pith. sign in

arxiv: 1907.10163 · v1 · pith:AFRVGC2Mnew · submitted 2019-07-23 · 💻 cs.GR

A system for efficient 3D printed stop-motion face animation

Rinat Abdrashitov , Alec Jacobson , Karan Singh This is my paper

Pith reviewed 2026-05-24 16:44 UTC · model grok-4.3

classification 💻 cs.GR
keywords 3D printingstop-motion animationmesh segmentationgraph-cut optimizationfacial animationdeforming meshesreplacement partsinterchangeable components
0
0 comments X

The pith

A pipeline converts deforming face mesh animations into compact libraries of interchangeable 3D printed parts assigned per frame to approximate the input motion with minimal printing and assembly.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Stop-motion animation requires a new face for nearly every frame, making full 3D printing of each frame too slow and costly. The paper shows how to take an input sequence of topology-invariant deforming meshes and produce a small library of printable replacement parts together with a schedule of which part to use on each frame. The process begins with user marks on preferred cut regions, locates curves of least deformation for the actual cuts, removes deformation at those boundaries so any replacement from one set fits any replacement from the others, and then solves a graph-cut problem independently on each segment to choose the smallest set of replacements that still stay close to the original motion. When the numbers are chosen to respect a printing budget or allowed error, the total printed volume drops sharply while the assembled animation remains visually faithful. Physical prints of several face sequences, reviewed by a professional animator, demonstrate the practical savings compared with printing every frame or using simpler segmentation.

Core claim

Given an input animation sequence of topology invariant deforming meshes, the method outputs a library of replacement parts and per-animation-frame assignment of the parts such that the animation is maximally approximated while the amount of 3D printing and assembly is minimized. User-specified preferred segmentation regions guide the search for minimal-deformation curves; boundary deformations are then zeroed so that parts from each replacement set can be interchanged without seams; each part is optimized separately by graph-cut to produce a user-chosen or budget-constrained number of replacements whose boundaries are shaped for easy printing and assembly.

What carries the argument

User-guided minimal-deformation curve segmentation followed by boundary deformation zeroing and independent graph-cut optimization of replacement sets per segment, which produces interchangeable 3D printable components.

If this is right

  • The number of replacements per part can be set by the user or computed automatically to respect a printing budget or allowed deviation from the original animation.
  • Part boundaries are shaped to simplify both 3D printing and physical assembly instrumentation.
  • The same pipeline produces usable results on both purely digital sequences and sequences that are actually printed and assembled.
  • The approach yields higher fidelity at lower cost than naive full-frame printing or simpler segmentation strategies.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same replacement libraries might be reusable across different animations that share a common base mesh and similar deformation patterns.
  • Extending the segmentation step to run without manual region hints would test whether the rest of the pipeline still succeeds on new mesh sequences.
  • The method could be applied to other topology-preserving deforming objects, such as cloth or articulated props, if the minimal-deformation curves remain well-defined.
  • Physical production trials could measure whether the added step of swapping parts during filming is offset by the large reduction in total printing time.

Load-bearing premise

Manually indicated preferred segmentation regions plus minimal-deformation curves, followed by boundary deformation zeroing, will produce interchangeable parts whose independent optimization still yields a faithful overall animation without visible artifacts at assembly seams.

What would settle it

If physical assemblies of the optimized parts show visible seams or distortions at the cut boundaries during playback, or if the total printed volume fails to fall substantially below the volume required to print every frame, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 1907.10163 by Alec Jacobson, Karan Singh, Rinat Abdrashitov.

Figure 1
Figure 1. Figure 1: Given an input mesh-animation sequence (a), our system segments and deforms the 3D mesh into parts that can be seamlessly joined (b). Each part is [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Modern stop-motion films such as Laika’s [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Algorithm overview. Input shape is segmented into parts and each [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Robustness of the part segmentation method with respect to per [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Left to right: average displacement visualized over the average face [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The input mesh-animation sequence is cut into two parts according to [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Increasing the number of pieces improves accuracy of the approxi [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Without saliency weights optimization sometimes fails to produce [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: A mouth opening animation (middle) is approximated using 2 pieces. Without the velocity term (top) the few frames (a) where the character slowly [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: An open-mouthed surprise animation (middle) is approximated using 2 pieces. The replacement and labeling without velocity term (top) snaps the [PITH_FULL_IMAGE:figures/full_fig_p008_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: For our physically realized results, we generated 20 cartoon charac [PITH_FULL_IMAGE:figures/full_fig_p008_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: We recreate a typical low-budget stop motion camera setup. [PITH_FULL_IMAGE:figures/full_fig_p009_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Number of pieces needed to be 3D printed in order to achieve a [PITH_FULL_IMAGE:figures/full_fig_p009_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Our method works on cartoon characters (a, b, c) as well as high [PITH_FULL_IMAGE:figures/full_fig_p009_17.png] view at source ↗
Figure 19
Figure 19. Figure 19: Top: for comparison, we fix the library to a uniform sampling of the deforming object over time (and then use our assignment optimization). Uniform [PITH_FULL_IMAGE:figures/full_fig_p011_19.png] view at source ↗
read the original abstract

Computer animation in conjunction with 3D printing has the potential to positively impact traditional stop-motion animation. As 3D printing every frame of a computer animation is prohibitively slow and expensive, 3D printed stop-motion can only be viable if animations can be faithfully reproduced using a compact library of 3D printed and efficiently assemblable parts. We thus present the first system for processing computer animation sequences (typically faces) to produce an optimal set of replacement parts for use in 3D printed stop-motion animation. Given an input animation sequence of topology invariant deforming meshes, our problem is to output a library of replacement parts and per-animation-frame assignment of the parts, such that we maximally approximate the input animation, while minimizing the amount of 3D printing and assembly. Inspired by current stop-motion workflows, a user manually indicates which parts of the model are preferred for segmentation; then, we find curves with minimal deformation along which to segment the mesh. We then present a novel algorithm to zero out deformations along the segment boundaries, so that replacement sets for each part can be interchangeably and seamlessly assembled together. The part boundaries are designed to ease 3D printing and instrumentation for assembly. Each part is then independently optimized using a graph-cut technique to find a set of replacements, whose size can be user defined, or automatically computed to adhere to a printing budget or allowed deviation from the original animation. Our evaluation is threefold: we show results on a variety of facial animations, both digital and 3D printed, critiqued by a professional animator; we show the impact of various algorithmic parameters; and compare our results to naive solutions. Our approach can reduce the printing time and cost significantly for stop-motion animated films.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper presents a pipeline for converting topology-invariant deforming mesh animation sequences (primarily faces) into a compact library of 3D-printable replacement parts together with per-frame part assignments. The objective is to maximize fidelity to the input animation while minimizing total printed material and assembly operations. The method begins with user-specified preferred segmentation regions, computes minimal-deformation cut curves, applies a boundary-deformation zeroing step to enable interchangeability, and then performs independent graph-cut optimization on each part to select a user- or budget-constrained replacement set. Results are demonstrated on several facial animations, with qualitative feedback from a professional animator and comparisons against naive baselines.

Significance. If the seam-free interchangeability claim holds, the work provides a concrete, user-controllable bridge between computer animation and traditional stop-motion that could materially lower the cost and time barriers to 3D-printed facial animation. The pipeline incorporates practical printing and assembly considerations and includes both parameter studies and professional critique, which are positive attributes for an applied graphics paper.

major comments (2)
  1. [Boundary deformation zeroing and independent part optimization] The boundary-deformation zeroing step (described after the minimal-deformation curve computation) fixes boundary geometry on the input sequence only; the subsequent independent per-part graph-cut optimization contains no explicit cross-part or post-zeroing constraint that would guarantee chosen replacements continue to match at the fixed boundaries. Consequently, replacements that alter geometry adjacent to the boundary can produce visible discontinuities even when each part individually satisfies its error budget.
  2. [Evaluation] The evaluation section reports results on digital and 3D-printed animations together with professional animator critique and comparisons to naive solutions, yet provides no quantitative per-seam error metric, no controlled measurement of assembled seam deviation, and no formal invariance argument establishing that arbitrary recombination of the independently optimized libraries remains seam-free.
minor comments (2)
  1. [Method overview] The abstract states that the part boundaries are 'designed to ease 3D printing and instrumentation for assembly,' but the manuscript does not detail the concrete geometric or topological criteria used for this design.
  2. [Graph-cut optimization] Notation for the user-specified number of replacements versus the automatically computed budget-constrained size could be clarified to avoid ambiguity when both options are discussed.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and positive assessment of the work's potential. We respond point-by-point to the major comments below.

read point-by-point responses

  1. Referee: [Boundary deformation zeroing and independent part optimization] The boundary-deformation zeroing step (described after the minimal-deformation curve computation) fixes boundary geometry on the input sequence only; the subsequent independent per-part graph-cut optimization contains no explicit cross-part or post-zeroing constraint that would guarantee chosen replacements continue to match at the fixed boundaries. Consequently, replacements that alter geometry adjacent to the boundary can produce visible discontinuities even when each part individually satisfies its error budget.

    Authors: The boundary-deformation zeroing is performed on the complete input sequence prior to segmentation or optimization. This step enforces identical boundary vertex positions across all frames for each user-specified region. Consequently every candidate replacement part in a library inherits exactly the same boundary geometry. Graph-cut selection then operates exclusively over these pre-zeroed candidates; arbitrary recombination therefore preserves boundary alignment by construction. We will add an explicit invariance paragraph and short proof sketch to the method section. revision: partial

  2. Referee: [Evaluation] The evaluation section reports results on digital and 3D-printed animations together with professional animator critique and comparisons to naive solutions, yet provides no quantitative per-seam error metric, no controlled measurement of assembled seam deviation, and no formal invariance argument establishing that arbitrary recombination of the independently optimized libraries remains seam-free.

    Authors: We agree that quantitative seam metrics would strengthen the evaluation. In revision we will add (i) average boundary-vertex deviation measured over all possible part recombinations and (ii) a controlled seam-deviation measurement on the assembled 3D-printed models. The formal invariance argument follows directly from the pre-optimization zeroing step described above; we will state it explicitly alongside the new metrics. revision: yes

Circularity Check

0 steps flagged

No circularity: algorithmic pipeline is externally evaluated and self-contained

full rationale

The paper presents a multi-stage algorithmic pipeline (user-specified regions, minimal-deformation curve segmentation, boundary zeroing, then independent per-part graph-cut optimization) whose outputs are not definitionally equivalent to its inputs. No equations, fitted parameters, or predictions reduce to the input data by construction; the method is compared to naive baselines and assessed via external animator critique rather than internal self-reference. No self-citations serve as load-bearing uniqueness theorems, and no ansatz or renaming patterns appear. The derivation chain therefore remains independent of the target result.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption of topology-invariant input meshes and the practical assumption that user-guided segmentation plus boundary conditioning suffices for seamless interchangeability; the number of replacements per part is a user- or budget-controlled parameter.

free parameters (1)
  • number of replacements per part
    User-defined or automatically computed to meet printing budget or allowed deviation from original animation.
axioms (2)
  • domain assumption Input animation consists of topology-invariant deforming meshes
    Stated directly in the problem formulation in the abstract.
  • domain assumption User can manually indicate preferred parts for segmentation
    Described as inspired by current stop-motion workflows.

pith-pipeline@v0.9.0 · 5847 in / 1393 out tokens · 25144 ms · 2026-05-24T16:44:14.180869+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    Representing Animations by Principal Compo- nents. Comput. Graph. Forum (2000). Jed Alger, Travis Knight, and Chris Butler

    work page 2000

  2. [2]

    ACM Trans

    Video Puppetry: A Performative Interface for Cutout Animation. ACM Trans. Graph. (2008). Miklós Bergou, Saurabh Mathur, Max Wardetzky, and Eitan Grinspun

    work page 2008

  3. [3]

    ACM Trans

    TRACKS: Toward Directable Thin Shells. ACM Trans. Graph. (2007). Bernd Bickel, Peter Kaufmann, Mélina Skouras, Bernhard Thomaszewski, Derek Bradley, Thabo Beeler, Phil Jackson, Steve Marschner, Wojciech Matusik, and Markus Gross

    work page 2007

  4. [4]

    ACM Transactions on Graphics (TOG) 31, 4 (2012),

    Physical face cloning. ACM Transactions on Graphics (TOG) 31, 4 (2012),

    work page 2012

  5. [5]

    Tangible and Embedded Interaction (2009)

    Stop-motion prototyping for tangible interfaces. Tangible and Embedded Interaction (2009). Yuri Boykov and Vladimir Kolmogorov

    work page 2009

  6. [6]

    IEEE transactions on pattern analysis and machine intelligence 26, 9 (2004), 1124–1137

    An experimental comparison of min- cut/max-flow algorithms for energy minimization in vision. IEEE transactions on pattern analysis and machine intelligence 26, 9 (2004), 1124–1137. Gabriel J. Brostow and Irfan Essa

    work page 2004

  7. [7]

    Perceptually-motivated Real-time Temporal Upsampling of 3D Content for High-refresh-rate Displays. Comput. Graph. Forum (2010). Marek Dvorožnák, Wilmot Li, Vladimir G Kim, and Daniel S`ykora

    work page 2010

  8. [8]

    ACM Transactions on Graphics (TOG) 37, 4 (2018),

    Toonsynth: example-based synthesis of hand-colored cartoon animations. ACM Transactions on Graphics (TOG) 37, 4 (2018),

    work page 2018

  9. [9]

    ACM Trans

    JALI: an animator- centric viseme model for expressive lip synchronization. ACM Trans. Graph. (2016). Michael Elad and Michal Aharon

    work page 2016

  10. [10]

    IEEE Transactions on Image processing 15, 12 (2006), 3736–3745

    Image denoising via sparse and redundant representations over learned dictionaries. IEEE Transactions on Image processing 15, 12 (2006), 3736–3745. Jakub Fišer, Ondřej Jamriška, David Simons, Eli Shechtman, Jingwan Lu, Paul Asente, Michal Lukáč, and Daniel S`ykora

    work page 2006

  11. [11]

    ACM Transactions on Graphics (TOG) 36, 4 (2017),

    Example-based synthesis of stylized facial animations. ACM Transactions on Graphics (TOG) 36, 4 (2017),

    work page 2017

  12. [12]

    NIPS (2012)

    From Deformations to Parts - Motion-based Segmentation of 3D Objects. NIPS (2012). Qin Gu, Jingliang Peng, and Zhigang Deng

    work page 2012

  13. [13]

    Computer Graphics Forum (2009)

    Compression of Human Motion Capture Data Using Motion Pattern Indexing. Computer Graphics Forum (2009). Gaël Guennebaud, Benoît Jacob, et al

    work page 2009

  14. [14]

    Eigen v3. (2010). http://eigen.tuxfamily.org. Xiaoguang Han, Hongbo Fu, Hanlin Zheng, Ligang Liu, and Jue Wang

    work page 2010

  15. [15]

    IEEE CG&A (2014)

    A Video- Based System for Hand-Driven Stop-Motion Animation. IEEE CG&A (2014). Philipp Herholz, Wojciech Matusik, and Marc Alexa

    work page 2014

  16. [16]

    Approximating Free-form Geometry with Height Fields for Manufacturing. Comput. Graph. Forum 34, 2 (May 2015), 239–251. https://doi.org/10.1111/cgf.12556 Ruizhen Hu, Honghua Li, Hao Zhang, and Daniel Cohen-Or

    work page doi:10.1111/cgf.12556 2015

  17. [17]

    ACM Trans

    Approximate Pyramidal Shape Decomposition. ACM Trans. Graph. 33, 6, Article 213 (Nov. 2014), 12 pages. https://doi.org/10.1145/2661229.2661244 Alec Jacobson, Daniele Panozzo, et al

    work page doi:10.1145/2661229.2661244 2014

  18. [18]

    libigl: A simple C++ geometry processing library. (2017). http://libigl.github.io/libigl/. Alec Jacobson, Elif Tosun, Olga Sorkine, and Denis Zorin

    work page 2017

  19. [19]

    ACM Trans

    Skinning Mesh Animations. ACM Trans. Graph. 24, 3 (July 2005), 399–407. Se-Won Jeong and Jae-Young Sim

    work page 2005

  20. [20]

    APSIPA (2014)

    Multiscale saliency detection for 3D meshes using random walk. APSIPA (2014). Brian Knep, Craig Hayes, Rick Sayre, and Tom Williams

    work page 2014

  21. [21]

    Vladimir Kolmogorov and Ramin Zabin

    Objet Geometries’ 3-D Printers Play Starring Role in New Animated Film Coraline.http://www.prnewswire.co.uk/news-releases/objet-geometries- 3-d-printers-play-starring-role-in-new-animated-film-coraline-155479455.html (2009). Vladimir Kolmogorov and Ramin Zabin

    work page 2009

  22. [22]

    Jan Eric Kyprianidis, John Collomosse, Tinghuai Wang, and Tobias Isenberg

    What energy functions can be minimized via graph cuts? IEEE TPAMI (2004). Jan Eric Kyprianidis, John Collomosse, Tinghuai Wang, and Tobias Isenberg

    work page 2004

  23. [23]

    IEEE Transactions on Visualization and Computer Graphics 19, 5 (May 2013), 866–885

    State of the "Art”: A Taxonomy of Artistic Stylization Techniques for Images and Video. IEEE Transactions on Visualization and Computer Graphics 19, 5 (May 2013), 866–885. https://doi.org/10.1109/TVCG.2012.160 Binh Huy Le and Zhigang Deng

    work page doi:10.1109/tvcg.2012.160 2013

  24. [24]

    ACM Transactions on Graphics (2013)

    Two-layer sparse compression of dense-weight blend skinning. ACM Transactions on Graphics (2013). Binh Huy Le and Zhigang Deng

    work page 2013

  25. [25]

    ACM Trans

    Robust and accurate skeletal rigging from mesh sequences. ACM Trans. Graph. (2014). Tong-Yee Lee, Yu-Shuen Wang, and Tai-Guang Chen

    work page 2014

  26. [26]

    The Visual Computer 22, 9-11 (2006), 729–739

    Segmenting a deforming mesh into near-rigid components. The Visual Computer 22, 9-11 (2006), 729–739. John P Lewis, Ken Anjyo, Taehyun Rhee, Mengjie Zhang, Frédéric H Pighin, and Zhigang Deng

    work page 2006

  27. [27]

    Eurographics (2014)

    Practice and Theory of Blendshape Facial Models. Eurographics (2014). Stuart P. Lloyd

    work page 2014

  28. [28]

    IEEE Transactions on Informa- tion Theory 28 (1982), 129–137

    Least squares quantization in pcm. IEEE Transactions on Informa- tion Theory 28 (1982), 129–137. Linjie Luo, Ilya Baran, Szymon Rusinkiewicz, and Wojciech Matusik

    work page 1982

  29. [29]

    ACM Trans

    Chopper: Partitioning Models into 3D-printable Parts. ACM Trans. Graph. (2012). Mekado Murphy

    work page 2012

  30. [30]

    https://www.nytimes.com/interactive/2015/12/18/movies/anomalisa-behind-the- scenes.html (2015)

    Showing the Seams in ‘Anomalisa’. https://www.nytimes.com/interactive/2015/12/18/movies/anomalisa-behind-the- scenes.html (2015). Thomas Neumann, Kiran Varanasi, Stephan Wenger, Markus Wacker, Marcus A Magnor, and Christian Theobalt

    work page 2015

  31. [31]

    Sparse localized deformation components.ACM Trans. Graph. (2013). Seung-Tak Noh and Takeo Igarashi

    work page 2013

  32. [32]

    Facial Retargeting with Automatic Range of Motion Alignment.ACM Trans. Graph. (2017). 2019-07-25 00:40 page 10 (pp. 1-11) A system for efficient 3D printed stop-motion face animation • 39:11 0 20 40 λ = 1 100 200 300 400 500 600 700 0 20 40 λ = 5 Naive uniform sampling 0 50 100 150 200 Velocity Energy Position Energy Number of Frames 0 20 40 λ = 0 Our opt...

    work page 2017

  33. [33]

    ACM Trans

    Compression and direct manipulation of complex blendshape models. ACM Trans. Graph. (2011). Ariel Shamir

    work page 2011

  34. [34]

    A survey on Mesh Segmentation Techniques. Comput. Graph. Forum (2008). Karan Singh and Eugene Fiume

    work page 2008

  35. [35]

    ACM Transactions on Graphics (TOG) 31, 6 (2012),

    Recursive interlocking puzzles. ACM Transactions on Graphics (TOG) 31, 6 (2012),

    work page 2012

  36. [36]

    Computer Graphics Forum 33, 2 (2014), 145–154

    Compressing dynamic meshes with geometric laplacians. Computer Graphics Forum 33, 2 (2014), 145–154. Jue Wang, Yingqing Xu, Heung-Yeung Shum, and Michael F. Cohen

    work page 2014

  37. [37]

    InProceedings of the 37th Annual Conference of the European Association for Computer Graphics (EG ’16)

    Improved Surface Quality in 3D Printing by Optimizing the Printing Direction. InProceedings of the 37th Annual Conference of the European Association for Computer Graphics (EG ’16) . Eurographics Association, Goslar Germany, Germany, 59–70. https://doi.org/10.1111/cgf.12811 Brian Whited, Gioacchino Noris, Maryann Simmons, Robert W Sumner, Markus H Gross, ...

    work page doi:10.1111/cgf.12811

  38. [38]

    BetweenIT - An Interactive Tool for Tight Inbe- tweening. Comput. Graph. Forum (2010). Andrew P Witkin and Michael Kass

    work page 2010

  39. [39]

    Siggraph (1988)

    Spacetime constraints. Siggraph (1988). John Wright, Yi Ma, Julien Mairal, Guillermo Sapiro, Thomas S Huang, and Shuicheng Yan

    work page 1988

  40. [40]

    Sparse representation for computer vision and pattern recognition. Proc. IEEE 98, 6 (2010), 1031–1044. Y Y. Boykov, O Veksler R., and R Zabih

    work page 2010

  41. [41]

    20 (01 2001)

    Efficient Approximate Energy Minimiza- tion via Graph Cuts. 20 (01 2001). Xiaoting Zhang, Xinyi Le, Athina Panotopoulou, Emily Whiting, and Charlie C. L. Wang

    work page 2001

  42. [42]

    ACM Trans

    Perceptual Models of Preference in 3D Printing Direction. ACM Trans. Graph. (2015). 2019-07-25 00:40 page 11 (pp. 1-11)

    work page 2015