the upper arms and upper legs of the scape model, since the LBS model need them for a better approximation of muscle bulging on these parts (indicated by red arrows). To the end, our approach keeps the 4 bones to capture the non-linear deformation effect. The gray models are the ground-truth.From the PDF

Verification

Claims the paper makes

0 Valid · 1 Not valid · 21 Not checked · 72 Own results and interpretations

49 more claims

Create a free account to see all 94 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 25
  • Probably not a substantial claim. 19

Create a free account to see these claims.

Citations

Cites

Show all 58 references

Cited by

See all 115

Metadata

Authors
Le, B. H., & Deng, Z.
Affiliations
University of Houston (US)
Venue
ACM Transactions on Graphics
Published
2014-07-27
Volume
33
Issue
4
Pages
1-10
ISSN
  • 0730-0301 (ISSN-L)
  • 1557-7368
OpenAlex
W2152005648
Cited by
111
References
46
Citations by year
2026: 2 · 2025: 8 · 2024: 7 · 2023: 9 · 2022: 2 · 2021: 15 · 2020: 11 · 2019: 12 · 2018: 12 · 2017: 2 · 2016: 19 · 2015: 10 · 2014: 2
FWCI
9.98
Type
journal article
Language
en
Open access
closed
Locations
ACM Transactions on Graphics (publishedVersion)
Funders
  • National Institutes of Health
  • National Natural Science Foundation of China (61328204)
  • Division of Information and Intelligent Systems
Topics
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › 3D Shape Modeling and Analysis
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Human Motion and Animation
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Computer Graphics and Visualization Techniques
Keywords
linear blend skinning, skeleton estimation, joint position estimation, skeleton-based animation, 3D character animation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Robust and accurate skeletal rigging from mesh sequences. https://app.cochat.ai/papers/10.1145/2601097.2601161
The paper
Le, B. H., & Deng, Z. (2014). Robust and accurate skeletal rigging from mesh sequences. ACM Transactions on Graphics. https://doi.org/10.1145/2601097.2601161

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.