Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

A Generic Hybrid Framework for 2D Visual Reconstruction

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Pairing a deep-learning compatibility model with a genetic solver reconstructs Portuguese tile panels at 95.2% and 89.4% neighbor accuracy, besting the prior eroded-puzzle method by up to 35.1 points.

desk verdict A capable hybrid DL+GA jigsaw solver with broad experiments; the Portuguese-tile SOTA claims are provisional until train/test overlap is ruled out. read the letter →

arxiv 2501.19325 v1 pith:W5D2XSNP submitted 2025-01-31 cs.CV

Daniel Rika , Dror Sholomon , Eli David , Alexandre Pais , Nathan S. Netanyahu This is my paper · ORCID
classification cs.CV
keywords jigsawpuzzleproblemvisualreconstructiondeeplearningcompatibilitymeasuregeneticalgorithmPortuguesetilepanelserodedboundariesshreddeddocumentsconvolutionalneuralnetworks
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 for real-world image reassembly tasks, the limiting factor is the compatibility measure: when tile edges are degraded or the imagery is repetitive, boundary-only color comparisons fail and greedy placement locks in early errors. The proposed remedy is a hybrid in which a compact convolutional network scores whole pairs of tiles as potential neighbors, and a genetic algorithm searches for the global arrangement using those scores. The authors report 95.2% and 89.4% neighbor accuracy on known-dimension Type-1 and Type-2 Portuguese tile panels, up to 96.9% on standard synthetic benchmarks, and average gains of 16.2 and 35.1 percentage points over the prior GAN-based method on 7% and 14% eroded puzzles. If this holds, the same two-part recipe—learned whole-piece compatibility plus evolutionary global search—applies across domains with little per-task engineering, and it could turn an archaeology-scale manual reassembly effort into a machine-assisted workflow.

What carries the argument

The load-bearing object is the DLCM–GA combination. The DLCM is a convolutional network that ingests a whole piece pair as a $P\times 2P$ image rather than comparing boundary pixels, so the compatibility signal can come from interior texture, color, and structure; the GA then treats the resulting pairwise scores as a fitness landscape and searches globally with a hierarchical crossover that places tiles by parent agreement, best-buddy relations, and fallback compatibility, plus random phase-skipping as mutation. Post-processing of the score matrix—per-edge min–max normalization and symmetrization—is also load-bearing, since it lifts DLCM Top-1 accuracy from 64.5% to 69.9% on Type-1 panels.

What would settle it

Re-score the 24 museum test panels using ground-truth tile boundaries obtained by human marking or high-resolution seam detection instead of the automated 50x50 crop, and recompute neighbor accuracy. If the 95.2% Type-1 and 89.4% Type-2 figures drop substantially, the reported state-of-the-art is an artifact of misaligned crop labels rather than a true reconstruction capability.

Watch

Extended reading notes

Core claim

The paper's central claim is that a compatibility measure which sees entire pieces, not just their abutting edges, is enough to make large real-world jigsaw puzzles tractable. The proposed DLCM is a compact convolutional network that takes a concatenated pair of $P\times 2P$ tiles and outputs a scalar score; for the tile domain it is an ensemble of four such networks, one per color channel plus an RGB network, trained with binary cross-entropy on sampled positive and negative pairs and augmented with boundary degradation and pixel shifts. Raw scores are min–max normalized per edge and symmetrized so $C(e_i,e_j)=C(e_j,e_i)$. The companion solver is a genetic algorithm whose crossover grows a kernel through hierarchical phases, including parent-confidence phases and a best-buddies phase, with mutation that skips phases to escape local optima. On the paper's evaluation this yields 95.2% and 89.4% known-dimension neighbor accuracy for Type-1 and Type-2 Portuguese tile panels, new best average results across the standard synthetic Type-1 and Type-2 benchmarks, and average gains of 16.2 and 35.1 percentage points over the previous GAN-based method on 7% and 14% eroded puzzles. The same pipeline reconstructs a 765-strip shredded-document puzzle at 97.1% accuracy.

Load-bearing premise

The claim depends on the 24 Portuguese museum test panels being cut into tiles along their true boundaries: if the automatic 50x50 cropping is even slightly misaligned, the 'correct neighbor' labels are wrong, and the reported 95.2% and 89.4% accuracy would be inflated.

Editorial extensions

If this is right

  • Portuguese tile panels with known dimensions can be assembled automatically to near-perfect neighbor accuracy, reducing a decades-long manual effort to a machine-assisted task.
  • The same trained compatibility network transfers to synthetic jigsaw benchmarks and to eroded-boundary puzzles with only a change in training data, so the hybrid is a general recipe rather than a tile-specific method.
  • Heavily eroded pieces (14% of boundary pixels removed) remain reconstructible at 85–92% neighbor accuracy, roughly 35 points above the prior GAN-based method.
  • On strip-cut shredded documents, the framework reaches 97.1% accuracy on a 765-strip multi-page puzzle, enough to recover the text content.
  • The GA's stochastic restarts matter: best-of-50 accuracy is 95.2% Type-1 and 89.4% Type-2, while average-of-runs is 93.1% and 79%, so the reported state-of-the-art figures require multiple runs per puzzle.

Reading between the lines

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

  • Because the reconstruction accuracy (95.2% Type-1) far exceeds the DLCM's Top-1 compatibility accuracy (69.9% Type-1), the GA is not merely summing evidence—it is actively correcting many wrong first choices; a testable consequence is that improving the compatibility measure may matter less than improving the solver's global search in this regime.
  • The automated-cropping caveat applies to absolute accuracy; however, since every compared method is scored on the same labels, the relative gap over prior methods is more trustworthy than the headline numbers.
  • A direct stress test would evaluate on panels whose tile seams are known from the physical tiles or manually marked, so the sensitivity to crop alignment can be quantified.
  • The computational bottleneck the paper flags—computing $16N^2$ pairwise scores—means the practical ceiling on the number of pieces is set by the CNN, not the GA; embedding-based compatibility could extend the same recipe to tens of thousands of pieces.
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

4 major / 4 minor

Summary. The paper proposes a two-stage framework for square-piece visual reconstruction: a deep CNN-based compatibility measure (DLCM) trained on full piece pairs via binary cross-entropy, followed by a genetic algorithm solver with a hierarchical crossover. The framework is evaluated on Portuguese tile panels (24 MNAz test panels), synthetic JPP benchmarks, eroded-boundary puzzles, and shredded documents. The authors report state-of-the-art results, including 95.2% and 89.4% neighbor accuracy for Type-1 and Type-2 Portuguese tile panels with known dimensions (Table II), and 16.2%/35.1% average improvements over Bridger et al. on eroded puzzles (Table VI). The contribution also includes a new benchmark dataset of Portuguese tiles.

Significance. If the reported numbers are reproducible under clean held-out conditions, the paper demonstrates a substantial empirical advance, particularly for Type-2 puzzles and eroded boundaries, and the ablation study (Table III) is a useful decomposition of the GA phases. The release of the Portuguese tile benchmark is a valuable asset to the community. However, the central SOTA claims are currently supported by comparisons that are vulnerable to data-contamination risk and to best-of-N reporting bias; the paper provides no statistical evidence for the advantage over baselines. These issues must be resolved before the claims can be accepted.

major comments (4)
  1. [V-A, Tables I-II] Section V-A states that the 24 MNAz test panels were excluded from CNN training, but the 208 Internet-acquired training images (some 'taken by casual tourists') are not checked for duplication or near-duplication with the test panels. Because the DLCM is trained on concatenated piece pairs, a tourist photo containing any test panel would provide direct supervision for exactly the adjacency pairs used in the Top-1 evaluation of Table I and the neighbor-accuracy evaluation of Table II. The reported 95.2% and 89.4% accuracies therefore rest on an unverified assumption of clean held-out generalization. The authors should verify, by image retrieval or manual inspection, that no test panel appears in the training images, or retrain on a deduplicated training set and report the resulting Top-1 and neighbor-accuracy numbers.
  2. [V-C/D, Tables II, III, VI] The proposed results are reported as the 'best result, after running our enhanced GA module 50 times on each image' (Section V-D), while the baseline methods are not described as receiving the same multiple-run treatment. With stochastic solvers, best-of-50 systematically inflates the expected reported accuracy relative to a single run, and the paper gives no error bars, confidence intervals, or significance tests. The very large gaps in Table II (e.g., 95.2% vs. 28% for Bridger et al.) could be partly an artifact of comparing a selected best run against a single run. Please report mean and standard deviation (or median and IQR) over the 50 runs for the proposed method, run each baseline for the same number of trials with the same stopping rule, and report paired significance tests (e.g., Wilcoxon signed-rank over images). The same issue applies to the erosion experiments in Table VI and the ablation results in Table III.
  3. [V-A, III-B] The paper does not specify how the 24 MNAz test panels are converted into 50x50 pieces for evaluation. In Section V-A, automated piece-cropping is described and immediately followed by the concession that 'automated cropping may not always align perfectly with actual piece boundaries.' If the test-set tiles are produced by the same automated procedure without manual verification, the ground-truth adjacency labels used to compute the neighbor accuracies in Table II may be incorrect, which would affect all compared methods to different degrees and undermine the SOTA comparison. The authors should state the exact cropping procedure for the test panels and, if automated, quantify the alignment error or provide manual verification for the 24 test panels.
  4. [VI-A, Tables IV-V] For synthetic JPP, the manuscript reports 'average best results obtained over five runs of our scheme per image' (Section VI-A), while the Portuguese-tile section reports the best result over 50 runs; the aggregate statistic is ambiguous. Please clarify the number of runs and the aggregation rule used in Tables IV and V, and report variance so that the SOTA claims on synthetic puzzles can be assessed on the same footing as the baseline numbers.
minor comments (4)
  1. [V-D] Section V-D duplicates Section V-C almost verbatim (the text on elitism, roulette-wheel selection, phase-skip probabilities, and the 50-run protocol appears twice); the duplicate should be removed.
  2. [V-A] The sentence 'Although automated cropping may not always align perfectly with actual piece boundaries, such occurrences are rare and may contribute positively by reducing the risk of overfitting' is unconvincing: misaligned training crops reduce label quality rather than mitigating overfitting, and the claim should be reworded or supported.
  3. [Table I] Top-1 accuracies in Table I are reported as point estimates without any measure of variation across the 24 test panels; a per-panel standard deviation or confidence interval would strengthen the comparison with the baseline CMs.
  4. [Abstract] The phrase 'our unique hybrid methodology' overstates novelty given that the authors' prior work [15] already combines a DL-based CM with a GA solver; the abstract should describe the specific extensions rather than claiming uniqueness.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the DLCM is trained on a separate set with held-out test panels, the GA maximizes a CM-sum fitness that is distinct from the ground-truth accuracy metric, and SOTA claims are empirical comparisons rather than reductions to inputs.

full rationale

The paper's central claims are empirical systems results rather than analytic derivations, and no load-bearing step reduces to its own inputs by construction. The DLCM is a supervised compatibility model trained (Section III-A) on triplets whose positive pairs are true adjacent edges from 208 Internet-sourced Portuguese tile images, with a separate 18-image validation set (Section V-A); the 24 MNAz test panels are explicitly excluded from training: 'The 24 high-resolution test images from the MNAz were excluded from the CNN training process.' Reported Top-1 and neighbor accuracies are then computed against ground-truth adjacency on that held-out test set, so the CM's evaluation labels are not the same data that define its training objective. The GA solver (Section IV) maximizes a fitness defined as the sum of pairwise DLCM scores, while reconstruction accuracy is measured separately as 'the fraction of correctly assigned adjacent edges with respect to ground truth,' so the optimization target and the evaluation metric are distinct. The authors adopt the GA from their own prior work [13] and compare against their prior hybrid [15], but these are component choices and baseline comparisons, not citation chains invoked to force a conclusion; Table III ablates the GA phases, and Tables II, IV, V, and VI benchmark against external baselines. The post-processing normalization and symmetry averaging (Equations 4 and 5) are straightforward score transformations that do not encode the answer. The only notable risk is an unverified data-hygiene assumption: the 208 training images, some 'taken by casual tourists,' are not checked for overlap with the 24 test panels, so test-set leakage could inflate the headline numbers; however, this is an external-validity and fairness concern, not a circularity by construction, because no fitted parameter or equation is shown to be equivalent to the target result.

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

The framework rests on standard machine-learning practice plus several domain-specific choices. The free parameters are the GA and training hyperparameters, chosen empirically. The main axioms are about learnability and ground-truth correctness. No invented entities are introduced.

free parameters (8)
  • alpha0 = 0.8
    Initial compatibility threshold in GA crossover Phase 1.1. Tuned by trial and error (Section IV).
  • GA phase-skip probabilities = 0.10 for Phases 1.1/1.2, 0.20 for Phases 2/3
    Mutation probabilities that skip hierarchical phases to escape local optima. Set empirically (Section V-C/V-D).
  • GA population size = 100
    Number of chromosomes per generation, set for all experiments (Section V-D).
  • GA termination patience = 50 generations
    Stops when no improved chromosome appears for 50 generations (Section V-D).
  • CNN learning rate = 0.0001
    Adam optimizer learning rate for DLCM training (Section V).
  • CNN batch size = 64
    Training batch size (Section V).
  • Piece resolution = 50x50 (Portuguese, shredded), 64x64 (eroded), 28x28 (synthetic)
    Input tile size chosen per domain; affects the CNN input dimensions and puzzle scale.
  • Augmentation policy = degradation and shift applied only to Portuguese tiles
    Data augmentations (Section III-A) were beneficial for Portuguese tiles but excluded from other domains, so this is a per-domain modeling choice tuned on validation performance.
assumptions (5)
  • domain assumption The adjacency of two tiles in the original image is learnable from the pixel content of the whole tiles, and the CNN generalizes from the training set to the held-out MNAz test panels.
    The DLCM is trained on 208 internet panels and 18 validation images, then applied to 24 MNAz test panels. If the CNN overfits to the training distribution, the compatibility scores used by the GA would be unreliable and the SOTA claim would fail.
  • domain assumption The compatibility scores are pairwise additive, so that the GA fitness function (sum of scores across all boundaries) is a valid proxy for reconstruction quality.
    The GA in Section IV maximizes the sum of pairwise DLCM scores; the paper assumes that higher total score correlates with more correct adjacencies, which is not proven but is empirically supported by the fitness-versus-ground-truth comparison.
  • domain assumption The automated cropping of the MNAz test panels into 50x50 tiles aligns with the true tile boundaries, making the ground-truth adjacency labels correct.
    Neighbor accuracy is computed against the grid produced by automated cropping (Section V-A). If the grid is misaligned, the reported accuracy numbers are not a true measure of reconstruction quality.
  • domain assumption The 24 MNAz test panels are representative of the Portuguese tile reconstruction problem, and the manual determination of row/column counts is accurate.
    The SOTA claim for Portuguese tiles is based on these 24 panels; a biased or small sample would limit generality.
  • standard math The JPP is NP-complete and the GA is a reasonable heuristic for it.
    Cited from [1], [2]; the GA is used as a global optimization heuristic, not a polynomial-time solver.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Generic Hybrid Framework for 2D Visual Reconstruction." pith.science (2026). https://pith.science/paper/W5D2XSNP

@misc{pith2026250119325,
  author       = {Pith},
  title        = {Pith review of: A Generic Hybrid Framework for 2D Visual Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W5D2XSNP}},
  note         = {Machine review of arXiv:2501.19325}
}
read the original abstract

This paper presents a versatile hybrid framework for addressing 2D real-world reconstruction tasks formulated as jigsaw puzzle problems (JPPs) with square, non-overlapping pieces. Our approach integrates a deep learning (DL)-based compatibility measure (CM) model that evaluates pairs of puzzle pieces holistically, rather than focusing solely on their adjacent edges as traditionally done. This DL-based CM is paired with an optimized genetic algorithm (GA)-based solver, which iteratively searches for a global optimal arrangement using the pairwise CM scores of the puzzle pieces. Extensive experimental results highlight the framework's adaptability and robustness across multiple real-world domains. Notably, our unique hybrid methodology achieves state-of-the-art (SOTA) results in reconstructing Portuguese tile panels and large degraded puzzles with eroded boundaries.

Figures

Figures reproduced from arXiv: 2501.19325 by the authors.

Figure 1
Figure 1. Reconstruction of 460-tile Portuguese panel with unknown piece orientation and panel dimensions using our proposed system: [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Artificially eroded boundaries of puzzle with 150 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Piece augmentation through degradation and shift: [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (11 more)
Figure 5
Figure 5. Figure 5: Given the monochromatic nature of Portuguese tiles, [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: Sub-model architecture with input size P × 2P × C, where the piece size is P × P pixels and C is the number of channels; the architecture includes four convolutional layers with 3 × 3 kernels and ReLU activation function; max pooling is applied after the second and thi…
Figure 5
Figure 5. Figure 5: Our DLCM architecture, consisting of four sub-models: RGB-Net, [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Illustration of tile placement through chromosome pairing in different phases of the proposed genetic algorithm (GA); starting with a single piece, [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Top-i accuracy plots comparing traditional and DL-based CMs with the proposed DLCM for Type-1 (top) and Type-2 (bottom) Portuguese tile panels; DLCM achieves superior performance over all tested methods, including SqueezeNet- and GAN-based CMs. Specifically, we compare…
Figure 8
Figure 8. Figure 8: Evolutionary reconstruction of three Portuguese tile panels using the enhanced GA-based solver; each sub-image shows the intermediate solution [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Reconstruction of 70- and 150-piece puzzles with 14% erosion: (a), [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Reconstruction of a puzzle made up of nine shredded documents: (a) Scrambled puzzle with 765 strips of width 2.47mm, and (b) readable [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Compatibility measure for strip-cut pair. Left to right: Two strips, [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Compatibility map of puzzle tiles: (Top) 256-piece Portuguese tile [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: Illustration of in-painting results generated by the [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Pairwise Alignment & Compatibility for Arbitrarily Irregular Image Fragments

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A hybrid geometric-pictorial compatibility method achieves state-of-the-art neighborhood precision and recall on the RePAIR 2D fragments benchmark, with a new erosion-model dataset.

Reference graph

Works this paper leans on

71 extracted references · 66 canonical work pages · cited by 1 Pith paper

  1. [1]

    Solving the jigsaw puzzle problem in linear time,

    T. Altman, “Solving the jigsaw puzzle problem in linear time,” Applied Artificial Intelligence an International Journal , vol. 3, no. 4, pp. 453– 462, 1989

    work page 1989

  2. [2]

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

    E. Demaine and M. Demaine, “Jigsaw puzzles, edge matching, and polyomino packing: Connections and complexity,” Graphs and Com- binatorics, vol. 23, pp. 195–208, 2007

    work page 2007

  3. [3]

    The patch transform and its applications to image editing,

    T. Cho, M. Butman, S. Avidan, and W. Freeman, “The patch transform and its applications to image editing,” in IEEE Conference on Computer Vision and Pattern Recognition , 2008, pp. 1–8

    work page 2008

  4. [4]

    Automated assembly of shredded pieces from multiple photos,

    H. Liu, S. Cao, and S. Yan, “Automated assembly of shredded pieces from multiple photos,” IEEE Transactions on Multimedia, vol. 13, no. 5, pp. 1154–1162, 2011. 14 Portuguese Panel DLCM Score Map Fig. 12. Compatibility map of puzzle tiles: (Top) 256-piece Portuguese tile panel from the MNAz test set, and (bottom) compatibility score matrix; the i-th row re...

    work page 2011

  5. [5]

    Reconstructing strip-shredded documents using color as feature matching,

    M. Marques and C. Freitas, “Reconstructing strip-shredded documents using color as feature matching,” in Proceedings of the ACM Symposium on Applied Computing , 2009, pp. 893–894

    work page 2009

  6. [6]

    Reconstructing shredded doc- uments through feature matching,

    E. Justino, L. Oliveira, and C. Freitas, “Reconstructing shredded doc- uments through feature matching,” Forensic Science International, vol. 160, no. 2, pp. 140–147, 2006

    work page 2006

  7. [7]

    Semi-automatic assembly of real cross-cut shredded documents,

    A. Deever and A. Gallagher, “Semi-automatic assembly of real cross-cut shredded documents,” in Proceedings of the International Conference on Image Processing, 2012, pp. 233–236

    work page 2012

  8. [8]

    A system for high-volume acquisition and matching of fresco fragments: Reassem- bling Teheran wall paintings,

    B. Brown, C. Toler-Franklin, D. Nehab, M. Burns, D. Dobkin, A. Vla- chopoulos, C. Doumas, S. Rusinkiewicz, and T. Weyrich, “A system for high-volume acquisition and matching of fresco fragments: Reassem- bling Teheran wall paintings,” ACM Transactions on Graphics , vol. 27, no. 3, p. 84, 2008

    work page 2008

Show all 71 references
  1. [9]

    Computer-aided reconstruction and new matches in the forma urbis Romae,

    D. Koller and M. Levoy, “Computer-aided reconstruction and new matches in the forma urbis Romae,” Bullettino della Commissione Archeologica Comunale di Roma , pp. 103–125, 2006

    2006

  2. [10]

    Automatic reconstruction of ancient Portuguese tile panels,

    F. A. Andal ´o, G. Carneiro, G. Taubin, S. Goldenstein, and L. Velho, “Automatic reconstruction of ancient Portuguese tile panels,” Instituto Nacional de Matem ´atica Pura e Aplicada, Tech. Rep. A773/2016, 2016

    2016

  3. [11]

    A puzzle solver and its application in speech descrambling,

    Y . Zhao, M. Su, Z. Chou, and J. Lee, “A puzzle solver and its application in speech descrambling,” in Proceedings of the WSEAS International Conference on Computer Engineering and Applications , 2007, pp. 171– 176

    2007

  4. [12]

    On the security of block scrambling-based ETC systems against jigsaw puzzle solver attacks,

    T. Chuman, K. Kurihara, and H. Kiya, “On the security of block scrambling-based ETC systems against jigsaw puzzle solver attacks,” in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing , 2017, pp. 2157–2161

    2017

  5. [13]

    A genetic algorithm- based solver for very large jigsaw puzzles,

    D. Sholomon, O. David, and N. S. Netanyahu, “A genetic algorithm- based solver for very large jigsaw puzzles,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2013, pp. 1767–1774

    2013

  6. [14]

    Genetic algorithm- based solver for very large multiple jigsaw puzzles of unknown dimen- sions and piece orientation,

    D. Sholomon, E. O. David, and N. S. Netanyahu, “Genetic algorithm- based solver for very large multiple jigsaw puzzles of unknown dimen- sions and piece orientation,” in Proceedings of the ACM Conference on Genetic and Evolutionary Computation , 2014, pp. 1191–1198

    2014

  7. [15]

    A novel hybrid scheme using genetic algorithms and deep learning for the reconstruction of Portuguese tile panels,

    D. Rika, D. Sholomon, E. O. David, and N. S. Netanyahu, “A novel hybrid scheme using genetic algorithms and deep learning for the reconstruction of Portuguese tile panels,” in Proceedings of the ACM Conference on Genetic and Evolutionary Computation , 2019, pp. 1319– 1327

    2019

  8. [16]

    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, vol. EC-13, no. 2, pp. 118–127, 1964

    1964

  9. [17]

    Jigsaw puzzle matching using a boundary-centered polar encoding,

    G. M. Radack and N. I. Badler, “Jigsaw puzzle matching using a boundary-centered polar encoding,” Computer Graphics and Image Processing, vol. 19, no. 1, pp. 1–17, 1982

    1982

  10. [18]

    Solving jigsaw puzzles by computer,

    H. Wolfson, E. Schonberg, A. Kalvin, and Y . Lamdan, “Solving jigsaw puzzles by computer,” Annals of Operations Research , vol. 12, no. 1, pp. 51–64, 1988

    1988

  11. [19]

    On solving 2D and 3D puzzles using curve matching,

    W. Kong and B. B. Kimia, “On solving 2D and 3D puzzles using curve matching,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, vol. II, 2001, pp. 583–590

    2001

  12. [20]

    A global approach to automatic solution of jigsaw puzzles,

    D. Goldberg, C. Malon, and M. Bern, “A global approach to automatic solution of jigsaw puzzles,” in Proceedings of the Eighteenth ACM Annual Symposium on Computational Geometry , 2002, pp. 82–87

    2002

  13. [21]

    An automatic jigsaw puzzle solver,

    D. A. Kosiba, P. M. Devaux, S. Balasubramanian, T. L. Gandhi, and K. Kasturi, “An automatic jigsaw puzzle solver,” in Proceedings of the 12th IAPR International Conference on Pattern Recognition , vol. 1, 1994, pp. 616–618

    1994

  14. [22]

    Jigsaw puzzle solver using shape and color,

    M. G. Chung, M. M. Fleck, and D. A. Forsyth, “Jigsaw puzzle solver using shape and color,” in Proceedings of the Fourth IEEE International Conference Signal Processing, vol. 2, 1998, pp. 877–880

    1998

  15. [23]

    A shape and image merging technique to solve jigsaw puzzles,

    F.-H. Yao and G.-F. Shao, “A shape and image merging technique to solve jigsaw puzzles,” Pattern Recognition Letters, vol. 24, no. 12, pp. 1819–1835, 2003

    2003

  16. [24]

    A new technique for solving a jigsaw puzzle,

    M. Makridis and N. Papamarkos, “A new technique for solving a jigsaw puzzle,” in Proceedings of the International Conference on Image Processing, 2006, pp. 2001–2004

    2006

  17. [25]

    A texture based matching approach for automated assembly of puzzles,

    M. S. Sagiroglu and A. Erc ¸il, “A texture based matching approach for automated assembly of puzzles,” in Proceedings of the 18th IEEE International Conference on Pattern Recognition, vol. 3, 2006, pp. 1036– 1041

    2006

  18. [26]

    Solving jigsaw puzzles using image features,

    T. R. Nielsen, P. Drewsen, and K. Hansen, “Solving jigsaw puzzles using image features,” Pattern Recognition Letters, vol. 29, no. 14, pp. 1924–1933, 2008. 15

    1924

  19. [27]

    A probabilistic image jigsaw puzzle solver,

    T. Cho, S. Avidan, and W. Freeman, “A probabilistic image jigsaw puzzle solver,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2010, pp. 183–190

    2010

  20. [28]

    Particle filter with state permutations for solving image jigsaw puzzles,

    X. Yang, N. Adluru, and L. J. Latecki, “Particle filter with state permutations for solving image jigsaw puzzles,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2011, pp. 2873–2880

    2011

  21. [29]

    A fully automated greedy square jigsaw puzzle solver,

    D. Pomeranz, M. Shemesh, and O. Ben-Shahar, “A fully automated greedy square jigsaw puzzle solver,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2011, pp. 9– 16

    2011

  22. [30]

    Jigsaw puzzles with pieces of unknown orientation,

    A. C. Gallagher, “Jigsaw puzzles with pieces of unknown orientation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2012, pp. 382–389

    2012

  23. [31]

    A generalized genetic algorithm-based solver for very large jigsaw puzzles of complex types,

    D. Sholomon, O. David, and N. S. Netanyahu, “A generalized genetic algorithm-based solver for very large jigsaw puzzles of complex types,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2014, pp. 2839–2845

    2014

  24. [32]

    Solving square jigsaw puzzles with loop constraints,

    K. Son, J. Hays, and D. B. Cooper, “Solving square jigsaw puzzles with loop constraints,” in Proceedings of the European Conference on Computer Vision. Springer, 2014, pp. 32–46

    2014

  25. [33]

    Solving small-piece jigsaw puzzles by growing consensus,

    S. Kilho, D. Moreno, J. Hays, and D. Cooper, “Solving small-piece jigsaw puzzles by growing consensus,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2016, pp. 1193–1201

    2016

  26. [34]

    Solving square jigsaw puzzle by hierarchical loop constraints,

    ——, “Solving square jigsaw puzzle by hierarchical loop constraints,” in IEEE Transactions on Pattern Analysis and Machine Intelligence , 2019, pp. 2222–2235

    2019

  27. [35]

    Solving jigsaw puzzles with linear programming,

    R. Yu, C. Russell, and L. Agapito, “Solving jigsaw puzzles with linear programming,” 2015. [Online]. Available: http://arxiv.org/abs/ 1511.04472

    2015 arXiv

  28. [36]

    Solving jigsaw puzzles by the graph connection laplacian,

    V . Huroyan, G. Lerman, and H. Wu, “Solving jigsaw puzzles by the graph connection laplacian,” CoRR, 2018. [Online]. Available: http://arxiv.org/abs/1811.03188

    2018 arXiv

  29. [37]

    Solving multiple square jigsaw puzzles with missing pieces,

    G. Paikin and A. Tal, “Solving multiple square jigsaw puzzles with missing pieces,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2015, pp. 4832–4839

    2015

  30. [38]

    PSQP – puzzle solving by quadratic programming,

    F. A. Andal ´o, G. G. Taubin, and S. Goldenstein, “PSQP – puzzle solving by quadratic programming,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 2, pp. 385–396, 2017

    2017

  31. [39]

    DNN-buddies: A deep neural network-based estimation metric for the jigsaw puzzle problem,

    D. Sholomon, O. E. David, and N. S. Netanyahu, “DNN-buddies: A deep neural network-based estimation metric for the jigsaw puzzle problem,” in Proceedings of the International Conference on Artificial Neural Networks. Springer International Publishing, 2016, pp. 170–178

    2016

  32. [40]

    A deep learning-based compatibility score for reconstruction of strip-shredded text documents,

    T. M. Paixao, R. F. Berriel, M. C. Boeres, C. Badue, A. F. De Souza, and T. Oliveira-Santos, “A deep learning-based compatibility score for reconstruction of strip-shredded text documents,” in Proceedings of the IEEE 31st SIBGRAPI Conference on Graphics, Patterns and Images , ...

    2018

  33. [41]

    SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size,

    N. I. Forrest, H. Song, W. Matthew, A. Khalid, J. William, and K. Kurt, “SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size,” 2015

    2015

  34. [42]

    Mo- bileNetV2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. Chen, “Mo- bileNetV2: Inverted residuals and linear bottlenecks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2018, pp. 4510–4520

    2018

  35. [43]

    Fast(er) reconstruction of shredded text documents via self-supervised deep asymmetric metric learning,

    T. M. Paix ˜ao, R. F. Berriel, M. C. S. Boeres, A. L. Koerich, C. Badue, A. F. D. Souza, and T. Oliveira-Santos, “Fast(er) reconstruction of shredded text documents via self-supervised deep asymmetric metric learning,” in Proceedings of the IEEE Conference on Computer Vision a...

    2020

  36. [44]

    Solving jigsaw puzzles with eroded boundaries,

    D. Bridger, D. Danon, and A. Tal, “Solving jigsaw puzzles with eroded boundaries,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2020, pp. 3526–3535

    2020

  37. [45]

    Generative adversarial networks,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial networks,” arXiv:1406.2661v1, 2014

    2014 arXiv

  38. [46]

    Relaxation labeling meets GANs: Solving jigsaw puzzles with missing borders,

    M. Khoroshiltseva, A. Traviglia, M. Pelillo, and S. Vascon, “Relaxation labeling meets GANs: Solving jigsaw puzzles with missing borders,” in Proceedings of the International Conference on Image Analysis and Processing. Springer, LNCS 13233, 2022, pp. 27–38

    2022

  39. [47]

    Jigsaw puzzle solving as a consistent labeling problem,

    M. Khoroshiltseva, B. Vardi, A. Torcinovich, A. Traviglia, O. Ben- Shahar, and M. Pelillo, “Jigsaw puzzle solving as a consistent labeling problem,” in Proceedings of the Computer Analysis of Images and Patterns, 2021, pp. 392–402

    2021

  40. [48]

    Solving jigsaw puzzle of large eroded gaps using puzzlet discriminant network,

    X. Song, X. Yang, J. Ren, R. Bai, and X. Jiang, “Solving jigsaw puzzle of large eroded gaps using puzzlet discriminant network,” inProceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2023, pp. 1–5

    2023

  41. [49]

    Unsupervised visual repre- sentation learning by context prediction,

    C. Doersch, A. Gupta, and A. A. Efros, “Unsupervised visual repre- sentation learning by context prediction,” in Proceedings of the IEEE International Conference on Computer Vision , 2015, pp. 1422–1430

    2015

  42. [50]

    Unsupervised learning of visual represen- tations by solving jigsaw puzzles,

    M. Noroozi and P. Favaro, “Unsupervised learning of visual represen- tations by solving jigsaw puzzles,” arXiv:1603.09246, 2016

    2016 arXiv

  43. [51]

    Neural combinatorial optimization for solving jigsaw puzzles: A step towards unsupervised pre-training,

    L. Dery, R. Mengistu, and O. Awe, “Neural combinatorial optimization for solving jigsaw puzzles: A step towards unsupervised pre-training,” http://cs231n.stanford.edu/reports/2017/pdfs/110.pdf, 2017

    2017

  44. [52]

    DeepPermNet: Visual permutation learning,

    R. Santa Cruze, B. Fernando, A. Cherian, and S. Gould, “DeepPermNet: Visual permutation learning,” arXiv:1704.02729v1, 2017

    2017 arXiv

  45. [53]

    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

  46. [54]

    JigsawGAN: Aux- iliary learning for solving jigsaw puzzles with generative adversarial networks,

    R. Li, S. Liu, G. Wang, G. Liu, and B. Zeng, “JigsawGAN: Aux- iliary learning for solving jigsaw puzzles with generative adversarial networks,” IEEE Transactions on Image Processing , vol. 31, pp. 513– 524, 2021

    2021

  47. [55]

    GANzzle: Reframing jigsaw puzzle solving as a retrieval task using a generative mental image,

    D. Talon, A. D. Bue, and S. James, “GANzzle: Reframing jigsaw puzzle solving as a retrieval task using a generative mental image,” in Proceedings of the International Conference on Image Processing , 2022, pp. 4083–4087

    2022

  48. [56]

    Learning deep graph matching with channel-independent embedding and hungarian attention,

    T. Yu, R. Wang, J. Yan, and B. Li, “Learning deep graph matching with channel-independent embedding and hungarian attention,” in Proceed- ings of the International Conference on Learning Representations, 2020, pp. 1–15

    2020

  49. [57]

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

    X. Song, J. Jin, C. Yao, S. Wang, J. Ren, and R. Bai, “Siamese- discriminant deep reinforcement learning for solving jigsaw puzzles with large eroded gaps,” in Proceedings of the AAAI Conference on Artificial Intelligence, 2023, p. 2303–2311

    2023

  50. [58]

    Solving masked jigsaw puzzles with diffusion vision transformers,

    J. Liu, W. Teshome, S. Ghimire, M. Sznaier, and O. Camps, “Solving masked jigsaw puzzles with diffusion vision transformers,” 2024

    2024

  51. [59]

    de Matos and Museu Nacional do Azulejo, Azulejos: Masterpieces of the National Tile Museum of Lisbon

    M. de Matos and Museu Nacional do Azulejo, Azulejos: Masterpieces of the National Tile Museum of Lisbon . Chandeigne, 2011

    2011

  52. [60]

    A. N. Pais, Curator of Museu Nacional do Azulejo, personal communi- cation, 2018

    2018

  53. [61]

    J. T. Fonseca, Montagem Autom ´atica de Pain ´eis de Azulejos . M.Sc. Thesis, Instituto Superior T ´ecnico, Universidade T ´ecnica de Lisboa, 2012

    2012

  54. [62]

    Robust solvers for square jigsaw puzzles,

    D. Mondal, Y . Wang, and S. Durocher, “Robust solvers for square jigsaw puzzles,” in Proceedings of the International Conference on Computer and Robot Vision , 2013, pp. 249–256

    2013

  55. [63]

    An investigation into automated shredded document recon- struction using heuristic search algorithms,

    A. Skeoch, “An investigation into automated shredded document recon- struction using heuristic search algorithms,” Ph.D. dissertation, Univer- sity of Bath, UK, 2006

    2006

  56. [64]

    Reconstructing shredded documents,

    R. Ranca, “Reconstructing shredded documents,” School of Informatics, University of Edinburgh, UK, Tech. Rep., 2013

    2013

  57. [65]

    J. H. Holland, Adaptation in Natural and Artificial Systems. Ann Arbor, MI: University of Michigan Press, 1975

    1975

  58. [66]

    Learning represen- tations by back-propagating errors,

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning represen- tations by back-propagating errors,” Nature, vol. 323, no. 6088, p. 533, 1986

    1986

  59. [67]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” CoRR, vol. abs/1412.6980, 2014. [Online]. Available: http://arxiv.org/abs/1412.6980

    2014 arXiv

  60. [68]

    A biologically inspired algorithm for the recovery of shading and reflectance images,

    A. Olmos and F. A. A. Kingdom, “A biologically inspired algorithm for the recovery of shading and reflectance images,” in Perception, no. 33, 2004, pp. 1463–1473

    2004

  61. [69]

    NTIRE 2017 challenge on single image super-resolution: Dataset and study,

    E. Agustsson and R. Timofte, “NTIRE 2017 challenge on single image super-resolution: Dataset and study,” in Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition Workshops , July 2017

    2017

  62. [70]

    TEN: Twin embedding networks for the jigsaw puzzle problem with eroded boundaries,

    D. Rika, D. Sholomon, E. O. David, and N. S. Netanyahu, “TEN: Twin embedding networks for the jigsaw puzzle problem with eroded boundaries,” arXiv:2203.06488, 2022

    2022 arXiv

  63. [71]

    Edge2Vec: A high quality embedding for the jisaw puzzle problem,

    ——, “Edge2Vec: A high quality embedding for the jisaw puzzle problem,” arXiv:2211.07771, 2022

    2022 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.