Pith. sign in

REVIEW 2 major objections 6 minor 48 references

Classical versus Deep Mirror-Symmetry Scoring: A Benchmark of Thirteen Methods

T0 review · 2 major / 6 minor · reviewed 2026-07-10 · grok-4.5

Pith's one-line read A tuned classical HOG nearly matches frozen deep networks at measuring mirror symmetry, at roughly 300× lower CPU cost.

desk verdict Solid first head-to-head of symmetry scorers: frozen deep features barely beat tuned HOG, with a clean mid-scale story and open code. read the letter →

arxiv 2607.08379 v1 pith:CJCE6ZTP submitted 2026-07-09 cs.CV eess.IV

classification cs.CVeess.IV
keywords symmetryscoringmirrorreflectionaxisdiscriminationhistogramoforientedgradientsdeepfeaturesbenchmark
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 asks a practical question that the field has never answered head-to-head: among existing methods, which ones actually measure how mirror-symmetric an image is about a given axis, and at what cost? The authors put thirteen scorers—from raw pixel correlation through oriented filters to frozen deep-network readouts—on the same reflection-exact discrimination task across nine public datasets, scoring each method by how well it ranks a true axis above controlled wrong ones. Deep mid-stage features lead, but a carefully tuned histogram-of-oriented-gradients descriptor trails by only a small skill margin, is statistically inseparable from a classical CNN-filter measure, and runs hundreds of times faster on CPU. Two independent ablations (network stage and HOG cell size) converge on the same mechanism: discrimination lives in mid-scale, unsigned, oriented structure. The result is an actionable cost–skill map and an open toolkit, while leaving open whether scorers trained specifically for the task could widen the gap.

What carries the argument

A single representation–comparison–aggregation template, s(I) = A(C(T(I), T(MI))), evaluated under a reflection-exact crop protocol with chance-anchored discrimination skill (2·AUC − 1) and paired-bootstrap significance testing. The representation T is the decisive variable; mid-scale unsigned oriented structure is the operative signal.

What would settle it

Train a deep scorer end-to-end for axis discrimination on the same protocol and show a large, significant skill gain over HOG that survives multi-axis and domain-transfer checks; or re-run the benchmark on native medical volumes and show domain asymmetry indices overtake HOG and frozen deep readouts.

Watch

Extended reading notes

Core claim

Among existing methods, frozen deep features offer little over a tuned classical HOG descriptor for measuring mirror symmetry. DeepFeat leads at mean single-axis skill 0.83; HOG trails by only 0.031 (significant but small), is not statistically separable from AlexNet-C2 at 0.81, and runs about 300× faster on CPU. Discrimination concentrates in mid-scale unsigned oriented features: deep backbones peak at a low or mid stage, HOG peaks at a mid cell size, and columnar Vision Transformers without spatial hierarchy fall behind.

Load-bearing premise

That ranking the true axis above controlled wrong axes within an image is the right test of what applications need, rather than a calibrated score of how symmetric an image is that can be compared across images.

Editorial extensions

If this is right

  • For most practical scoring, a tuned mid-cell HOG is the default: near-top skill at ~0.5 ms per crop on CPU.
  • Deep frozen readouts are justified only when the final ~0.03 skill justifies two orders of magnitude more compute, or as part of a complementary ensemble with HOG.
  • Detection pipelines that wrap a scorer around candidate axes can substitute a cheap mid-scale oriented scorer without large accuracy loss.
  • Architecture choice for frozen scoring matters less than stage: low/mid hierarchical features win; single-resolution tokenizers underperform.
  • Whether task-trained deep scorers can beat this classical ceiling remains the open next experiment.

Reading between the lines

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

  • The mid-scale oriented principle may transfer to other bilateral-structure tasks (figure-ground, aesthetic balance) that currently default to deep features without checking classical baselines.
  • A natural extension is to drive axis search with HOG first, then refine with a mid-stage deep readout only on hard candidates—hybrid detection that the paper leaves open.
  • If annotation noise at the finest perturbations is the true floor, future benchmarks need inter-rater axis spread, not only consensus labels.
  • Medical and industrial domains that already use domain-specific indices may still prefer those methods; the natural-image ranking should not be assumed universal without re-benchmarking.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper presents the first head-to-head benchmark of thirteen mirror-symmetry scoring methods (nine from the literature, four introduced or adapted here), spanning intensity, binary-shape, gradient, spectral, oriented-filter, patch-correspondence, and frozen deep-feature families. All methods are cast as instances of a common representation–comparison–aggregation template and evaluated under a reflection-exact crop protocol with chance-anchored discrimination skill (2·AUC−1), paired bootstrap significance testing, fixed-extent and multi-axis controls, and local-versus-global negative stratification across four single-axis and five multi-axis datasets (~3950 axis units). The central empirical claim is that a tuned classical HOG descriptor trails the best frozen deep readout (DeepFeat) by only +0.031 skill (Holm-adjusted p=0.044), is not statistically separable from AlexNet-C2, and runs ~300× faster on CPU; ablations locate discrimination in mid-scale unsigned oriented features. The scorers and harness are released as the open toolkit imgsym.

Significance. If the ranking and cost–skill map hold, the result is immediately actionable for any application that consumes a scalar symmetry score about a given axis (aesthetics, medical asymmetry indices, developmental biology). The mid-scale oriented-feature principle is of independent scientific interest and is supported by two independent routes (network stage and HOG cell size) plus a columnar-ViT control. Strengths that raise the contribution above a routine bake-off include: the reflection-exact extraction with fixed-extent control, chance-anchored within-image skill with paired bootstrap and Holm correction, explicit multi-axis generalization including DENDI, transparent quantification of protocol artifacts (direction inference, crop geometry, axis-count cap), and the full open release of scorers, harness, and analysis scripts. The paper also states its scope boundary clearly: relative within-image ranking, not calibrated cross-image degree of symmetry, and frozen rather than task-trained deep scorers.

major comments (2)
  1. §5.1–5.2 and Table 4: On the multi-axis protocol, absolute skill levels fall substantially (DeepFeat 0.83→0.70; HOG 0.80→0.65), and on DENDI the best methods reach only ~0.5. The relative ranking and the HOG–DeepFeat gap are preserved, but the practical claim that HOG is a near-substitute for frozen deep features should be caveated more explicitly for hard multi-axis / in-the-wild settings where absolute discrimination remains modest for every method. A short quantitative statement of residual error rates (or of how often the true axis is not ranked first) would make the cost–skill recommendation more usable.
  2. §4 and §5.3: Exactly two methods (HOG cell/bins/sign and DeepFeat backbone/stage) were selected on single-axis ablation subsets of the same public data used for the leaderboard. The paper shows held-out generalization and treats both methods symmetrically, which is good practice, but the abstract and conclusion phrasing “a classical HOG descriptor” / “among existing methods” slightly understates that the competitive classical entry is a re-engineered, benchmark-tuned formulation rather than an off-the-shelf literature default. A one-sentence clarification in the abstract and §7 that the HOG result is for the tuned configuration introduced here would keep the claim precise without weakening it.
minor comments (6)
  1. §3.5: The direction-inference step (inferring score polarity from evaluation data) is quantified (~0.8 SE inflation for near-chance methods) and is irrelevant for the leaders, but it would be cleaner to fix polarity a priori from each method’s definition and report any residual flips as a diagnostic.
  2. Figure 2 / Table 4: Whiskers show per-dataset range rather than bootstrap CIs; adding the paired-bootstrap intervals (or a small table of them) next to the means would make the non-separation claims easier to read without consulting the text.
  3. §5.3 / Appendix A: The hard-subset backbone ordering is correctly flagged as unstable (n=40, overlapping intervals). Consider moving the full per-backbone table to the main text or a short supplementary figure so readers do not over-read the point estimates.
  4. §5.4.1: CPU-only timings are appropriate for the classical methods; a single GPU number for DeepFeat / AlexNet-C2 (even as a footnote) would complete the cost map for readers who will run the deep scorers on accelerators.
  5. §2.2 / Table 1: The exclusion of SSIM-style quality indices, per-pixel symmetry maps, and pure shape scores is justified in the text; a one-line note in the table caption would make the pool boundary self-contained.
  6. Typos / polish: “reflection” vs “reflection” is consistent in the PDF but mixed in some arXiv metadata; “Somers’ D (Gini)” is fine but a parenthetical pointer to the rank-correlation literature would help non-stats readers. The Acknowledgments AI-use statement is clear and appropriate.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical benchmark with disclosed hyperparameter selection on subsets; no load-bearing circular derivation.

  1. fitted input called prediction [Section 4 (methods) and Section 5.3 / Limitations 6.1]
    "Exactly two methods are tuned on benchmark data: HOG’s descriptor grid and DeepFeat’s backbone and stage were selected on the single-axis ablation subsets of Section 5.3 rather than taken from a source... Both winning configurations (HOG’s cell, bin count, and sign, and DeepFeat’s stage) were selected on single-axis benchmark subsets"

    The two leading configurations were chosen by maximizing skill on subsets of the same evaluation data later used for the leaderboard. This is ordinary hyperparameter selection, not a first-principles prediction, and the paper reports held-out checks showing the ranking is preserved; still, the reported peak skills for HOG and DeepFeat are not fully independent of the data they are scored on.

full rationale

This is a head-to-head empirical benchmark of 13 existing or newly formalized scorers on public datasets under a fixed reflection-exact protocol and chance-anchored within-image rank metric (skill = 2·AUC−1). The central ranking and cost–skill claims are produced by evaluating methods on held-out axis units with paired bootstrap; they do not reduce by construction to any fitted parameter or self-citation. The only mild circularity-adjacent practice is ordinary hyperparameter selection: HOG cell/bins/sign and DeepFeat backbone/stage were chosen on single-axis ablation subsets (Section 5.3), then reported on the full leaderboard. The paper discloses this, applies identical treatment to both winners, and checks generalization (HOG 0.81 outside its tuning pool; on PIX2PER-nat outside both pools DeepFeat/AlexNet-C2 still lead). Direction inference from evaluation scores slightly inflates near-chance methods (~0.8 SE) but is irrelevant for the leaders and is quantified. No self-definitional loop, no uniqueness theorem imported from the author, no ansatz smuggled via self-citation, and no renaming of a known result as a first-principles derivation. Score 1 reflects only the disclosed subset tuning; the derivation chain itself is self-contained against external data.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a well-specified empirical protocol rather than free parameters or invented entities. The main free choices are the HOG descriptor grid and DeepFeat backbone/stage (selected on ablation subsets) and the perturbation-magnitude cut used for the main leaderboard. Domain assumptions are standard for computer-vision benchmarks: public axis annotations are treated as ground truth, and within-image ranking is treated as the right proxy for scoring quality. No new physical or mathematical entities are postulated.

free parameters (4)
  • HOG cell size / bins / signedness = cell 16, 18 bins, unsigned
    Selected on the single-axis ablation pool (cell 16, 18 bins, unsigned); worth +0.058 skill over the common default. Held-out checks are reported, but the value is data-chosen.
  • DeepFeat backbone and stage = MambaOut-Base, stage 1
    MambaOut-Base stage-1 chosen as default after sweeping 16 backbones and stages on easy/hard subsets. Architecture choice affects absolute skill only mildly; stage choice is load-bearing for the mid-scale claim.
  • Main-result perturbation magnitude cut = ≥3% / ≥3°
    Main leaderboard uses only coarser half of the ladder (shifts ≥3%, rotations ≥3°) because finest perturbations are near-duplicates no method can separate. Full-range curves are reported separately.
  • Multi-axis axis-count cap K=10 and 256 px crop cap = K=10, 256 px
    Tractability choices for dense DENDI images; control experiments claim ranking unchanged and skill shifts ≤0.05.
assumptions (4)
  • domain assumption Within-image ranking of a true annotated axis above controlled wrong axes (chance-anchored skill = 2·AUC−1) is a valid measure of symmetry-scoring quality for the applications that consume scalar scores.
    Stated in Sections 3.1 and 3.5; the Discussion explicitly notes this measures relative discrimination, not a calibrated cross-image degree of symmetry.
  • domain assumption Public competition and perception-dataset axis annotations are sufficiently accurate ground truth for discrimination (with the caveat that finest perturbations may be confounded by annotation noise).
    Datasets listed in Section 3.2; annotation-noise floor acknowledged in Limitations.
  • domain assumption Frozen ImageNet-pretrained features and classical descriptors can be compared fairly when both receive the same reflection-exact crops and no task-specific fine-tuning.
    Section 4.7 and the scope statement that no scorer is trained for the task.
  • standard math Standard statistical tools (paired bootstrap, Holm correction, Somers’ D / Gini transform of AUC) correctly quantify significance of skill differences.
    Section 3.5; medical precedent of Hogeweg et al. cited for AUC-based discrimination.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Classical versus Deep Mirror-Symmetry Scoring: A Benchmark of Thirteen Methods." pith.science (2026). https://pith.science/paper/CJCE6ZTP

@misc{pith2026260708379,
  author       = {Pith},
  title        = {Pith review of: Classical versus Deep Mirror-Symmetry Scoring: A Benchmark of Thirteen Methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CJCE6ZTP}},
  note         = {Machine review of arXiv:2607.08379}
}
read the original abstract

Quantifying how mirror-symmetric an image is about a given axis (symmetry scoring) underpins applications from visual aesthetics to medical imaging, yet proposed scoring methods have never been compared on a common, statistically grounded protocol. We benchmark 13 scoring methods (nine collected from literature, four introduced here) spanning from classical features to frozen deep features, across four single-axis and five multi-axis datasets under a reflection-exact protocol with a chance-anchored, significance-tested discrimination skill. Deep backbones perform best on single-axis and harder multi-axis protocols. However, a classical histogram-of-oriented-gradients (HOG) descriptor trails the best frozen-network readout by a small (but significant) margin, is not statistically separable from the runner-up (a CNN-filter measure), and runs ~300x faster on CPU. Our results show that discrimination concentrates in mid-scale oriented features, where deep backbones peak at a low or mid stage, and HOG peaks at a mid cell size. Among existing methods, frozen deep features thus offer little over a tuned classical descriptor for measuring symmetry; whether task-trained deep scorers can do better remains open. We release the scorers and harness in imgsym, an open toolkit for image symmetry detection and measurement.

Figures

Figures reproduced from arXiv: 2607.08379 by the authors.

Figure 1
Figure 1. The discrimination task in practice. Top: an input image with its true axis and one negative axis of each kind (Section 3.4), and the canonical crop pair the scorer receives: the whole crop I and its mirror M I (Section 3.3). Bottom: four benchmark crops spanning the outcome range; thin lines mark each crop’s 12 standard perturbed negatives (orange = shifted, blue = rotated) around the true axis, and the verdicts gi… view at source ↗
Figure 2
Figure 2. Mean discrimination skill of all 13 methods under the single-axis (solid) and multi-axis (hatched) protocols; whiskers span the per-dataset range. 5.1.1. Sensitivity to Perturbation Magnitude Sweeping the perturbation magnitude shows where the ranking is decided ( [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Discrimination skill vs. perturbation magnitude (left: perpendicular shift, % of size; right: rotation, degrees). All 13 methods are shown (gray = context); the three leaders, Gabor (mid-pack), and EROS (floor) are highlighted. 5.2. Is the Ranking an Artifact of the Protocol? Three checks say no. Fixed-extent crops leave the ranking essentially unchanged (Sec￾tion 3.3); the per-axis protocol below preserves it on fi… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Discrimination skill by backbone stage across 16 backbones on the easy (NYU + sym￾Comp17, left) and hard (PIX2PER-art, right) subsets; each line ends at its backbone’s last stage (back￾bones expose three to five stages). Hierarchical networks are gray, MambaOut-Base (D…
Figure 5
Figure 5. Figure 5: HOG ablations. Left: skill vs. cell size (signed and unsigned gradients) at the default 64-px window. Right: cell size swept against descriptor window size (unsigned, 18 bins). 5.4. What Does Scoring Cost in Practice? 5.4.1. Speed versus Accuracy Runtime is measured by…
Figure 6
Figure 6. Figure 6: Speed versus accuracy: median CPU scoring time per crop in milliseconds (log scale; whiskers = interquartile range) against mean single-axis discrimination skill. 5.4.2. Complementarity and Ensembles The top methods make different mistakes: the per-image success correl…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    Detection of Interest Points Using Symmetry

    Reisfeld, D.; Wolfson, H.; Yeshurun, Y . Detection of Interest Points Using Symmetry . In Proceedings of the [1990] Proceedings Third International Conference on Computer Vision, Osaka, Japan, 1990; pp. 62–65. https://doi.org/10.1109/ICCV .1990.139494

  2. [2]

    Quantification of Local Symmetry: Application to Tex- ture Discrimination

    Bonneh, Y .; Reisfeld, D.; Yeshurun, Y . Quantification of Local Symmetry: Application to Tex- ture Discrimination. In Human Symmetry Perception and Its Computational Analysis ; Psychology Press, 2003; pp. 310–325

  3. [3]

    Symmetry reCAPTCHA

    Funk, C.; Liu, Y . Symmetry reCAPTCHA. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las V egas, NV , USA, 2016; pp. 5165–5174. https://doi.org/10.1109/CVPR.2016.558

  4. [4]

    Accurate Robust Symmetry Estimation

    Smith, S.; Jenkinson, M. Accurate Robust Symmetry Estimation. In Medical Image Computing and Computer-Assisted Intervention – MICCAI’99 ; Goos, G.; Hartmanis, J.; Van Leeuwen, J.; Tay- lor, C.; Colchester, A., Eds.; Springer Berlin Heidelberg: Berlin, Heidelberg, 1999; V ol. 1679, pp. 308–317. https://doi.org/10.1007/10704282_34

  5. [5]

    Fast and Effective Quantification of Symmetry in Medical Images for Pathology Detection: Application to Chest Radiography

    Hogeweg, L.; Sánchez, C.I.; Maduskar, P .; Philipsen, R.H.; Van Ginneken, B. Fast and Effective Quantification of Symmetry in Medical Images for Pathology Detection: Application to Chest Radiography . Med. Phys. 2017, 44, 2242–2256. https://doi.org/10.1002/mp.12127

  6. [6]

    A Method for Quantifying Rotational Symmetry

    Frey , F.M.; Robertson, A.; Bukoski, M. A Method for Quantifying Rotational Symmetry . New Phytologist 2007, 175, 785–791. https://doi.org/10.1111/j.1469-8137.2007.02146.x. 20 of 22

  7. [7]

    Quantifying Visual Aesthetics Based on Processing Fluency Theory: Four Algorithmic Measures for Antecedents of Aesthetic Preferences

    Mayer, S.; Landwehr, J.R. Quantifying Visual Aesthetics Based on Processing Fluency Theory: Four Algorithmic Measures for Antecedents of Aesthetic Preferences. Psychology of Aesthetics, Creativity, and the Arts 2018, 12, 399–431. https://doi.org/10.1037/aca0000187

  8. [8]

    Predicting Perceived Visual Complexity of Abstract Patterns Using Com- putational Measures: The Influence of Mirror Symmetry on Complexity Perception

    Gartus, A.; Leder, H. Predicting Perceived Visual Complexity of Abstract Patterns Using Com- putational Measures: The Influence of Mirror Symmetry on Complexity Perception. PLoS ONE 2017, 12, e0185276. https://doi.org/10.1371/journal.pone.0185276

Show all 48 references
  1. [9]

    Combining Appearance and Gradient Information for Image Symmetry Detection

    Gnutti, A.; Guerrini, F.; Leonardi, R. Combining Appearance and Gradient Information for Image Symmetry Detection. IEEE T rans. on Image Process. 2021, 30, 5708–5723. https://doi. org/10.1109/TIP .2021.3085202

  2. [10]

    A New Symmetry Measure Based on Gabor Filters

    Shaker, F.; Monadjemi, A. A New Symmetry Measure Based on Gabor Filters. In Proceedings of the 2015 23rd Iranian Conference on Electrical Engineering, Tehran, Iran, 2015; pp. 705–710. https://doi.org/10.1109/IranianCEE.2015.7146305

  3. [11]

    Symmetry Analysis for 2D Images by Using DCT Coefficients

    Gunlu, G.; Bilge, H.S. Symmetry Analysis for 2D Images by Using DCT Coefficients. In Proceedings of the 2009 Fifth International Conference on Soft Computing, Computing with Words and Perceptions in System Analysis, Decision and Control, Famagusta, Cyprus, 2009; pp. 1–4. https:...

  4. [12]

    Using Convolutional Neural Network Filters to Measure Left-Right Mirror Symmetry in Images

    Brachmann, A.; Redies, C. Using Convolutional Neural Network Filters to Measure Left-Right Mirror Symmetry in Images. Symmetry 2016, 8, 144. https://doi.org/10.3390/sym8120144

  5. [13]

    Symmetry Detection from RealWorld Images Competition 2013: Summary and Results

    Liu, J.; Slota, G.; Zheng, G.; Wu, Z.; Park, M.; Lee, S.; Rauschert, I.; Liu, Y . Symmetry Detection from RealWorld Images Competition 2013: Summary and Results. In Proceedings of the 2013 IEEE Conference on Computer Vision and Pattern Recognition Workshops, OR, USA, 2013; pp....

  6. [14]

    2017 ICCV Challenge: Detecting Symmetry in the Wild

    Funk, C.; Lee, S.; Oswald, M.R.; Tsogkas, S.; Shen, W.; Cohen, A.; Dickinson, S.; Liu, Y . 2017 ICCV Challenge: Detecting Symmetry in the Wild. In Proceedings of the 2017 IEEE Inter- national Conference on Computer Vision Workshops (ICCVW), V enice, 2017; pp. 1692–1701. https:...

  7. [15]

    Detecting Symmetry and Symmetric Constellations of Features

    Loy , G.; Eklundh, J.O. Detecting Symmetry and Symmetric Constellations of Features. In Com- puter Vision – ECCV 2006; Leonardis, A.; Bischof, H.; Pinz, A., Eds.; Springer Berlin Heidelberg: Berlin, Heidelberg, 2006; V ol. 3952, pp. 508–521. https://doi.org/10.1007/11744047_39

  8. [16]

    Bilateral Symmetry Detection via Symmetry-Growing

    Cho, M.; Lee, K.M. Bilateral Symmetry Detection via Symmetry-Growing

  9. [17]

    An Adaptive Symmetry Detection Algorithm Based on Local Features

    Cai, D.; Li, P .; Su, F.; Zhao, Z. An Adaptive Symmetry Detection Algorithm Based on Local Features. In Proceedings of the 2014 IEEE Visual Communications and Image Processing Conference, Valletta, Malta, 2014; pp. 478–481. https://doi.org/10.1109/VCIP .2014.7051610

  10. [18]

    Reflection Symmetry Detection Using Locally Affine Invariant Edge Correspondence

    Wang, Z.; Tang, Z.; Zhang, X. Reflection Symmetry Detection Using Locally Affine Invariant Edge Correspondence. IEEE T rans. on Image Process. 2015, 24, 1297–1301. https://doi.org/10 .1109/TIP .2015.2393060

  11. [19]

    Fast Reflectional Symmetry Detection Using Orientation Histograms

    Sun, C.; Si, D. Fast Reflectional Symmetry Detection Using Orientation Histograms. Real-Time Imaging 1999, 5, 63–74. https://doi.org/10.1006/rtim.1998.0135

  12. [21]

    Wavelet-Based Reflection Sym- metry Detection via Textural and Color Histograms

    Elawady , M.; Ducottet, C.; Alata, O.; Barat, C.; Colantoni, P . Wavelet-Based Reflection Sym- metry Detection via Textural and Color Histograms. In Proceedings of the 2017 IEEE Inter- national Conference on Computer Vision Workshops (ICCVW), V enice, 2017; pp. 1725–1733. https...

  13. [22]

    Image Symmetries: The Right Balance between Evenness and Perception

    Guerrini, F.; Gnutti, A.; Leonardi, R. Image Symmetries: The Right Balance between Evenness and Perception. In Proceedings of the 2017 International Conference on Systems, Signals and Image Processing (IWSSIP), Pozna ´ n, Poland, 2017; pp. 1–5. https://doi.org/10.1109/IWSSIP ....

  14. [23]

    Using Phase Information for Symmetry Detection

    Xiao, Z.; Hou, Z.; Miao, C.; Wang, J. Using Phase Information for Symmetry Detection. Pattern Recognition Letters 2005, 26, 1985–1994. https://doi.org/10.1016/j.patrec.2005.02.003

  15. [24]

    Finding Mirror Symmetry via Registration and Optimal Symmetric Pairwise Assignment of Curves

    Cicconet, M.; Hildebrand, D.G.C.; Elliott, H. Finding Mirror Symmetry via Registration and Optimal Symmetric Pairwise Assignment of Curves. In Proceedings of the 2017 IEEE Inter- national Conference on Computer Vision Workshops (ICCVW), V enice, 2017; pp. 1749–1758. https://do...

  16. [25]

    Learning-Based Symmetry Detection in Natural Images

    Tsogkas, S.; Kokkinos, I. Learning-Based Symmetry Detection in Natural Images. In Computer Vision – ECCV 2012; Hutchison, D.; Kanade, T.; Kittler, J.; Kleinberg, J.M.; Mattern, F.; Mitchell, J.C.; Naor, M.; Nierstrasz, O.; Pandu Rangan, C.; Steffen, B.; et al., Eds.; Springer ...

  17. [26]

    SRN: Side-Output Residual Network for Object Symmetry Detection in the Wild

    Ke, W.; Chen, J.; Jiao, J.; Zhao, G.; Ye, Q. SRN: Side-Output Residual Network for Object Symmetry Detection in the Wild. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, 2017; pp. 302–310. https://doi.org/10 .1109/CVP...

  18. [27]

    Reflection and Rotation Symmetry Detection via Equiv- ariant Learning

    Seo, A.; Kim, B.; Kwak, S.; Cho, M. Reflection and Rotation Symmetry Detection via Equiv- ariant Learning. In Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, 2022; pp. 9529–9538. https: //doi.org/10.1109/CVPR5...

  19. [29]

    CLIPSym: Delving into Symmetry Detection with CLIP, 2025, [arXiv:cs.CV/2508.14197]

    Yang, T.; Rahman, M.A.; Yeh, R.A. CLIPSym: Delving into Symmetry Detection with CLIP, 2025, [arXiv:cs.CV/2508.14197]. https://doi.org/10.48550/arXiv .2508.14197

  20. [31]

    Reflection Symmetry Detection of Shapes Based on Shape Signatures

    Nguyen, T.P .; Truong, H.P .; Nguyen, T.T.; Kim, Y .G. Reflection Symmetry Detection of Shapes Based on Shape Signatures. Pattern Recognition 2022, 128, 108667. https://doi.org/10.1016/j. patcog.2022.108667

  21. [32]

    Detection of the Optimal Reflection Symmetry Axis with the Jaccard Index and the Radon Transform

    Lomov , N.; Seredin, O.; Kushnir, O. Detection of the Optimal Reflection Symmetry Axis with the Jaccard Index and the Radon Transform. In Proceedings of the 2022 International Russian Automation Conference (RusAutoCon), Sochi, Russian Federation, 2022; pp. 489–498. https: //doi...

  22. [33]

    Mirror Symmetry Detection in Digital Images:

    Mestetskiy , L.; Zhuravskaya, A. Mirror Symmetry Detection in Digital Images:. In Proceedings of the Proceedings of the 15th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications, Valletta, Malta, 2020; pp. 331–337. https: //d...

  23. [34]

    ALGORITHMS FOR ADJUSTMENT OF SYM- METRY AXIS FOUND FOR 2D SHAPES BY THE SKELETON COMPARISON METHOD

    Kushnir, O.A.; Seredin, O.S.; Fedotova, S.A. ALGORITHMS FOR ADJUSTMENT OF SYM- METRY AXIS FOUND FOR 2D SHAPES BY THE SKELETON COMPARISON METHOD. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. 2019, XLII-2/W12, 129–136. https: //doi.org/10.5194/isprs-archives-XLII-2-W12-129-2019

  24. [35]

    Detection of Mirror-Symmetric Image Patches

    Patraucean, V .; V on Gioi, R.G.; Ovsjanikov , M. Detection of Mirror-Symmetric Image Patches. In Proceedings of the 2013 IEEE Conference on Computer Vision and Pattern Recognition Workshops, OR, USA, 2013; pp. 211–216. https://doi.org/10.1109/CVPRW.2013.38

  25. [37]

    The Gradient Product Transform: An Image Filter for Sym- metry Detection

    Dalitz, C.; Wilberg, J.; Jeltsch, M. The Gradient Product Transform: An Image Filter for Sym- metry Detection. Image Processing On Line 2019, 9, 413–431. https://doi.org/10.5201/ipol.20 19.270

  26. [38]

    Computational Modeling and Experimental Investigation of Effects of Compositional Elements on Interface and Design Aesthetics

    Bauerly , M.; Liu, Y . Computational Modeling and Experimental Investigation of Effects of Compositional Elements on Interface and Design Aesthetics. International Journal of Human- Computer Studies 2006, 64, 670–682. https://doi.org/10.1016/j.ijhcs.2006.01.002

  27. [39]

    Comparison of Objective Measures for Predicting Perceptual Bal- ance and Visual Aesthetic Preference

    Hübner, R.; Fillinger, M.G. Comparison of Objective Measures for Predicting Perceptual Bal- ance and Visual Aesthetic Preference. Front. Psychol. 2016, 7. https://doi.org/10.3389/fpsyg. 2016.00335

  28. [40]

    Robust Midsagittal Plane Extraction from Coarse, Patho- logical 3D Images

    Liu, Y .; Collins, R.T.; Rothfus, W.E. Robust Midsagittal Plane Extraction from Coarse, Patho- logical 3D Images. In Medical Image Computing and Computer-Assisted Intervention – MIC- CAI 2000 ; Goos, G.; Hartmanis, J.; Van Leeuwen, J.; Delp, S.L.; DiGoia, A.M.; Jaramaz, B., 22...

  29. [41]

    Extracting the Symmetry of the Human Face from Digital Photographs

    Renero-C, F.J.; Romero-H, R.A.; Peregrina-B, H. Extracting the Symmetry of the Human Face from Digital Photographs. Bio-Algorithms and Med-Systems 2017, 13, 103–109. https://doi.org/ 10.1515/bams-2017-0002

  30. [42]

    A Toolbox for Calculating Objective Image Properties in Aesthetics Research

    Redies, C.; Bartho, R.; Koßmann, L.; Spehar, B.; Hübner, R.; Wagemans, J.; Hayn-Leichsenring, G.U. A Toolbox for Calculating Objective Image Properties in Aesthetics Research

  31. [43]

    An Image Quality Metric Based on Corner, Edge and Symmetry Maps

    Cui, L.; Allen, A. An Image Quality Metric Based on Corner, Edge and Symmetry Maps. In Proceedings of the Procedings of the British Machine Vision Conference 2008, Leeds, 2008; pp. 38.1–38.10. https://doi.org/10.5244/C.22.38

  32. [44]

    SymmMap: Estimation of the 2-D Reflection Symmetry Map and Its Applications

    Nagar, R.; Raman, S. SymmMap: Estimation of the 2-D Reflection Symmetry Map and Its Applications. In Proceedings of the 2017 IEEE International Conference on Computer Vision Workshops (ICCVW), V enice, Italy , 2017; pp. 1715–1724. https://doi.org/10.1109/ICCVW.20 17.201

  33. [45]

    From Pixels to Perception: A Benchmark for Human-like Symmetry Detection

    Muradás Odriozola, G.; Koßmann, L.; Tuytelaars, T.; Wagemans, J. From Pixels to Perception: A Benchmark for Human-like Symmetry Detection. Vision Research 2026, 245, 108825. https: //doi.org/10.1016/j.visres.2026.108825

  34. [46]

    Detecting Symmetry in Grey Level Images: The Global Optimization Approach

    Kiryati, N.; Gofman, Y . Detecting Symmetry in Grey Level Images: The Global Optimization Approach. International Journal of Computer Vision 1998, 29, 29–45. https://doi.org/10.1023/A: 1008034529558

  35. [47]

    Histograms of Oriented Gradients for Human Detection

    Dalal, N.; Triggs, B. Histograms of Oriented Gradients for Human Detection. In Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), San Diego, CA, USA, 2005; V ol. 1, pp. 886–893. https://doi.org/10.1109/CVPR.20 05.177

  36. [48]

    Representing Shape with a Spatial Pyramid Kernel

    Bosch, A.; Zisserman, A.; Munoz, X. Representing Shape with a Spatial Pyramid Kernel. In Proceedings of the Proceedings of the 6th ACM International Conference on Image and Video Retrieval, Amsterdam The Netherlands, 2007; pp. 401–408. https://doi.org/10.1145/1282280. 1282340

  37. [49]

    PyTorch Image Models, 2019

    Wightman, R. PyTorch Image Models, 2019. https://doi.org/10.5281/zenodo.4414861

  38. [50]

    MambaOut: Do We Really Need Mamba for Vision?

    Yu, W.; Wang, X. MambaOut: Do We Really Need Mamba for Vision?

  39. [51]

    The Canonical Deep Neural Network as a Model for Human Symme- try Processing

    Bonneh, Y .S.; Tyler, C.W. The Canonical Deep Neural Network as a Model for Human Symme- try Processing. iScience 2025, 28, 111540. https://doi.org/10.1016/j.isci.2024.111540

  40. [52]

    Mirror Reflection Invariant HOG Descriptors for Object Detection

    Kanezaki, A.; Mukuta, Y .; Harada, T. Mirror Reflection Invariant HOG Descriptors for Object Detection. In Proceedings of the 2014 IEEE International Conference on Image Processing (ICIP), Paris, France, 2014; pp. 1594–1598. https://doi.org/10.1109/ICIP .2014.7025319

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.