Pith. sign in

REVIEW 2 major objections 5 minor 56 references

Enhancing Quantum-ready QUBO-based Suppression for Object Detection with Appearance and Confidence Features

T0 review · 2 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Adding SSIM and confidence to QUBO suppression gains up to 4.5 mAP

desk verdict A useful incremental QUBO-suppression extension with a genuinely fast SSIM implementation, but the paper's own Section 5.1 shows the new terms only shrink existing penalties, so the stated appearance/confidence mechanism is untested without a scale-matched control. read the letter →

arxiv 2502.02895 v1 pith:K6QTEPE4 submitted 2025-02-05 cs.CV

Keiichiro Yamamura , Toru Mitsutake , Hiroki Ishikura , Daiki Kusuhara , Akihiro Yoshida , Katsuki Fujisawa This is my paper · ORCID
classification cs.CV
keywords QUBOobjectdetectionnon-maximumsuppressionSSIMocclusionconfidencescorequantum-readyCrowdHuman
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

Suppression is the stage of object detection that removes duplicate boxes around one object; standard greedy methods tend to delete genuinely distinct objects that happen to overlap because one is occluded. This paper proposes changing the quadratic objective used by QUBO-based suppression so that its pairwise penalty is modulated by two extra signals: the visual similarity of the two predicted crops, measured by SSIM, and the product of their confidence scores. The authors call the new formulations QAQS and QAQS-C and show that they keep more true positives than the state-of-the-art QUBO baseline, gaining up to 4.54 points in mAP and 9.89 points in large-object recall on the crowded CrowdHuman benchmark while adding only a few milliseconds per image. The deeper point is that a suppression objective can, in principle, tell two boxes on one person apart from two people with overlapping boxes by looking beyond geometry.

What carries the argument

The load-bearing object is the modified QUBO coefficient matrix. SSIM (Structural SIMilarity, a standard image-quality measure of brightness, contrast, and structure between two image patches) is computed on every pair of resized detection crops that overlap, yielding a symmetric matrix $A$ with entries at most 1; QAQS forms $Q = w_1 L - (w_2 P_1 + w_3 P_2) \odot A$, and QAQS-C forms $Q = w_1 L - v((w_2 P_1 + w_3 P_2) \odot A) v^{\top}$. Because SSIM is at most 1 and confidence is at most 1, the new terms shrink the pairwise penalty relative to QSQS, which the paper identifies in its discussion as one reason the new formulations suppress fewer true positives. To make the appearance term affordable, the paper parallelizes SSIM on a GPU with a divide-and-conquer scheme and reorders the intersection matrix to skip pairs with zero overlap.

What would settle it

Repeat the CrowdHuman and COCO experiments with the SSIM matrix replaced by a constant matrix equal to the dataset's mean SSIM value. If QAQS still gains the reported 4.4-4.5 mAP and 9.5-9.9 recall over QSQS, the appearance signal itself is not responsible for the improvement, and only the penalty-shrinking effect remains.

Watch

Extended reading notes

Core claim

The central claim is that the identity of the object behind a box can be read off the QUBO coefficients themselves. The paper modifies the QSQS coefficient matrix $Q = w_1 L - (w_2 P_1 + w_3 P_2)$ by element-wise multiplying the pairwise overlap terms by an SSIM appearance matrix $A$, giving QAQS, and then additionally sandwiching with the confidence vector $v$, giving QAQS-C: $Q = w_1 L - v((w_2 P_1 + w_3 P_2) \odot A) v^{\top}$. The authors state that these terms let the objective distinguish whether overlap is occlusion or redundancy, and report consistent gains on COCO and CrowdHuman; on CrowdHuman, QAQS and QAQS-C exceed the baseline's mAP by 4.39 and 4.54 points and its large-object recall by 9.47 and 9.89 points. They also report a GPU SSIM implementation that cuts appearance-feature cost from seconds to single-digit milliseconds per image.

Load-bearing premise

The formulation works only if the appearance of two overlapping crops and the product of their confidence scores actually encode whether the overlap is redundant or occluded; if the observed gains instead come from the fact that these multipliers shrink every pairwise penalty, the specific features are not doing the claimed work.

Editorial extensions

If this is right

  • On crowded benchmarks, the proposed formulations keep more true positives that NMS-style suppression drops, so dense-scene recall improves without a meaningful speed penalty.
  • The speedup of SSIM from seconds to milliseconds makes appearance-aware suppression practical for near-real-time pipelines.
  • Because the QUBO is unchanged in form from the baseline, the same suppression objective can be handed to a quantum annealer or QAOA circuit when hardware with enough logical qubits exists.
  • Replacing IoU with a clipped sparser overlap metric roughly halves suppression runtime while preserving the accuracy gains, helping scale to larger proposal sets.
  • The confidence-weighted variant QAQS-C is robust to adding or removing a soft-scoring post-process, whereas the baseline's result depends strongly on that post-process.

Reading between the lines

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

  • A direct test of whether the semantic content of SSIM matters would replace A with a constant equal to the average SSIM across pairs; if mAP/mAR match the reported gains, the improvement is just a scaled-down penalty rather than the appearance signal.
  • The confidence-product term could be made category-dependent, since occlusion statistics differ strongly across object classes, which the paper does not explore.
  • The paper's runtime breakdown implies the largest remaining gain is in faster QUBO solving, not more feature engineering; swapping the classical solver for a faster quantum one is the bottleneck the authors themselves identify.
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 / 5 minor

Summary. The paper proposes two new QUBO-based suppression formulations for object detection, QAQS and QAQS-C, that extend the existing QSQS method by multiplying the pairwise penalty terms with an SSIM-based appearance similarity and, for QAQS-C, with confidence-score products. The stated goal is to distinguish overlapping predictions caused by occlusion of distinct objects from redundant predictions of the same object. The authors evaluate on COCO 2017 and CrowdHuman with a Faster R-CNN detector, reporting mAP and mAR improvements over QSQS (up to 4.54 mAP and 9.89 mAR on CrowdHuman) and a GPU/divide-and-conquer SSIM implementation that reduces per-image SSIM runtime to milliseconds. They also provide a quantum-ready software interface. The paper's central interpretive claim is that the appearance and confidence features allow the QUBO objective to discriminate occlusion from redundancy, and it concludes that the results empirically validate this hypothesis.

Significance. If the proposed mechanism is real, the work offers a simple and computationally cheap improvement to QUBO-based suppression, with a useful engineering contribution in the fast SSIM implementation and public software. The authors wisely fix QUBO weights to prior values rather than tuning them to the benchmark, and they compare QUBO solutions directly without soft-scoring, which is a clean way to isolate the formulation quality. However, the central conceptual claim is currently underdetermined: because all new factors are bounded by 1, the reported gains are equally consistent with a uniform reduction in suppression strength. The paper itself notices this in Section 5.1 but does not run the control experiment needed to separate the shrinkage effect from the hypothesized appearance/confidence mechanism. The results are still valuable as an empirical improvement, but the mechanism claim needs additional evidence.

major comments (2)
  1. [Section 3.1, Eq. (8)] Equation (8) is dimensionally invalid as written: with v ∈ R^{n×1}, the expression v ((w2P1 + w3P2) ⊙ A) v^T is a scalar, not an n×n coefficient matrix. The intended element-wise confidence weighting should be written as Q = w1L − diag(v) ((w2P1 + w3P2) ⊙ A) diag(v), or equivalently Q = w1L − (v v^T) ⊙ ((w2P1 + w3P2) ⊙ A). As it appears in the text, the definition cannot be implemented without guessing the author's intention.
  2. [Sections 4.4, 5.1, and 7] Because SSIM ≤ 1 and confidence scores lie in [0,1], every non-diagonal penalty in QAQS and QAQS-C is the corresponding QSQS penalty multiplied by a factor in (0,1]. The paper acknowledges this in Section 5.1: "The absolute value of non-diagonal elements of the QAQS coefficient matrix is smaller than that of QSQS because SSIM is less than or equal to 1. Similarly, the absolute value of non-diagonal elements of the QAQS-C coefficient matrix is smaller than that of QAQS because the confidence score is between 0 and 1." Consequently, the mAP/mAR gains in Tables 1 and 2 are consistent with a simple shrinkage of the pairwise suppression strength, even if the SSIM and confidence features carry no discriminative information about occlusion versus redundancy. To support the central hypothesis, the ablation in Table 2 must include a control where the QSQS pairwise terms are reduced by a constant factor (for example, scaling w2 and w3 by the mean of the SSIM/confidence products, or replacing A by a constant matrix). Without such a control, the claim in Section 7 that "These results empirically validate our hypothesis" is not justified.
minor comments (5)
  1. [Section 4.5] There is a typo: "CroudHuman" should be "CrowdHuman."
  2. [Equations (4)–(6)] The SSIM definition uses non-standard constants C1=0.012, C2=0.032, and C3=2C2; please clarify whether these correspond to pixel values in [0,1] or [0,255], since the standard formulation typically uses C=(K L)^2 with L being the dynamic range.
  3. [Section 3.2] The statement "Empirically, resizing to around 48 × 48 does not affect performance" is given without supporting data or a reference; a small sensitivity analysis would strengthen the reproducibility of the runtime claims.
  4. [Table 6 caption] The caption notes that the second-best tie is not underlined; this is an unusual presentation choice and could confuse readers comparing Table 6 with Tables 1 and 5.
  5. [Appendix B, Table 4] All three methods produce identical numbers in Experiment No. 1; a sentence explaining why would help the reader interpret the result (e.g., soft-scoring dominates the final output after confidence-based preprocessing).

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the QUBO extensions are heuristic, evaluated against external benchmarks, and not fitted to the target result; Section 5.1 reveals an interpretation confound but not circularity.

full rationale

This paper does not make a derivation-to-fit circular move. The proposed QUBO extensions in Eqs. (7) and (8) are heuristics motivated by two stated hypotheses, not quantities fitted to the benchmark; the QUBO weights are carried over from the original QSQS implementation (Sec. 4.2), and SSIM and confidence are fixed functions of the detector output. The main claims are empirical comparisons against QSQS on external CrowdHuman and COCO evaluation sets, so the reported mAP/mAR improvements are observed outcomes, not consequences of the definitions. Section 5.1 does explicitly concede that, because SSIM is at most 1 and confidence is in [0,1], the new pairwise penalties are pointwise smaller than in QSQS. This is a genuine confound for the paper's semantic interpretation that appearance and confidence encode the occlusion-versus-redundancy distinction, and the conclusion's statement that the results 'empirically validate our hypothesis' is not fully supported without a control that scales the QSQS pairwise weights by the same average factors. However, that is an inference and experimental-design limitation, not circularity: no equation is defined in terms of the target metric and no fitted parameter is relabeled as a prediction. There are no load-bearing self-citations, no imported uniqueness theorems, and no known result is merely renamed. The empirical comparison is self-contained against external benchmarks, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The ledger reflects that the method inherits the QUBO suppression model from prior work and adds two unproven feature hypotheses. There are no genuinely invented entities; the three free parameters are inherited or hand-chosen. The hypotheses about appearance and confidence are the main domain assumptions, and they are not independently tested.

free parameters (3)
  • QUBO weights w1, w2, w3 = w1=0.4, w2=0.3, w3=0.3
    Weights balance unary confidence against pairwise penalties; authors tuned via Optuna then fixed to the same values as QSQS. They affect all scores and are not derived from first principles.
  • Confidence threshold for preprocessing = 0.25
    Predictions below score 0.25 are discarded before QUBO; the threshold was chosen by balancing quantitative and qualitative results, and affects the final mAP and mAR.
  • Resize size for SSIM crops = 48 x 48 pixels
    All predicted boxes are resized to 48 by 48 for SSIM; the paper says this size was chosen empirically.
assumptions (3)
  • domain assumption Maximizing x^T Q x with a diagonal confidence reward and negative pairwise overlap penalties is a valid model for suppression.
    Adopted from QF and QSQS; the paper does not justify this model beyond prior empirical success. Invoked in Section 3.1, Eqs. (7) and (8).
  • domain assumption SSIM between resized bounding-box crops is a useful proxy for whether two detections are redundant or correspond to distinct objects.
    Central hypothesis in Section 3.1; the paper's own Section 5.1 shows gains may be explained by penalty shrinkage rather than by this proxy.
  • domain assumption The product of confidence scores scales pairwise penalties in a way that helps recover occluded true positives.
    Motivated by DPP [28]; hypothesis stated in Section 3.1. No independent evidence isolates this effect from the appearance feature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Quantum-ready QUBO-based Suppression for Object Detection with Appearance and Confidence Features." pith.science (2026). https://pith.science/paper/K6QTEPE4

@misc{pith2026250202895,
  author       = {Pith},
  title        = {Pith review of: Enhancing Quantum-ready QUBO-based Suppression for Object Detection with Appearance and Confidence Features},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K6QTEPE4}},
  note         = {Machine review of arXiv:2502.02895}
}
read the original abstract

Quadratic Unconstrained Binary Optimization (QUBO)-based suppression in object detection is known to have superiority to conventional Non-Maximum Suppression (NMS), especially for crowded scenes where NMS possibly suppresses the (partially-) occluded true positives with low confidence scores. Whereas existing QUBO formulations are less likely to miss occluded objects than NMS, there is room for improvement because existing QUBO formulations naively consider confidence scores and pairwise scores based on spatial overlap between predictions. This study proposes new QUBO formulations that aim to distinguish whether the overlap between predictions is due to the occlusion of objects or due to redundancy in prediction, i.e., multiple predictions for a single object. The proposed QUBO formulation integrates two features into the pairwise score of the existing QUBO formulation: i) the appearance feature calculated by the image similarity metric and ii) the product of confidence scores. These features are derived from the hypothesis that redundant predictions share a similar appearance feature and (partially-) occluded objects have low confidence scores, respectively. The proposed methods demonstrate significant advancement over state-of-the-art QUBO-based suppression without a notable increase in runtime, achieving up to 4.54 points improvement in mAP and 9.89 points gain in mAR.

Figures

Figures reproduced from arXiv: 2502.02895 by the authors.

Figure 2
Figure 2. Visualization of Intersection Matrix I ∈ {0, 1} n×n . Zero-value elements are colored in purple, and one-value elements are colored in yellow. the permuted I has large zero matrix blocks, where corre￾sponding SSIM computations are efficiently omitted. The RCM algorithm is based on breadth-first search and mini￾mizes the bandwidth of the matrix, i.e., the maximum dis￾tance between non-zero and diagonal elements. The … view at source ↗
Figure 3
Figure 3. Visualization of suppressed predictions. Confidence scores of false negatives (Fig. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Breakdown of QUBO-based suppression runtime per [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Ablation study of SSIM computation. Naive is the se￾quential computation on the CPU based on the implementation of scikit-image. GPU represents the GPU parallelization. Rec repre￾sents the recursive computation using the divide-and-conquer al￾gorithm. Ord shows the reo…
Figure 6
Figure 6. Figure 6: Example usage of our software. We implement our proposed methods as modularized soft￾ware for portability. A minimal sample codes are shown in [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 53 canonical work pages

  1. [1]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowl- edge Discovery & Data Mining, KDD 2019, Anchorage, AK, USA, August 4-8, 2019, pages 2623–2631, 2019. 5

    work page 2019

  2. [2]

    Fast explicit diffusion for accelerated features in non- linear scale spaces

    Pablo Fern ´andez Alcantarilla, Jes´us Nuevo, and Adrien Bar- toli. Fast explicit diffusion for accelerated features in non- linear scale spaces. In British Machine Vision Conference, BMVC 2013, Bristol, UK, September 9-13, 2013, 2013. 8

    work page 2013

  3. [3]

    Yolov1 to yolov10: A comprehensive review of YOLO variants and their application in the agricultural domain

    Mujadded Al Rabbani Alif and Muhammad Hussain. Yolov1 to yolov10: A comprehensive review of YOLO variants and their application in the agricultural domain. CoRR, abs/2406.10139, 2024. 8

    arXiv 2024

  4. [4]

    Bardin, Rami Barends, Rupak Biswas, Sergio Boixo, Fernando G

    Frank Arute, Kunal Arya, Ryan Babbush, Dave Ba- con, Joseph C. Bardin, Rami Barends, Rupak Biswas, Sergio Boixo, Fernando G. S. L. Brandao, David A. Buell, Brian Burkett, Yu Chen, Zijun Chen, Ben Chiaro, Roberto Collins, William Courtney, Andrew Dunsworth, Edward Farhi, Brooks Foxen, Austin Fowler, Craig Gid- ney, Marissa Giustina, Rob Graff, Keith Guerin...

    work page 2019

  5. [5]

    Algorithms for hyper-parameter optimization

    James Bergstra, R ´emi Bardenet, Yoshua Bengio, and Bal´azs K´egl. Algorithms for hyper-parameter optimization. In Ad- vances in Neural Information Processing Systems 24: 25th Annual Conference on Neural Information Processing Sys- tems 2011. Proceedings of a meeting held 12-14 December 2011, Granada, Spain, pages 2546–2554, 2011. 5

    work page 2011

  6. [6]

    James Bergstra, Daniel Yamins, and David D. Cox. Mak- ing a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures. In Pro- ceedings of the 30th International Conference on Machine Learning, ICML 2013, Atlanta, GA, USA, 16-21 June 2013 , pages 115–123, 2013. 5

    work page 2013

  7. [7]

    Quantum permutation synchronization

    Tolga Birdal, Vladislav Golyanik, Christian Theobalt, and Leonidas J Guibas. Quantum permutation synchronization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13122–13133, 2021. 8

    work page 2021

  8. [8]

    Navaneeth Bodla, Bharat Singh, Rama Chellappa, and Larry S. Davis. Soft-nms - improving object detection with one line of code. In 2017 IEEE/CVF International Confer- ence on Computer Vision, ICCV 2017, Venice, Italy, October 22-29, 2017, pages 5562–5570, 2017. 1, 2, 8

    work page 2017

Show all 56 references
  1. [9]

    DATE: dual assignment for end-to-end fully convolutional object detection

    Yiqun Chen, Qiang Chen, Qinghao Hu, and Jian Cheng. DATE: dual assignment for end-to-end fully convolutional object detection. CoRR, abs/2211.13859, 2022. 8

    2022 arXiv

  2. [10]

    Cuthill and J

    E. Cuthill and J. McKee. Reducing the bandwidth of sparse symmetric matrices. In Proceedings of the 1969 24th Na- tional Conference , page 157–172, New York, NY , USA,

    1969

  3. [11]

    A hybrid quantum-classical algorithm for robust fitting

    Anh-Dzung Doan, Michele Sasdelli, David Suter, and Tat- Jun Chin. A hybrid quantum-classical algorithm for robust fitting. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 417–427, 2022. 8

    2022

  4. [12]

    A new dataset and performance evaluation of a region-based CNN for urban object detection

    Alex Dominguez-Sanchez, Sergio Orts-Escolano, Jos´e Garc ´ıa Rodr ´ıguez, and Miguel Cazorla. A new dataset and performance evaluation of a region-based CNN for urban object detection. In 2018 International Joint Conference on Neural Networks, IJCNN 2018, Rio de Janeiro, Brazi...

    2018

  5. [13]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  6. [14]

    Rex A. Dwyer. A faster divide-and-conquer algorithm for constructing delaunay triangulations. Algorithmica, 2(1): 137–151, 1987. 2

    1987

  7. [15]

    Everingham, S

    M. Everingham, S. M. A. Eslami, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The pascal visual ob- ject classes challenge: A retrospective.International Journal of Computer Vision, 111(1):98–136, 2015. 2

    2015

  8. [16]

    A quantum approximate optimization algorithm, 2014

    Edward Farhi, Jeffrey Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm, 2014. 2

    2014

  9. [17]

    Visibility guided NMS: efficient boosting of amodal object detection in crowded traffic scenes

    Nils G ¨ahlert, Niklas Hanselmann, Uwe Franke, and Joachim Denzler. Visibility guided NMS: efficient boosting of amodal object detection in crowded traffic scenes. In Ma- chine Learning for Autonomous Driving Workshop at the 33rd Conference on Neural Information Processing Sys...

    2019

  10. [18]

    Are we ready for autonomous driving? the KITTI vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the KITTI vision benchmark suite. In IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, CVPR 2012, Providence, RI, USA, June 16-21, 2012, pages 3354–3361, 2012. 1

    2012

  11. [19]

    A quantum com- putational approach to correspondence problems on point sets

    Vladislav Golyanik and Christian Theobalt. A quantum com- putational approach to correspondence problems on point sets. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 9182–9191,

  12. [20]

    Eisenstat

    Ming Gu and Stanley C. Eisenstat. A divide-and-conquer al- gorithm for the symmetric tridiagonal eigenproblem. SIAM Journal on Matrix Analysis and Applications , 16(1):172– 191, 1995. 2

    1995

  13. [21]

    Gurobi Optimizer Reference Manual, 2024

    Gurobi Optimization, LLC. Gurobi Optimizer Reference Manual, 2024. 2, 5

    2024

  14. [22]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016, pages 770–778, 2016. 1, 4

    2016

  15. [23]

    Bounding box regression with uncer- tainty for accurate object detection

    Yihui He, Chenchen Zhu, Jianren Wang, Marios Savvides, and Xiangyu Zhang. Bounding box regression with uncer- tainty for accurate object detection. In IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, CVPR 2019, Long Beach, CA, USA, June 16-20, 2019, pages 2888...

    2019

  16. [24]

    Image quality metrics: PSNR vs

    Alain Hor ´e and Djemel Ziou. Image quality metrics: PSNR vs. SSIM. In 20th International Conference on Pattern Recognition, ICPR 2010, Istanbul, Turkey, 23-26 August 2010, pages 2366–2369, 2010. 3, 8

    2010

  17. [25]

    Visible feature guidance for crowd pedestrian detection

    Zhida Huang, Kaiyu Yue, Jiangfan Deng, and Feng Zhou. Visible feature guidance for crowd pedestrian detection. In Computer Vision - ECCV 2020 Workshops - Glasgow, UK, August 23-28, 2020, Proceedings, Part V , pages 277–290,

    2020

  18. [26]

    Wood, Jake Lishman, Julien Gacon, Simon Martiel, Paul D

    Ali Javadi-Abhari, Matthew Treinish, Kevin Krsulich, Christopher J. Wood, Jake Lishman, Julien Gacon, Simon Martiel, Paul D. Nation, Lev S. Bishop, Andrew W. Cross, Blake R. Johnson, and Jay M. Gambetta. Quantum comput- ing with Qiskit, 2024. 9

    2024

  19. [27]

    Quantum an- nealing in the transverse ising model.Phys

    Tadashi Kadowaki and Hidetoshi Nishimori. Quantum an- nealing in the transverse ising model.Phys. Rev. E, 58:5355– 5363, 1998. 2

    1998

  20. [28]

    Individualness and determinantal point processes 12 for pedestrian detection

    Donghoon Lee, Geonho Cha, Ming-Hsuan Yang, and Songh- wai Oh. Individualness and determinantal point processes 12 for pedestrian detection. In Computer Vision - ECCV 2016 - 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part VI, pages 330–346,

    2016

  21. [29]

    Quantum-soft QUBO sup- pression for accurate object detection

    Junde Li and Swaroop Ghosh. Quantum-soft QUBO sup- pression for accurate object detection. In Computer Vision - ECCV 2020 - 16th European Conference, Glasgow, UK, Au- gust 23-28, 2020, Proceedings, Part XXIX , pages 158–173,

    2020

  22. [30]

    Girshick, Kaiming He, Bharath Hariharan, and Serge J

    Tsung-Yi Lin, Piotr Doll ´ar, Ross B. Girshick, Kaiming He, Bharath Hariharan, and Serge J. Belongie. Feature pyra- mid networks for object detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017, pages 936–9...

    2017

  23. [31]

    Lawrence Zitnick, and Piotr Doll ´ar

    Tsung-Yi Lin, Michael Maire, Serge Belongie, Lubomir Bourdev, Ross Girshick, James Hays, Pietro Perona, Deva Ramanan, C. Lawrence Zitnick, and Piotr Doll ´ar. Microsoft coco: Common objects in context, 2015. 2, 4, 5

    2015

  24. [32]

    Adaptive NMS: refining pedestrian detection in a crowd

    Songtao Liu, Di Huang, and Yunhong Wang. Adaptive NMS: refining pedestrian detection in a crowd. In IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, CVPR 2019, Long Beach, CA, USA, June 16-20, 2019, pages 6459– 6468, 2019. 1, 8

    2019

  25. [33]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In 2021 IEEE/CVF International Conference on Computer Vi- sion, ICCV 2021, Montreal, QC, Canada, October 10-17, ...

    2021

  26. [34]

    David G. Lowe. Object recognition from local scale- invariant features. In Proceedings of the International Conference on Computer Vision, Kerkyra, Corfu, Greece, September 20-25, 1999, pages 1150–1157, 1999. 8

    1999

  27. [35]

    Detrs beat yolos on real-time object detection

    Wenyu Lv, Shangliang Xu, Yian Zhao, Guanzhong Wang, Jinman Wei, Cheng Cui, Yuning Du, Qingqing Dang, and Yi Liu. Detrs beat yolos on real-time object detection. CoRR, abs/2304.08069, 2023. 8

    2023 arXiv

  28. [36]

    Prokhorov, and Dacheng Tao

    Xue Mei, Zhibin Hong, Danil V . Prokhorov, and Dacheng Tao. Robust multitask multiview tracking in videos. IEEE Trans. Neural Networks Learn. Syst. , 26(11):2874–2890,

  29. [37]

    An iterative quantum approach for transformation estimation from point sets

    Natacha Kuete Meli, Florian Mannel, and Jan Lellmann. An iterative quantum approach for transformation estimation from point sets. In IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 519–527, 2022. 8

    2022

  30. [38]

    Mathematical foundation of quantum annealing

    Satoshi Morita and Hidetoshi Nishimori. Mathematical foundation of quantum annealing. Journal of Mathematical Physics, 49(12), 2008. 2

    2008

  31. [39]

    Solv- ing traffic data occlusion problems in computer vision al- gorithms using deepsort and quantum computing

    Frank Ngeni, Judith Mwakalonge, and Saidi Siuhi. Solv- ing traffic data occlusion problems in computer vision al- gorithms using deepsort and quantum computing. Journal of Traffic and Transportation Engineering (English Edition), 11(1):1–15, 2024. 8

    2024

  32. [40]

    Nonnegative/binary matrix fac- torization with a d-wave quantum annealer

    Daniel O’Malley, Velimir V Vesselinov, Boian S Alexandrov, and Ludmil B Alexandrov. Nonnegative/binary matrix fac- torization with a d-wave quantum annealer. PloS one , 13 (12):e0206653, 2018. 8

    2018

  33. [41]

    Faster exact solution of sparse maxcut and qubo problems

    Daniel Rehfeldt, Thorsten Koch, and Yuji Shinano. Faster exact solution of sparse maxcut and qubo problems. Math- ematical Programming Computation, 15(3):445–470, 2023. 8

    2023

  34. [42]

    Girshick, and Jian Sun

    Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: towards real-time object detection with re- gion proposal networks. In Advances in Neural Information Processing Systems 28: Annual Conference on Neural In- formation Processing Systems 2015, December 7-12...

    2015

  35. [43]

    Reid, and Silvio Savarese

    Hamid Rezatofighi, Nathan Tsoi, JunYoung Gwak, Amir Sadeghian, Ian D. Reid, and Silvio Savarese. Generalized intersection over union: A metric and a loss for bounding box regression. In IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, CVPR 2019, Long Beach, CA...

    2019

  36. [44]

    Sitapa Rujikietgumjorn and Robert T. Collins. Optimized pedestrian detection for multiple and occluded people. In IEEE/CVF Conference on Computer Vision and Pattern Recognition,CVPR 2018, Portland, OR, USA, June 23-28, 2013, pages 3690–3697, 2013. 1, 2, 3, 8

    2018

  37. [45]

    Crowdhuman: A bench- mark for detecting human in a crowd

    Shuai Shao, Zijian Zhao, Boxun Li, Tete Xiao, Gang Yu, Xiangyu Zhang, and Jian Sun. Crowdhuman: A bench- mark for detecting human in a crowd. arXiv preprint arXiv:1805.00123, 2018. 2, 4

    2018 arXiv

  38. [46]

    Confidence propagation cluster: Unleash full potential of object detectors

    Yichun Shen, Wanli Jiang, Zhen Xu, Rundong Li, and Junghyun Kwon. Confidence propagation cluster: Unleash full potential of object detectors. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 1141–1151,

    2022

  39. [47]

    Confluence: A robust non-iou alternative to non- maxima suppression in object detection.IEEE Trans

    Andrew Shepley, Gregory Falzon, Paul Kwan, and Ljiljana Brankovic. Confluence: A robust non-iou alternative to non- maxima suppression in object detection.IEEE Trans. Pattern Anal. Mach. Intell., 45(10):11561–11574, 2023. 1, 4, 8

    2023

  40. [48]

    Very deep con- volutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep con- volutional networks for large-scale image recognition. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Confer- ence Track Proceedings, 2015. 1

    2015

  41. [49]

    Weighted boxes fusion: Ensembling boxes from different ob- ject detection models

    Roman Solovyev, Weimin Wang, and Tatiana Gabruseva. Weighted boxes fusion: Ensembling boxes from different ob- ject detection models. Image and Vision Computing , 107: 104117, 2021. 8

    2021

  42. [50]

    Body-face joint detection via embedding and head hook

    Junfeng Wan, Jiangfan Deng, Xiaosong Qiu, and Feng Zhou. Body-face joint detection via embedding and head hook. In 2021 IEEE/CVF International Conference on Computer Vi- sion, ICCV 2021, Montreal, QC, Canada, October 10-17, 2021, pages 2939–2948, 2021. 4

    2021

  43. [51]

    Bovik, Hamid R

    Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE Trans. Image Process., 13(4): 600–612, 2004. 2, 3, 8

    2004

  44. [52]

    WIDER FACE: A face detection benchmark

    Shuo Yang, Ping Luo, Chen Change Loy, and Xiaoou Tang. WIDER FACE: A face detection benchmark. In IEEE/CVF 13 Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016, pages 5525–5533, 2016. 1

    2016

  45. [53]

    Adiabatic quantum com- puting for multi object tracking

    Jan-Nico Zaech, Alexander Liniger, Martin Danelljan, Dengxin Dai, and Luc Van Gool. Adiabatic quantum com- puting for multi object tracking. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 8801–8812,

    2022

  46. [54]

    Efros, Eli Shecht- man, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, p...

    2018

  47. [55]

    Loss functions for image restoration with neural networks

    Hang Zhao, Orazio Gallo, Iuri Frosio, and Jan Kautz. Loss functions for image restoration with neural networks. IEEE Trans. Computational Imaging, 3(1):47–57, 2017. 8

    2017

  48. [56]

    Cad: Scale invariant framework for real-time object detection

    Huajun Zhou, Zechao Li, Chengcheng Ning, and Jinhui Tang. Cad: Scale invariant framework for real-time object detection. In Proceedings of the IEEE international confer- ence on computer vision workshops , pages 760–768, 2017. 8 14

    2017

Pith tools

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