Pith. sign in

REVIEW 5 major objections 6 minor 44 references

Pairwise Alignment & Compatibility for Arbitrarily Irregular Image Fragments

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper's PolEx measure computes pairwise compatibility for arbitrarily irregular, eroded image fragments by combining multiscale polygonal edge matching with diffusion-extrapolated pictorial bands, and a beam-search solver built on it…

desk verdict A genuinely new and useful compatibility pipeline for eroded fragments, but the SOTA claim is under-isolated and the diffusion-extrapolation substrate needs pairwise validation. read the letter →

arxiv 2507.09767 v1 pith:P2KGQJIJ submitted 2025-07-13 cs.CV

Ofir Itzhak Shahar , Gur Elkin , Ohad Ben-Shahar This is my paper · ORCID
classification cs.CV
keywords jigsawpuzzlefragmentreassemblypairwisecompatibilitydiffusionextrapolationerosionmodelingarchaeologicalreconstructionirregularfragmentsmultiscalepolygonalapproximation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that pairwise compatibility for arbitrarily irregular, eroded image fragments can be computed without restricting their shapes, sizes, or pictorial content, and that doing so improves archaeological puzzle solving. It introduces PolEx, which first uses a diffusion model to extrapolate each fragment's pictorial content into an eroded band, then represents fragment boundaries with multiscale polygons, prunes candidate alignments by edge-length ratios, and scores each alignment by LAB color dissimilarity of random patches over the overlapping extrapolated regions. The paper also contributes a 1000-puzzle synthetic dataset with Voronoi-based fragmentation, Perlin-noise erosion, and pairwise evaluation metrics. On the RePAIR 2D dataset, a beam-search solver built on PolEx reports state-of-the-art neighborhood precision, recall, and F1, which the paper reads as direct evidence that better pairwise compatibility lifts reconstruction quality.

What carries the argument

The load-bearing mechanism is the extrapolated band: a strip of diffusion-generated pictorial content around each fragment boundary that stands in for content lost to erosion. Around it, the pipeline uses a multiscale polygonal approximation (polygonal simplification plus synthesized augmented edges) to define candidate contact edges, edge-length ratio matching (min/max edge length above a threshold) to prune the infinite alignment space, rigid transforms that place paired edges opposite each other with a small gap, and a p-norm aggregation of random-patch LAB differences over the shared band region to produce the final compatibility score. The extrapolated band does the work of making pictorial comparison possible on real eroded pieces; the polygon machinery does the work of making the search finite without uniform sampling.

What would settle it

Rerun the RePAIR 2D neighbor benchmark with the extrapolated bands replaced by constant-color or nearest-edge-color extrapolations of the same width, keeping all PolEx parameters fixed. If neighborhood precision, recall, and F1 stay within noise of the reported values, the claimed role of learned extrapolation is not carrying the result.

Watch

Extended reading notes

Core claim

The central claim is that compatibility of two fragments is best decided by fusing geometric shape evidence with diffusion-extrapolated pictorial evidence: the geometry selects a sparse, informed set of candidate relative configurations, and the extrapolated bands supply the pictorial signal that ranks them. Concretely, PolEx builds a compact polygon per fragment, augments it with synthetic long edges for larger-scale structure, matches edges whose length ratio exceeds a threshold, derives a rigid transform per match, then scores the transformed pair by aggregating per-patch LAB differences in the intersection of the extrapolated bands, with an amplification factor for small non-coherent regions. The paper asserts that this unrestricted compatibility measure, embedded in a beam-search reconstruction, gives the best neighbor-based results on RePAIR 2D among the compared methods, and that the accompanying dataset isolates pairwise compatibility as a testable task.

Load-bearing premise

The whole scoring chain depends on the diffusion extrapolator generating plausible, distinctive pictorial content inside the eroded bands; if those extrapolations are hallucinated or uninformative, the pictorial ranking has no reliable signal.

Editorial extensions

If this is right

  • If PolEx is correct, fragment reassembly no longer needs square templates, low-degree polygonal approximations, or known image priors, because compatibility is computed directly from each pair's own geometry and extrapolated appearance.
  • State-of-the-art neighborhood precision and recall on RePAIR 2D imply that improving the pairwise stage alone can lift an end-to-end archaeological puzzle solver.
  • The pairwise metrics (Relative Position Score, Anchored RMSE) and the synthetic erosion dataset give later work a standardized way to benchmark compatibility independent of full assembly.
  • Ranking several top candidates instead of one is enough to capture most true neighbor matches, which suggests downstream global solvers can safely rely on a short candidate list from PolEx.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reliance on diffusion extrapolation may transfer to other pairwise registration tasks where boundary regions are incomplete, such as torn documents, broken tiles, or fragmented artworks, as long as the image content is locally coherent.
  • Because the edge-length-ratio filter is content-free, the candidate generation cost is controlled by fragment geometry; a natural test is whether a stricter ratio plus more pictorial candidates scales to puzzles with many more fragments.
  • The paper's own pitfall example - large uniform regions being aligned to the wrong shared area - suggests that adding texture- or patch-distribution-based features to the LAB score could close the remaining gap; this is an extension not explored in the paper.
  • The synthetic dataset's Voronoi/Delaunay structure gives exact ground-truth neighbor pairs, so it could double as a diagnostic for whether a compatibility method's failures come from eroded geometry or from ambiguous appearance.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes PolEx, a hybrid geometric-pictorial compatibility measure for arbitrarily irregular and eroded image fragments. PolEx first extrapolates each fragment's pictorial content with a Stable Diffusion model to create extrapolated bands, then obtains a multiscale polygonal approximation of the fragment boundary, extracts candidate matching edges, computes rigid alignment configurations from those edges, and scores each configuration by the LAB dissimilarity of random patch pairs sampled from the overlap of the two fragments' extrapolated bands. The authors introduce a synthetic dataset of 1000 archaeological puzzles generated via Voronoi tessellation and a Perlin-noise-based erosion model, plus two pairwise evaluation metrics. They embed PolEx into a beam-search reconstruction framework and report state-of-the-art neighborhood precision, recall, and F1 on the RePAIR 2D dataset.

Significance. If the claims hold, the paper addresses a genuine gap: pairwise compatibility for fragments with unrestricted shapes, arbitrary pictorial content, and realistic erosion. The proposed method is conceptually novel—using diffusion-based extrapolation for pairwise scoring is not, to my knowledge, standard in the puzzle-solving literature—and the dataset with a controllable erosion model plus the pairwise evaluation metrics could be useful to the community. The reported SOTA neighbor-based metrics on RePAIR 2D are potentially significant for archaeological puzzle solving. The strengths include the clear problem formulation, the explicit handling of erosion via extrapolated bands, and the provision of pairwise metrics rather than only assembly-level scores. However, the current evidence does not yet isolate PolEx's compatibility contribution from the new beam-search framework, and the central pairwise discrimination capability is not directly validated.

major comments (5)
  1. [§5, Table 1] The SOTA claim is confounded because the comparison changes two variables at once: the compatibility measure (PolEx) and the search strategy (beam search versus the greedy/genetic strategies of the prior works). Without an ablation that, for example, runs a previously published compatibility measure inside the same beam-search framework, or runs PolEx inside a prior greedy framework, Table 1 cannot be read as evidence that PolEx's compatibility measure is what improves neighbor-based metrics. Additionally, the composite score used in Algorithm 1 ("dissimilarity + overlap penalty") is not formally specified, so the contribution of the overlap penalty to the beam-search result is unknown.
  2. [§3.1, §3.3] The central premise is that two independently extrapolated bands from Stable Diffusion will agree at the pixel level for true neighbor pairs. Since neither extrapolation has access to the other fragment or to the true eroded content, this is a strong assumption that is never directly tested. The evaluation in §5 is restricted to true neighbor pairs—it asks whether the correct alignment among several candidate alignments is ranked highly, but it does not test whether PolEx can distinguish true neighbor pairs from non-neighbor pairs. This pairwise discrimination ability is load-bearing for the claimed compatibility improvement. Please add a pairwise-level experiment (e.g., ROC curves or precision-recall for true versus false pairs) on both the proposed synthetic dataset and RePAIR, or at minimum report the distributions of PolEx scores for true and false pairs.
  3. [§5, parameter selection] The sentence "All values were selected after simple rough grid search on their ranges" does not state whether the search was performed on a held-out validation set or directly on the RePAIR test set. If the test set was used for parameter selection, the reported precision, recall, and F1 numbers are optimistically biased and not directly comparable to baselines whose parameters were not tuned on the same test set. The authors should specify the exact validation protocol and report sensitivity of the results to the key parameters (γ, g, and the pictorial scoring parameters).
  4. [§2.2, §2.3] The dataset is claimed to mimic real-world archaeological erosion, but no validation supports this assertion. There is no comparison of the erosion statistics to those of real eroded fragments (e.g., from RePAIR), no ablation of the erosion-rate parameter against real data, and no quantitative justification for the fixed perpendicular gap g beyond an assertion that it accounts for erosion. Since the synthetic dataset is used to claim pairwise alignment performance, the realism of the erosion model should be demonstrated, or the claim should be softened to "a controllable erosion model" rather than "mimics real-world archaeological erosion."
  5. [§6 versus Table 1] The conclusion states that PolEx "achieves SOTA results on the RePAIR 2D dataset in all neighbor or pairwise metrics presented in their work [41]", but Table 1 reports only Qpos, RMSE (R and T), Precision, Recall, and F1. Several metrics from the RePAIR 2D benchmark appear to be missing from the table, so the conclusion overstates what is demonstrated. The claim should be limited to the metrics actually reported, or the missing RePAIR metrics should be added.
minor comments (6)
  1. [§2.2] Typo: "Delauney triangulation" should be "Delaunay triangulation."
  2. [§3.1] The text says "Stable-Diffusion 1.4 ... trained for the inpainting task," but the base Stable Diffusion 1.4 model is not an inpainting model; inpainting requires a fine-tuned variant or a specialized pipeline. Please clarify the exact model and inpainting configuration used.
  3. [§3.3] The amplification factor λ is introduced to "scale the score if prominent color exceptions are detected," but its computation is never defined. Please provide the formula or algorithmic description for λ.
  4. [Appendix A] The "composite score combining pictorial dissimilarity and an overlap penalty" used in the beam-search is not specified. Please give the exact formula used in the experiments.
  5. [General] The abstract calls the approach "efficient," but the paper provides no runtime or complexity comparison against the baselines. Adding wall-clock times or a complexity discussion would strengthen the claim.
  6. [General] No link to code or dataset is provided. Releasing the dataset generation code and the PolEx implementation would substantially improve reproducibility and community uptake.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the RePAIR SOTA claim is an external-benchmark comparison and PolEx compatibility is computed from extrapolated bands and alignment geometry, not from the target output.

full rationale

PolEx's compatibility score (Sec. 3.3) is defined by LAB color differences between extrapolated bands after a rigid alignment derived from polygonal edge lengths and angles (Secs. 3.2.2-3.2.3). None of these definitions uses ground-truth neighbor labels, the evaluation metrics of Sec. 4, or the assembly hypotheses of Appendix A. The central SOTA claim (Sec. 5, Table 1) is measured on the external RePAIR 2D benchmark [41] against published baselines [14], [41]; it is therefore an empirical comparison rather than a quantity entailed by the method's own inputs. The new synthetic dataset's ground truth arises from the Voronoi/Delaunay generation process and is used only for evaluation, not as a solver input. Self-citations ([17], [22], [34]) appear in dataset prompt styling, related-work positioning, and baseline descriptions; none supplies a uniqueness theorem or otherwise constrains the compatibility design, so they are not load-bearing. The grid-search parameter statement in Sec. 5 lacks an explicit validation split, which is a potential overfitting concern about the empirical claim, but the paper never states that RePAIR test labels were used to fit parameters, so it cannot be exhibited as fitted-input-called-prediction circularity. Accordingly, no step in the paper's derivation reduces to its own output.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim depends on many hand-set thresholds and on the unverified reliability of diffusion extrapolation and the synthetic erosion model. None of these parameters are derived from first principles or measured against independent physical data; they are choices of the authors, which is the honest price of the method.

free parameters (8)
  • gamma (length-ratio threshold for candidate edges) = 0.5
    Filters edge pairs by min(Lt,Ls)/max(Lt,Ls) >= gamma; set by grid search in Sec 5, affects which alignments are even considered.
  • alpha (polygon simplification tolerance scale) = 0.005
    epsilon = alpha*p in RDP simplification; chosen by grid search, controls how faithfully the shape is retained.
  • Delta_alpha (angle threshold for vertex removal) = 10 degrees
    Drops middle vertices with angle below Delta_alpha; grid-searched, changes polygon topology.
  • g (erosion gap offset) = 10 px
    Shifts target edge midpoint outward by g along the normal; a single global gap models erosion, grid-searched.
  • k_smooth (Gaussian smoothing kernel) = 3
    Binary mask smoothing before contour extraction; affects boundary noise handling; grid-searched.
  • l_min (minimum edge length) = 15 px
    Short edges are discarded before candidate extraction; grid-searched, reported only in Sec 5.
  • erosion rate parameter (dataset generation) = not reported
    Controls the Perlin-noise-scaled deletion radius along Voronoi boundaries; never given a numeric value or validated against real erosion.
  • pictorial scoring hyperparameters (P_min, P_max, p-norm exponent, lambda) = not reported
    Patch size range, aggregation exponent, and amplification factor determine the compatibility score but no values or sensitivity analysis are provided.
assumptions (5)
  • standard math Delaunay triangulation of the Voronoi sites exactly encodes neighboring fragment pairs.
    Sec 2.2: used to extract ground-truth pairs; standard duality result.
  • domain assumption RDP polygonization at tolerance alpha*p preserves the boundary segments that carry true alignments.
    Sec 3.2.1: if true matching curves are broken into non-collinear pieces beyond what augmented edges can bridge, the correct alignment is never generated.
  • domain assumption Stable Diffusion extrapolation of eroded fragments yields pictorial content whose similarity is a reliable proxy for true adjacency.
    Sec 3.1 and 3.3: every candidate alignment is scored only on extrapolated bands; hallucinated content can promote false matches, as the paper acknowledges in Sec 5 and Fig 7.
  • ad hoc to paper A fixed perpendicular gap g along the target edge normal accounts for erosion in real and synthetic fragments.
    Sec 3.2.3: one scalar cannot represent spatially varying erosion, so true alignments may be systematically offset.
  • domain assumption RePAIR 2D ground truth and the synthetic dataset are compatible evaluation regimes.
    Sec 5: parameters tuned on synthetic puzzles are applied to real scans without domain shift adjustment.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pairwise Alignment & Compatibility for Arbitrarily Irregular Image Fragments." pith.science (2026). https://pith.science/paper/P2KGQJIJ

@misc{pith2026250709767,
  author       = {Pith},
  title        = {Pith review of: Pairwise Alignment & Compatibility for Arbitrarily Irregular Image Fragments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P2KGQJIJ}},
  note         = {Machine review of arXiv:2507.09767}
}
read the original abstract

Pairwise compatibility calculation is at the core of most fragments-reconstruction algorithms, in particular those designed to solve different types of the jigsaw puzzle problem. However, most existing approaches fail, or aren't designed to deal with fragments of realistic geometric properties one encounters in real-life puzzles. And in all other cases, compatibility methods rely strongly on the restricted shapes of the fragments. In this paper, we propose an efficient hybrid (geometric and pictorial) approach for computing the optimal alignment for pairs of fragments, without any assumptions about their shapes, dimensions, or pictorial content. We introduce a new image fragments dataset generated via a novel method for image fragmentation and a formal erosion model that mimics real-world archaeological erosion, along with evaluation metrics for the compatibility task. We then embed our proposed compatibility into an archaeological puzzle-solving framework and demonstrate state-of-the-art neighborhood-level precision and recall on the RePAIR 2D dataset, directly reflecting compatibility performance improvements.

Figures

Figures reproduced from arXiv: 2507.09767 by the authors.

Figure 1
Figure 1. Samples from publicly available puzzle datasets. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (a-h) Puzzle generation process. (i) Pair extraction. (j) An eroded fragment (pictorial) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Empirical properties obtained from multiple non-eroded puzzles with 10, 20, . . . , 100 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Illustration of the compatibility calculation process on two fragments from the RePAIR [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Fragment Extrapolation Process. First, the fragment’s binary mask ( [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative Reconstruction Results on pairs of fragments from our dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Potential pitfalls of our approach. Top: Initial positions of the two fragments. Middle: The reconstruction chosen by PolEx. Bottom: The ground truth reconstruction. considering multiple top-ranked candidates rapidly improves the overall scores in all measures. This ca…
Figure 8
Figure 8. Figure 8: Quantitative results on all pairs in the test set, with different [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Qualitative reconstruction results on a selected fragments group from the RePAIR [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 39 canonical work pages

  1. [41]

    Re-assembling the past: The repair dataset and benchmark for real world 2d and 3d puzzle solving,

    T. Tsesmelis, L. Palmieri, M. Khoroshiltseva,et al., “Re-assembling the past: The repair dataset and benchmark for real world 2d and 3d puzzle solving,” inAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave,et al., Eds., vol. 37, Curran Asso- ciates, Inc., 2024, pp. 30076–30105. [Online]. Available:https://proceedings.n...

    work page 2024

  2. [1]

    Archaeological Park of Pompeii,The RePAIR Project Begins – Robotics and Digitisation at the Service of Archaeology, Press Release, Sep. 2021. [Online]. Available:https://pompeiisites. org/en/comunicati/the- repair- project- begins- robotics- and- digitisation- at- the-service-of-archaeology

    work page 2021

  3. [2]

    Saffo: A sift based approach for digital anastylosis for fresco reconstruction,

    P. Barra, S. Barra, M. Nappi,et al., “Saffo: A sift based approach for digital anastylosis for fresco reconstruction,”Pattern Recognition Letters, vol. 138, pp. 123–129, 2020,issn: 0167-8655. doi: https://doi.org/10.1016/j.patrec.2020.07.008 . [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0167865520302543

    work page doi:10.1016/j.patrec.2020.07.008 2020

  4. [3]

    Solving jigsaw puzzles with eroded boundaries,

    D. Bridger, D. Danon, and A. Tal, “Solving jigsaw puzzles with eroded boundaries,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 3526–3535

    work page 2020

  5. [4]

    A system for high-volume acquisition and matching of fresco fragments: Reassembling theran wall paintings,

    B. J. Brown, C. Toler-Franklin, D. Nehab,et al., “A system for high-volume acquisition and matching of fresco fragments: Reassembling theran wall paintings,”ACM transactions on graphics (TOG), vol. 27, no. 3, pp. 1–9, 2008

    work page 2008

  6. [5]

    Tools for virtual reassembly of fresco fragments,

    B. J. Brown, L. Laken, P. Dutré,et al., “Tools for virtual reassembly of fresco fragments,” International Journal of Heritage in the Digital Era, vol. 1, pp. 313–329, 2012

    work page 2012

  7. [6]

    Language models are few-shot learners,

    T. Brown, B. Mann, N. Ryder,et al., “Language models are few-shot learners,”arXiv preprint arXiv:2005.14165, 2020

    arXiv 2005

  8. [7]

    2d irregular fragment reassembly with deep learning assistance,

    Y. Cao, Z. Fang, H. Tian,et al., “2d irregular fragment reassembly with deep learning assistance,” IEEE Access, vol. 12, 2024. doi: 10 . 1109 / ACCESS . 2024 . 3368004. [Online]. Available: https://ieeexplore.ieee.org/document/3368004

    work page arXiv 2024

Show all 44 references
  1. [8]

    Caple and V

    C. Caple and V. Garlick, Eds.,Studies in Archaeological Conservation. London: Routledge, 2020

    2020

  2. [9]

    Global consistency in the automatic assembly of fragmented artefacts,

    A. Castañeda, B. J. Brown, S. Rusinkiewicz,et al., “Global consistency in the automatic assembly of fragmented artefacts,” inVAST, 2011

    2011

  3. [10]

    Jigsaw-vit: Learning jigsaw puzzles in vision transformer,

    Y. Chen, X. Shen, Y. Liu,et al., “Jigsaw-vit: Learning jigsaw puzzles in vision transformer,” Pattern Recognition Letters, vol. 166, pp. 53–60, 2023,issn: 0167-8655. doi: https://doi. org/10.1016/j.patrec.2022.12.023 . [Online]. Available:https://www.sciencedirect. com/science...

    2023 doi

  4. [11]

    Recon- struction of frescoes by sequential layers of feature extraction,

    T. da Silva Teixeira, M. L. Sguario Coelho de Andrade, and M. Rodrigues da Luz, “Recon- struction of frescoes by sequential layers of feature extraction,”Pattern Recognition Letters, vol. 147, pp. 172–178, 2021,issn: 0167-8655. doi: https://doi.org/10.1016/j.patrec. 2021.04.01...

    2021 doi

  5. [12]

    Sur la sphère vide. a la mémoire de georges voronoï,

    B. Delaunay, “Sur la sphère vide. a la mémoire de georges voronoï,” French,Bulletin de l’Académie des Sciences de l’URSS. Classe des sciences mathématiques et naturelles, vol. 6, pp. 793–800, 1934

    1934

  6. [13]

    Jigsaw puzzles, edge matching, and polyomino packing: Connections and complexity,

    E. D. Demaine and M. L. Demaine, “Jigsaw puzzles, edge matching, and polyomino packing: Connections and complexity,”Graphs and Combinatorics, vol. 23, no. Suppl 1, pp. 195–208, 2007

    2007

  7. [14]

    Solving archaeological puzzles,

    N. Derech, A. Tal, and I. Shimshoni, “Solving archaeological puzzles,”Pattern Recognition, vol. 119, p. 108065, 2021. 12

    2021

  8. [15]

    Dafne: A dataset of fresco fragments for digital anastlylosis,

    P. Dondi, L. Lombardi, and A. Setti, “Dafne: A dataset of fresco fragments for digital anastlylosis,” Pattern Recognition Letters, vol. 138, pp. 631–637, 2020,issn: 0167-8655. doi: https://doi.org/10.1016/j.patrec.2020.09.015 . [Online]. Available:https://www. sciencedirect.co...

    2020 doi

  9. [16]

    Algorithms for the reduction of the number of points required to represent a digitized line,

    D. H. Douglas and T. K. Peucker, “Algorithms for the reduction of the number of points required to represent a digitized line,”Cartographica, vol. 10, no. 2, pp. 112–122, 1973

    1973

  10. [17]

    Recognizing artistic style of archaeological image fragments using deep style extrapolation,

    G. Elkin, O. I. Shahar, Y. Ohayon,et al., “Recognizing artistic style of archaeological image fragments using deep style extrapolation,” inCulture and Computing: 13th International Conference, C&C 2025, Held as Part of the 27th HCI International Conference, HCII 2025, Gothenbu...

    2025 doi

  11. [18]

    Review of computer-based methods for archaeological ceramic sherds reconstruction,

    D. Eslami, L. Di Angelo, P. Di Stefano,et al., “Review of computer-based methods for archaeological ceramic sherds reconstruction,”Virtual Archaeology Review, vol. 11, no. 23, pp. 34–49, 2020.doi: 10.4995/var.2020.13134

    2020

  12. [19]

    Apictorial jigsaw puzzles: The computer solution of a problem in pattern recognition,

    H. Freeman and L. Garder, “Apictorial jigsaw puzzles: The computer solution of a problem in pattern recognition,”IEEE Transactions on Electronic Computers, no. 2, pp. 118–127, 1964

    1964

  13. [20]

    Learning how to match fresco fragments,

    T. Funkhouser, H. Shin, C. Toler-Franklin,et al., “Learning how to match fresco fragments,” ACM Journal on Computing and Cultural Heritage, vol. 4, 7:1–7:13, 2011

    2011

  14. [21]

    From square pieces to brick walls: The next challenge in solving jigsaw puzzles,

    S. Gur and O. Ben-Shahar, “From square pieces to brick walls: The next challenge in solving jigsaw puzzles,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 4029–4037

    2017

  15. [22]

    Pictorial and apictorial polygonal jigsaw puzzles from arbitrary number of crossing cuts,

    P. Harel, O. I. Shahar, and O. Ben-Shahar, “Pictorial and apictorial polygonal jigsaw puzzles from arbitrary number of crossing cuts,”International Journal of Computer Vision, 2024, issn: 1573-1405. doi: 10.1007/s11263-024-02033-7. [Online]. Available:https://doi.org/ 10.1007/...

    2024 doi

  16. [23]

    Puzzlefusion: Unleashing the power of diffusion models for spatial puzzle solving,

    S. Hosseini, M. A. Shabani, S. Irandoust,et al., “Puzzlefusion: Unleashing the power of diffusion models for spatial puzzle solving,” inThirty-seventh Conference on Neural Information Pro- cessing Systems, 2023. [Online]. Available:https://openreview.net/forum?id=Z764QxwETf

    2023

  17. [24]

    Nash meets wertheimer: Using good continua- tion in jigsaw puzzles,

    M. Khoroshiltseva, L. Palmieri, and S. Aslan, “Nash meets wertheimer: Using good continua- tion in jigsaw puzzles,”Proceedings of the Asian Conference on Computer Vision (ACCV),

  18. [25]

    Solving jigsaw puzzles by predicting fragment’s coordinate based on vision transformer,

    G. Kim, H. Cho, and H. Nam, “Solving jigsaw puzzles by predicting fragment’s coordinate based on vision transformer,”Expert Systems with Applications, vol. 272, p. 126776, 2025, issn: 0957-4174. doi: https://doi.org/10.1016/j.eswa.2025.126776. [Online]. Available: https://www....

    2025

  19. [26]

    Fast and efficient reconstruction of digitized frescoes,

    N. Lermé, S. L. Hégarat-Mascle, B. Zhang, et al., “Fast and efficient reconstruction of digitized frescoes,”Pattern Recognition Letters, vol. 138, pp. 417–423, 2020,issn: 0167-8655. doi: https : / / doi . org / 10 . 1016 / j . patrec . 2020 . 08 . 006. [Online]. Available: htt...

    2020

  20. [27]

    A texture based matching approach for auto- mated assembly of puzzles,

    Mahmut Şamil Sağıroğlu and Aytül Erçil, “A texture based matching approach for auto- mated assembly of puzzles,” inProceedings of the 18th International Conference on Pattern Recognition (ICPR’06), Hong Kong, China: IEEE, Aug. 2006, pp. 1036–1041

    2006

  21. [28]

    Optimization for automated assembly of puzzles,

    Mahmut Şamil Sağıroğlu and Aytül Erçil, “Optimization for automated assembly of puzzles,” TOP, vol. 18, no. 2, pp. 321–338, Oct. 2010.doi: 10.1007/s11750-010-0156-6 . [Online]. Available: https://doi.org/10.1007/s11750-010-0156-6

    2010 doi

  22. [29]

    Interface area, edge length, and number of vertices in crystal aggregates with random nucleation,

    J. Meijering, “Interface area, edge length, and number of vertices in crystal aggregates with random nucleation,”Philips Research Reports, vol. 8, pp. 270–290, 1953

    1953

  23. [30]

    Deepzzle: Solving visual jigsaw puzzles with deep learning and shortest path optimization,

    M.-M. Paumard, D. Picard, and H. Tabia, “Deepzzle: Solving visual jigsaw puzzles with deep learning and shortest path optimization,”IEEE Transactions on Image Processing, vol. 29, pp. 3569–3581, 2020

    2020

  24. [31]

    An image synthesizer,

    K. Perlin, “An image synthesizer,”ACM Siggraph Computer Graphics, vol. 19, no. 3, pp. 287– 296, 1985

    1985

  25. [32]

    A survey of geometric analysis in cultural heritage,

    R. Pintus, K. Pal, Y. Yang,et al., “A survey of geometric analysis in cultural heritage,” Computer Graphics Forum, vol. 35, 2016

    2016

  26. [33]

    Geometric analysis in cultural heritage.,

    R. Pintus, K. Pal, Y. Yang,et al., “Geometric analysis in cultural heritage.,” inGCH, 2014, pp. 117–133

    2014

  27. [34]

    A fully automated greedy square jigsaw puzzle solver,

    D. Pomeranz, M. Shemesh, and O. Ben-Shahar, “A fully automated greedy square jigsaw puzzle solver,” inCVPR 2011, 2011, pp. 9–16.doi: 10.1109/CVPR.2011.5995331

    2011

  28. [35]

    A generic hybrid framework for 2d visual reconstruc- tion,

    D. Rika, D. Sholomon, E. David,et al., “A generic hybrid framework for 2d visual reconstruc- tion,” arXiv preprint arXiv:2501.19325, 2025

    2025 arXiv

  29. [36]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz,et al., “High-resolution image synthesis with latent diffusion models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2022, pp. 10684–10695

    2022

  30. [37]

    Serra,Image Analysis and Mathematical Morphology

    J. Serra,Image Analysis and Mathematical Morphology. San Diego, CA: Academic Press, 1982

    1982

  31. [38]

    Siamese-discriminant deep reinforcement learning for solving jigsaw puzzles with large eroded gaps,

    X. Song, J. Jin, C. Yao,et al., “Siamese-discriminant deep reinforcement learning for solving jigsaw puzzles with large eroded gaps,” inAAAI Conference on Artificial Intelligence, 2023. [Online]. Available:https://api.semanticscholar.org/CorpusID:259755242

    2023

  32. [39]

    Ganzzle++: Generative approaches for jigsaw puzzle solving as local to global assignment in latent spatial representations,

    D. Talon, A. Del Bue, and S. James, “Ganzzle++: Generative approaches for jigsaw puzzle solving as local to global assignment in latent spatial representations,”Pattern Recognition Letters, vol. 187, pp. 35–41, 2025,issn: 0167-8655. doi: https:/ /doi.org /10.1016 /j. patrec.20...

    2025

  33. [40]

    Multi-feature matching of fresco fragments,

    C. Toler-Franklin, B. J. Brown, T. Weyrich,et al., “Multi-feature matching of fresco fragments,” in SIGGRAPH 2010, 2010

    2010

  34. [42]

    Multi-phase relaxation labeling for square jigsaw puzzle solving,

    B. Vardi, A. Torcinovich, M. Khoroshiltseva,et al., “Multi-phase relaxation labeling for square jigsaw puzzle solving,”arXiv preprint arXiv:2303.14793, 2023. 14

    2023 arXiv

  35. [43]

    Self-supervised fragment alignment with gaps,

    M. Yang, Y. Svirsky, Z. Cheng,et al., “Self-supervised fragment alignment with gaps,”IEEE Transactions on Visualization and Computer Graphics, vol. 30, no. 9, pp. 6235–6246, 2024. doi: 10.1109/TVCG.2023.3330859. Appendix A PolEx compatibility in puzzle solving While this is no...

    2024

  36. [2024]

    [Online]. Available:https://openaccess.thecvf.com/content/ACCV2024/papers/ Khoroshiltseva _ Nash _ Meets _ Wertheimer _ Using _ Good _ Continuation _ in _ Jigsaw _ Puzzles_ACCV_2024_paper.pdf

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.