Pith. sign in

REVIEW 3 major objections 5 minor 49 references

The paper claims that topological postprocessing of synthetic particle images can be performed on a compact graph representation instead of the full 3D voxel image, and that this consistently improves the resulting image topology and geomet

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 17:17 UTC pith:UOGTP4HG

load-bearing objection Graph representation makes topological postprocessing feasible for large particle images, but the key graph-to-image topology transfer is unproven and the main metric is the optimized loss; still worth reviewing. the 3 major comments →

arxiv 2607.17680 v1 pith:UOGTP4HG submitted 2026-07-20 stat.ME cond-mat.mtrl-sci

Fitting the topology of synthetic particle systems with a novel graph representation

Martin Alexander Memmesheimer , Claudia Redenbach This is my paper
classification stat.ME cond-mat.mtrl-sci MSC 55N31
keywords particle systemspersistent homologytopological optimizationgraph representationrandom sequential adsorptionWasserstein distance3D image reconstructiongeometric tortuosity
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to show that the topology of synthetic 3D particle microstructures can be fitted to real data by optimizing a small graph rather than the full voxel volume. Each particle becomes a graph node; edges carry the endpoints and L1 lengths of the shortest unobstructed line-of-sight connections between particles. Persistent homology is computed on this graph and aligned to a target via a differentiable entropic-regularized Wasserstein loss, after which the optimized graph is converted back to an image by taking convex hulls of the edge points. If the claim holds, any existing particle-generation method gets a cheap, generation-agnostic postprocessing step that fixes connectivity, tunnels, and cavities without disturbing particle sizes or counts. The paper reports consistent improvements in all three homology dimensions and large reductions in geometric tortuosity across four concrete images.

Core claim

The central claim is that the persistence diagrams of a binarized particle image are well approximated by the Vietoris-Rips persistence of a thresholded graph whose nodes are particles and whose edge weights are the L1 distances between the endpoints of the shortest unobstructed line segment connecting two particles. Because the graph has only a few hundred nodes instead of billions of voxels, gradient-based topological optimization becomes feasible. Applying this optimization to RSA-generated concrete images and reconstructing via convex hulls yields image Wasserstein distances to real data that improve over both the raw RSA images and unoptimized reconstructions in every homology dimension

What carries the argument

The load-bearing object is the particle graph G with node features storing particle centers of mass and edge features storing the endpoints of the shortest admissible L1 line segment between particles. The edge weight w(e) = ||u−v||_1 turns the graph into a metric space whose Vietoris-Rips complex yields persistence diagrams. The loss is the entropic-regularized 2-Wasserstein distance between synthetic and target persistence diagrams; the regularization stabilizes the matching gradients. Reconstruction takes the convex hull of all edge endpoints and centers of mass for each particle, resolves overlaps by label order, then applies morphological closing and optional simplex noise to recover bo

Load-bearing premise

The load-bearing premise is that the Vietoris-Rips persistence of the thresholded graph with L1 edge weights faithfully represents the cubical persistence of the distance transform of the binarized image; the paper justifies this with a heuristic that short edges capture topologically relevant information, rather than a proof.

What would settle it

Construct a particle arrangement where an image loop is formed by a connection longer than the threshold τ; the thresholded graph would miss this edge, and optimizing the graph could not change that loop. A concrete experiment: deliberately create a synthetic image with a long-range arch-like connection, optimize the graph with τ=400, reconstruct, and compare the image persistence diagrams to the target; a counterexample showing no transfer of improvement would falsify the central claim.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the graph optimization truly controls image topology, then any particle generation method (RSA, tessellation, or other) can be postprocessed to match a target topology without changing particle identity or count.
  • Optimization on the graph consistently improves image topology across all three homology dimensions, with H1 improving 16–18% over RSA images and H2 improving approximately 13–20%.
  • The improvements are driven mainly by the 1-dimensional topology, and including all three dimensions yields the best geometric tortuosity, cutting the gap to real images by up to 99%.
  • The method preserves particle volume and surface area distributions because particles remain explicit and cannot merge or be newly created during optimization.

Where Pith is reading between the lines

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

  • A direct test of the transfer assumption would be to compare cubical persistence diagrams of reconstructed images against graph persistence diagrams across a wider range of particle systems; the paper validates on four concrete images, so the transfer claim is not yet generally established.
  • The convex-hull reconstruction systematically fills concavities, which likely explains the slight shift in particle volume distribution; replacing hulls with concave-capable models (e.g., alpha shapes with denser boundary sampling) could recover concavities while preserving the graph-based optimization.
  • Because edge features are local boundary contact points, the same graph representation could support optimizing other transport-relevant geometric properties—such as percolation paths or elastic response—by adding extra loss terms on the same graph.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a postprocessing method for synthetic 3D particle images that aims to improve topological agreement with real micro-CT data. A labeled image is converted into a particle graph whose nodes are particles and whose edges are shortest unobstructed line segments between particles, with edge features storing the segment endpoints. The image is later reconstructed from this graph by taking convex hulls of per-particle feature points. Persistent homology is computed on the graph using a Vietoris–Rips filtration with edge weights given by the L1 length of the edge feature (Eq. 12), and an entropic-regularized 2-Wasserstein loss (Eq. 15) is minimized to match the synthetic graph persistence diagrams to those of a real graph. After optimization, images are reconstructed and evaluated by cubical persistence of the distance transform of the image (Eq. 9), Dice overlap, particle volume/surface distributions, and geometric tortuosity. Experiments on four recycled-concrete µCT images with 16 RSA-generated synthetic copies each report consistent improvements in H0/H1/H2 Wasserstein distances and tortuosity relative to the RSA images and to reconstructions of the unoptimized graphs, while approximately preserving particle count and size distributions.

Significance. If the graph-to-image topology transfer is valid, this is a practically valuable contribution: it offers a computationally tractable way to apply persistence-based topology optimization to large 3D particle microstructures, and it is presented as independent of the image generation method. The paper includes a sensible unoptimized-reconstruction baseline, evaluates on multiple real images, and explicitly acknowledges the convex-hull limitation in the Discussion. The main significance hinges on a load-bearing proxy assumption—that VR persistence of the thresholded particle graph faithfully reflects cubical persistence of the reconstructed image—which is asserted rather than demonstrated. The statistical evidence in Tables 4 and 5 also lacks error bars, so the consistency claims are not yet fully supported.

major comments (3)
  1. [Persistent Homology / Topological Loss (Eqs. 12–15) vs. Validation (Eq. 9)] The central claim requires that the graph-domain persistence diagrams D_G^(s) optimized in Eq. (15) are a faithful proxy for the image-domain persistence diagrams of the reconstructed image computed via Eq. (9). The paper asserts this via the 'short edges' heuristic and chooses τ=400 to reproduce volume fraction, not topological fidelity. This is load-bearing: the reported image improvements could arise from the convex-hull reconstruction (Eq. 4) rather than from aligning true image topology. Please (i) report the graph-domain loss before/after optimization to confirm the optimization actually reduces it; (ii) provide a direct comparison of graph persistence distances vs. image persistence distances across settings and τ values; and (iii) include a sensitivity analysis of Table 4 with respect to τ. Without this, the main conclusion is not established.
  2. [Tables 4 and 5] All entries are means over 16 synthetic runs, but no standard deviations, confidence intervals, or paired tests are reported. The claimed 'consistent improvement' is small in several cases (e.g., H2 improvements of 7–17% over the unoptimized reconstruction; tortuosity reductions in Table 5). Without run-level variability, the reader cannot judge whether the differences are statistically meaningful or within optimization noise. Report standard deviations/confidence intervals and, ideally, paired tests or per-run scatter plots for the main comparisons.
  3. [Reconstruction and confounds (Eq. 4, Tables 3 and 4)] The unoptimized reconstruction already improves H0 and H2 relative to the RSA images (Table 4, Reconstruction row), and the convex-hull reconstruction is explicitly acknowledged in the Discussion to fill concavities. Because optimization changes edge features, which also changes reconstructed particle overlaps and volume fraction (Table 3), the image-topology gains may be an indirect effect of densification or convexification rather than of alignment of the graph topology. Please add a control that perturbs edge features to match the optimized volume fraction without a topological loss, or otherwise condition the comparison on matched volume fraction and particle overlap.
minor comments (5)
  1. [Introduction] 'persistence homology' appears in the Introduction; should be 'persistent homology'.
  2. [Table 4 caption] The caption says '1-Wasserstein Distances' but the text and equations use 2-Wasserstein distances; correct the caption.
  3. [Abstract and Introduction] The method is described as 'independent of the underlying generation method', but experiments only use RSA. Either test at least one additional generation method or soften the claim to 'applicable to a range of generation methods'.
  4. [Validation section] The statement that 0-dimensional persistence diagrams of the real images are 'empty' follows from discarding essential features. State this explicitly at the point of use, since it is conceptually important for interpreting the H0 column.
  5. [Table 2] Average particle counts such as 513.438 and 503.875 are means of integer counts; rounding or adding a decimal convention would improve readability.

Circularity Check

0 steps flagged

No circularity: graph-domain loss and image-domain evaluation are distinct; the unproven proxy assumption is a correctness risk, not a circular reduction.

full rationale

The derivation is not circular. The optimization loss (Eq. 15) minimizes an entropic-regularized 2-Wasserstein distance between persistence diagrams D_G^(s) of the thresholded particle graph (computed via Vietoris-Rips on edge weights w(e)=||u-v||_1, Eq. 12) and target graph diagrams, while the reported headline metrics H0/H1/H2 (Table 4) are 2-Wasserstein distances between cubical persistence diagrams of the binarized/reconstructed images and the real image (Eqs. 9-10). These are different filtrations on different objects: the particle graph versus the distance transform of the reconstructed voxel image. The paper contains no equation identifying the graph and image persistence diagrams. The reconstruction step (Eq. 4) introduces a genuine domain gap that, as the paper notes, can worsen H1 before optimization, so the improvements are not forced by the loss. The paper also uses no author-overlapping citations as load-bearing support; the external references [19, 46, 47] provide the differentiable optimal-transport machinery, and GeomLoss implements it. The main risk flagged in the text—that graph VR persistence may not faithfully track cubical image persistence, since the paper only 'expects' the topologically relevant information to be captured by short edges and chooses tau=400 by volume fraction—is an unproven modeling assumption, not a circular step. Independent metrics (Dice, tortuosity, volume/surface distributions) provide non-tautological corroboration.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The method relies mainly on standard TDA and optimal-transport machinery. The nonstandard, unproved assumptions are the graph-to-image topology transfer, the thresholding heuristic, and convex-hull reconstruction. No new physical entities are introduced.

free parameters (6)
  • Edge-length threshold tau = 400 voxels (full resolution; 100 on 4x downscaled images)
    Only graph edges with L1 length <= tau are kept. The authors state tau=400 is used 'as it most accurately reproduces the volume fraction of the particle systems', i.e. tuned on the test data; robustness for tau in [200,600] is reported.
  • Persistence noise filter = persistence <= 2 removed
    For evaluation H1/H2 metrics, features with d-b<=2 are deleted as noise; the cutoff is chosen by hand and affects the measured Wasserstein distances.
  • Entropic regularization gamma = 100
    Controls bias and stability of the Sinkhorn approximation to the 2-Wasserstein loss; set by the authors without a sensitivity study.
  • Optimization schedule = lr=0.5, decay=0.99, epochs=500
    Adam hyperparameters chosen by hand; they affect how far the topological loss decreases.
  • Morphological closing radius = 1 voxel
    Applied to reconstructed images and to RSA images before evaluation; closing changes connectivity and volume fraction.
  • Downsampling factor = 4 per dimension
    Particle graphs computed on downscaled images then features scaled by 4; introduces resolution-dependent approximation.
axioms (6)
  • domain assumption Topological features of the image are captured by short graph edges
    Section 'Graph Representation': 'the topologically relevant information is expected to be captured primarily by short edges' justifies thresholding E to E_tau; if false, the graph filtration omits features of the image.
  • domain assumption VR persistence on the weighted particle graph approximates cubical persistence on the distance-transformed image
    The entire optimization is performed in the graph domain, while the claim is about image topology; no theorem relates H_s(Rips_tau(G)) to cubical H_s(image distance transform).
  • domain assumption Convex-hull reconstruction preserves the topology and shape of particles sufficiently
    Reconstruction via conv(X_v) fills concavities and can merge overlapping hulls; authors acknowledge this limitation and use it to explain volume changes, but central validation assumes the transfer is not topology-breaking.
  • domain assumption Essential persistence points may be discarded
    Stated in 'Persistent Homology' section: 'their proportion in our data is well below one percent'; assumption that loss of essential points is negligible.
  • domain assumption Binarization plus distance transform yields a meaningful filtration of the particle system
    Standard TDA practice used in evaluation; particle-specific labels are not used in image persistence.
  • standard math Entropic regularization of Wasserstein approximates exact Wasserstein with controllable error
    Uses Lacombe et al. [46] and Cuturi [47]; not proved in paper but accepted background.

pith-pipeline@v1.3.0-alltime-deepseek · 16737 in / 16574 out tokens · 173522 ms · 2026-08-01T17:17:37.879053+00:00 · methodology

0 comments
read the original abstract

The shape and arrangement of particles in a material determine its macroscopic properties. The generation of synthetic data with varying particle structure, often represented as 3D voxel images, combined with simulation of macroscopic properties reveals structure-property relations. Most particle generation models focus on single-particle characteristics like shape and size. We aim at fitting the topology of the particle system using tools from persistent homology. However, the large size of the required 3D image data makes existing methods computationally infeasible. We bridge this gap by introducing a novel graph representation of particle systems and transferring the computation of persistent homology from the image domain to the graph domain. This yields a postprocessing method for synthetic images of particle systems, that is independent of the underlying generation method and improves topological and geometrical agreement with real particle systems while preserving morphological characteristics such as the particle size distribution.

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

49 extracted references · 2 canonical work pages

  1. [1]

    & Daouadji, A

    Radjai, F., Roux, J.-N. & Daouadji, A. Modeling granular materials: Century-long research across scales.Journal of Engineering Mechanics143, 04017002 (2017). URL https://ascelibrary.org/doi/abs/10.1061/(ASCE)EM.1943-7889.0001196

    arXiv 2017

  2. [2]

    URL https://www.sciencedirect.com/science/article/pii/S0008884606001013

    Erdogan, S.et al.Three-dimensional shape analysis of coarse aggregates: New techniques for and preliminary results on several different coarse aggregates and reference rocks.Cement and Concrete Research36, 1619–1627 (2006). URL https://www.sciencedirect.com/science/article/pii/S0008884606001013

    2006

  3. [3]

    & Lee, P

    Fonseca, J., O’Sullivan, C., Coop, M. & Lee, P. Non-invasive characterization of particle morphology of natural sands.Soils and Foundations52, 712–722 (2012). URL https://www.sciencedirect.com/science/article/pii/S0038080612000819

    2012

  4. [4]

    Garboczi, E. Three-dimensional mathematical analysis of particle shape using x-ray tomography and spherical harmonics: Application to aggregates used in concrete.Cement and Concrete Research32, 1621–1638 (2002). URL https: //www.sciencedirect.com/science/article/pii/S0008884602008360

    2002

  5. [5]

    & Chang, X

    Liu, J., Zhou, W., Ma, G., Yang, S. & Chang, X. Strong contacts, connectivity and fabric anisotropy in granular materials: A 3d perspective.Powder Technology 366, 747–760 (2020). URL https://www.sciencedirect.com/science/article/pii/ S0032591020302096

    2020

  6. [6]

    A., Daniels, K

    Papadopoulos, L., Porter, M. A., Daniels, K. E. & Bassett, D. S. Network analysis of particles and grains.Journal of Complex Networks6, 485–565 (2018). URL https://doi.org/10.1093/comnet/cny005

    doi:10.1093/comnet/cny005 2018

  7. [7]

    Walker, D. M. & Tordesillas, A. Topological evolution in dense granular materials: A complex networks perspective.International Journal of Solids and Struc- tures47, 624–639 (2010). URL https://www.sciencedirect.com/science/article/ pii/S0020768309004272

    2010

  8. [8]

    URL https://www.sciencedirect.com/science/article/pii/S0010938X26000405

    Zaninovi´ c, J.et al.Leaching in active protective coatings observed in-situ by nano-CT using synchrotron radiation.Corrosion Science263, 113631 (2026). URL https://www.sciencedirect.com/science/article/pii/S0010938X26000405

    2026

  9. [9]

    N., Stoyan, D., Kendall, W

    Chiu, S. N., Stoyan, D., Kendall, W. S. & Mecke, J.Stochastic geometry and its applications(John Wiley & Sons, 2013)

    2013

  10. [10]

    & Yun, G

    You, H., Kim, Y. & Yun, G. J. Computationally fast morphological descriptor- based microstructure reconstruction algorithms for particulate composites.Com- posites Science and Technology182, 107746 (2019). URL https://www. sciencedirect.com/science/article/pii/S0266353819302076. 23

    2019

  11. [11]

    URL https://www.sciencedirect.com/science/ article/pii/S0927025615003845

    Feinauer, J.et al.Stochastic 3d modeling of the microstructure of lithium-ion battery anodes via gaussian random fields on the sphere.Computational Materi- als Science109, 137–146 (2015). URL https://www.sciencedirect.com/science/ article/pii/S0927025615003845

    2015

  12. [12]

    M., L¨ osel, P

    Wang, B., Kingston, A. M., L¨ osel, P. D. & Creemers, W. Synthetic particle pack generation for augmentation and testing in geological tomographic segmentation. Tomography of Materials and Structures9, 100072 (2025). URL https://www. sciencedirect.com/science/article/pii/S2949673X25000257

    2025

  13. [13]

    & Chen, C.Topodiffusionnet: A topology-aware dif- fusion model

    Gupta, S., Samaras, D. & Chen, C.Topodiffusionnet: A topology-aware dif- fusion model. (eds Yue, Y., Garg, A., Peng, N., Sha, F. & Yu, R.) International Conference on Learning Representations, Vol. 2025, 31699– 31713 (2025). URL https://proceedings.iclr.cc/paper files/paper/2025/file/ 4e889e581d4a6f2be932c6f65e7792a8-Paper-Conference.pdf

    2025

  14. [14]

    Topolidm: Topology-aware lidar diffusion models for interpretable and realistic lidar point cloud generation

    Liu, J.et al. Topolidm: Topology-aware lidar diffusion models for interpretable and realistic lidar point cloud generation. (eds Laugier, C.et al.)2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 8180–8186 (IEEE, 2025)

    2025

  15. [15]

    & Chen, C.Topogan: A topology-aware generative adversarial network

    Wang, F., Liu, H., Samaras, D. & Chen, C.Topogan: A topology-aware generative adversarial network. (eds Vedaldi, A., Bischof, H., Brox, T. & Frahm, J.-M.)Com- puter Vision – ECCV 2020, 118–136 (Springer International Publishing, Cham, 2020)

    2020

  16. [16]

    & Ovsjanikov, M

    Poulenard, A., Skraba, P. & Ovsjanikov, M. Topological function optimization for continuous shape matching.Computer Graphics Forum37, 13–25 (2018). URL https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.13487

    work page doi:10.1111/cgf.13487 2018

  17. [17]

    & Guibas, L

    Br¨ uel-Gabrielsson, R., Ganapathi-Subramanian, V., Skraba, P. & Guibas, L. J. Topology-aware surface reconstruction for point clouds.Computer Graphics Forum39, 197–207 (2020). URL https://onlinelibrary.wiley.com/doi/abs/10. 1111/cgf.14079

    2020

  18. [18]

    Kissi, M., Pont, M., Levine, J. A. & Tierny, J. A practical solver for scalar data topological simplification.IEEE Transactions on Visualization and Computer Graphics31, 97–107 (2025)

    2025

  19. [19]

    Optimizing persistent homology based functions

    Carriere, M.et al. Optimizing persistent homology based functions. (eds Meila, M. & Zhang, T.)Proceedings of the 38th International Conference on Machine Learn- ing, Vol. 139 ofProceedings of Machine Learning Research, 1294–1303 (PMLR, 2021). URL https://proceedings.mlr.press/v139/carriere21a.html

    2021

  20. [20]

    Bresenham, J. E. Algorithm for computer control of a digital plotter.IBM Systems Journal4, 25–30 (1965). 24

    1965

  21. [21]

    & Schladitz, K.3D Images of Materials Structures(John Wiley & Sons, Ltd, 2009)

    Ohser, J. & Schladitz, K.3D Images of Materials Structures(John Wiley & Sons, Ltd, 2009)

    2009

  22. [22]

    Simplex noise demystified (2005)

    Gustavson, S. Simplex noise demystified (2005)

    2005

  23. [23]

    Noise hardware

    Perlin, K. Noise hardware. in real-time shading.SIGGRAPH Course Notes (2001)

    2001

  24. [24]

    Topological persistence and simplification

    Edelsbrunner, Letscher & Zomorodian. Topological persistence and simplification. Discrete & computational geometry28, 511–533 (2002)

    2002

  25. [25]

    & Carlsson, G

    Zomorodian, A. & Carlsson, G. Computing persistent homology.Discrete & Computational Geometry33, 249–274 (2005). URL https://doi.org/10.1007/ s00454-004-1146-y

    2005

  26. [26]

    Simplizialzerlegungen von beschr¨ ankter Flachheit.Annals of Mathematics43, 580–582 (1942)

    Freudenthal, H. Simplizialzerlegungen von beschr¨ ankter Flachheit.Annals of Mathematics43, 580–582 (1942). URL http://www.jstor.org/stable/1968813

    arXiv 1942

  27. [27]

    & Harer, J.Computational topology: an introduction(American Mathematical Soc., 2010)

    Edelsbrunner, H. & Harer, J.Computational topology: an introduction(American Mathematical Soc., 2010)

    2010

  28. [28]

    & Mrozek, M.Cubical Homology, 39–92 (Springer New York, New York, NY, 2004)

    Kaczynski, T., Mischaikow, K. & Mrozek, M.Cubical Homology, 39–92 (Springer New York, New York, NY, 2004). URL https://doi.org/10.1007/0-387-21597-2 2

    work page doi:10.1007/0-387-21597-2 2004

  29. [29]

    Mohapatra, R. K.et al.A comprehensive survey to study the utilities of image segmentation methods in clinical routine.Network Modeling Analysis in Health Informatics and Bioinformatics13, 2 (2023). URL https://doi.org/10.1007/ s13721-023-00436-z

    2023

  30. [30]

    & Nishikawa, N

    Carriere, M., Ike, Y., Lacombe, T. & Nishikawa, N. Persistence-based topological optimization: a survey.arXiv preprint arXiv:2603.24613(2026)

    arXiv 2026

  31. [31]

    Kingma, D. P. & Ba, J.Adam: A Method for Stochastic Optimization. (eds Ben- gio, Y. & LeCun, Y.)3rd International Conference on Learning Representations, Conference Track Proceedings (2015). URL https://mlanthology.org/iclr/2015/ kingma2015iclr-adam/

    2015

  32. [32]

    URL https://doi.org/10.1007/ 978-3-662-05088-0

    Soille, P.Morphological Image Analysis: Principles and Applications(Springer Berlin Heidelberg, Berlin, Heidelberg, 2004). URL https://doi.org/10.1007/ 978-3-662-05088-0

    2004

  33. [33]

    & Morozov, D.Vines and vineyards by updating persistence in linear time

    Cohen-Steiner, D., Edelsbrunner, H. & Morozov, D.Vines and vineyards by updating persistence in linear time. (eds Amenta, N. & Cheong, O.)Proceed- ings of the Twenty-Second Annual Symposium on Computational Geometry, SCG ’06, 119–126 (Association for Computing Machinery, New York, NY, USA, 2006). URL https://doi.org/10.1145/1137856.1137877. 25

    arXiv 2006

  34. [34]

    & Nelson, B

    Luo, Y. & Nelson, B. J. Accelerating iterated persistent homology computations with warm starts.Computational Geometry120, 102089 (2024). URL https: //www.sciencedirect.com/science/article/pii/S0925772124000117

    2024

  35. [35]

    & M¨ ucke, E

    Edelsbrunner, H. & M¨ ucke, E. P. Three-dimensional alpha shapes.ACM Trans. Graph.13, 43–72 (1994). URL https://doi.org/10.1145/174462.156635

    arXiv 1994

  36. [36]

    A., Davis, L

    Feldkamp, L. A., Davis, L. C. & Kress, J. W. Practical cone-beam algorithm.J. Opt. Soc. Am. A1, 612–619 (1984). URL https://opg.optica.org/josaa/abstract. cfm?URI=josaa-1-6-612

    1984

  37. [37]

    A threshold selection method from gray-level histograms.IEEE Transactions on Systems, Man, and Cybernetics9, 62–66 (1979)

    Otsu, N. A threshold selection method from gray-level histograms.IEEE Transactions on Systems, Man, and Cybernetics9, 62–66 (1979)

    1979

  38. [38]

    & Soille, P

    Vincent, L. & Soille, P. Watersheds in digital spaces: an efficient algorithm based on immersion simulations.IEEE Transactions on Pattern Analysis and Machine Intelligence13, 583–598 (1991)

    1991

  39. [39]

    & Breit, W

    Burgmann, S., Godehardt, M., Schladitz, K. & Breit, W. Separation of sand and gravel particles in 3d images using the adaptive h-extrema transform.Powder Technology404, 117468 (2022). URL https://www.sciencedirect.com/science/ article/pii/S003259102200362X

    2022

  40. [40]

    URL https://www.sciencedirect.com/science/article/ pii/S030094402400554X

    Schladitz, K.et al.Geometric modelling of corrosion inhibitor pigments in active protective coatings based on SR-nano-CT images.Progress in Organic Coat- ings197, 108762 (2024). URL https://www.sciencedirect.com/science/article/ pii/S030094402400554X

    2024

  41. [41]

    URL https://doi.org/10.1038/s41524-021-00567-9

    Furat, O.et al.Artificial generation of representative single li-ion electrode par- ticle architectures from microscopy data.npj Computational Materials7, 105 (2021). URL https://doi.org/10.1038/s41524-021-00567-9

    doi:10.1038/s41524-021-00567-9 2021

  42. [42]

    & Lacombe, T

    Divol, V. & Lacombe, T. Understanding the topology and the geometry of the space of persistence diagrams via optimal partial transport.Journal of Applied and Computational Topology5, 1–53 (2021). URL https://doi.org/10.1007/ s41468-020-00061-z

    2021

  43. [43]

    & Vialard, F.-X

    Chizat, L., Peyr´ e, G., Schmitzer, B. & Vialard, F.-X. Unbalanced optimal trans- port: Dynamic and kantorovich formulations.Journal of Functional Analysis 274, 3090–3123 (2018). URL https://www.sciencedirect.com/science/article/pii/ S0022123618301058

    2018

  44. [44]

    & Gigli, N

    Figalli, A. & Gigli, N. A new transportation distance between non-negative measures, with applications to gradients flows with dirichlet boundary conditions. Journal de Math´ ematiques Pures et Appliqu´ ees94, 107–130 (2010). URL https: //www.sciencedirect.com/science/article/pii/S0021782409001627. 26

    2010

  45. [45]

    & Vorotnikov, D

    Kondratyev, S., Monsaingeon, L. & Vorotnikov, D. A new optimal transport dis- tance on the space of finite Radon measures.Advances in Differential Equations 21, 1117 – 1164 (2016). URL https://doi.org/10.57262/ade/1476369298

    arXiv 2016

  46. [46]

    & Oudot, S.Large scale computation of means and clusters for persistence diagrams using optimal transport

    Lacombe, T., Cuturi, M. & Oudot, S.Large scale computation of means and clusters for persistence diagrams using optimal transport. (eds Bengio, S.et al.) Advances in Neural Information Processing Systems, Vol. 31 (Curran Asso- ciates, Inc., 2018). URL https://proceedings.neurips.cc/paper files/paper/2018/ file/b58f7d184743106a8a66028b7a28937c-Paper.pdf

    2018

  47. [47]

    (eds Burges, C., Bottou, L., Welling, M., Ghahramani, Z

    Cuturi, M.Sinkhorn distances: Lightspeed computation of optimal transport. (eds Burges, C., Bottou, L., Welling, M., Ghahramani, Z. & Weinberger, K.) Advances in Neural Information Processing Systems, Vol. 26 (Curran Asso- ciates, Inc., 2013). URL https://proceedings.neurips.cc/paper files/paper/2013/ file/af21d0c97db2e27e13572cbf59eb343d-Paper.pdf

    2013

  48. [48]

    Pytorch: An imperative style, high-performance deep learning library

    Paszke, A.et al. Pytorch: An imperative style, high-performance deep learning library. (eds Wallach, H.et al.)Advances in Neural Information Processing Sys- tems, Vol. 32 (Curran Associates, Inc., 2019). URL https://proceedings.neurips. cc/paper files/paper/2019/file/bdbca288fee7f92f2bfa9f7012727740-Paper.pdf

    2019

  49. [49]

    Interpolating between optimal transport and mmd using sinkhorn divergences

    Feydy, J.et al. Interpolating between optimal transport and mmd using sinkhorn divergences. (eds Chaudhuri, K. & Sugiyama, M.)Proceedings of the Twenty- Second International Conference on Artificial Intelligence and Statistics, Vol. 89 ofProceedings of Machine Learning Research, 2681–2690 (PMLR, 2019). URL https://proceedings.mlr.press/v89/feydy19a.html. ...

    2019