Pith. sign in

REVIEW 2 major objections 5 minor 86 references

Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Point-to-region matching makes semi-supervised point-based crowd counting work.

desk verdict P2R is a genuinely useful fix for a real failure in semi-supervised point-based crowd counting; the central diagnosis is correct and the paper deserves refereeing, with one unquantified vulnerability around teacher recall. read the letter →

arxiv 2505.21943 v1 pith:C6MK65ZN submitted 2025-05-28 cs.CV

Wei Lin , Chenyang Zhao , Antoni B. Chan This is my paper
classification cs.CV
keywords semi-supervisedcrowdcountingpointdetectionpoint-to-regionmatchingpseudo-labelingpoint-specificactivationmaplocalizationdomainadaptationpoint-based
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 tries to establish that point-based crowd counters can be trained with only a few annotated images plus large amounts of unlabeled data, provided pseudo-label confidence is spread over local regions instead of single points. The authors show that the usual point-to-point matching is ill-posed for semi-supervised learning: the background term of the cross-entropy loss vanishes for unlabeled data, so the model is trained with only positive examples and collapses. Their point-to-region (P2R) matching gives every pixel near a pseudo-point that point's confidence and explicitly marks distant pixels as reliable background, restoring the missing gradients. Under this scheme, semi-supervised counting reaches state-of-the-art mean absolute errors on ShTech A/B, UCF-QNRF, and JHU++, and the same loss transfers to unsupervised domain adaptation. A byproduct is that the Hungarian matching step is no longer needed, making loss computation roughly 68 times faster.

What carries the argument

The load-bearing mechanism is the point-to-region (P2R) matching matrix $M = M_f \odot (\beta 1_m^\top)$, where $M_f$ is a many-to-one nearest-point assignment from pixels to pseudo-points and $\beta$ is the indicator of pixels within radius $\mu$ of any point. A per-column minimum of the cost matrix selects the representative foreground pixel in each region, and the confidence matrix $Z = \mathrm{diag}(M_{st}\zeta + (1-\beta))$ propagates each pseudo-point's confidence to its whole region while marking far pixels as reliable background. The binary cross-entropy loss is then computed with $Z$ as weights, so the background term is no longer zero. PSAM, a gradient-based per-point activation map, is the diagnostic that shows the over-activation growth and motivates the region-based supervision.

What would settle it

A direct test: in a fixed semi-supervised run, randomly drop a known fraction (say 10-20%) of the teacher's pseudo-points before the P2R loss is computed. If P2R's final MAE rises by roughly the dropped fraction compared to the full-pseudo-label run, that confirms the reliable-background assumption is the load-bearing part and that imperfect teacher recall would hurt. If the MAE is unchanged, the background confidence term is not the decisive mechanism.

Watch

Extended reading notes

Core claim

The paper's central claim is that the observed training collapse in semi-supervised point-based counting is caused by confidence never reaching background pixels. With point-to-point matching, the pseudo-label confidence vector is applied only to matched foreground pixels, so the second (background) term in the binary cross-entropy for unlabeled data is identically zero; the proof is given through equations (25)-(34), and the loss reduces to supervision with only positive samples. Using the point-specific activation map (PSAM), the authors visualize that the decoder's activation regions around each predicted person grow during this ill-posed training until neighboring pixels are misread as people. The P2R strategy replaces each one-pixel match with a local region: every pixel in the region inherits the pseudo-point's confidence, and pixels farther than a radius $\mu$ from all pseudo-points are assigned confidence 1 with label 0. This re-introduces the background supervision that P2P drops, stabilizes training, and produces the reported gains in semi-supervised counting (5% ShTech A MAE 69.9 vs 83.7 for OT-M; 40% ShTech A MAE 55.6 vs 70.7) and in unsupervised domain adaptation.

Load-bearing premise

The load-bearing assumption is that every pixel farther than a fixed radius from every pseudo-point really is background; if the teacher misses a person (score below the 0.5 extraction threshold), that missed person is trained as background and the mistake can be reinforced.

Editorial extensions

If this is right

  • With only 5% of labels, P2R reaches 69.9 MAE on ShTech A, beating the previous semi-supervised method OT-M's 83.7, and with 40% labels it reaches 55.6 vs 70.7.
  • Under the fully-supervised 100% protocol, P2R improves over P2P on all four benchmark datasets (e.g., ShTech A MAE 51.02 vs 52.74) while cutting loss computation time from 0.4307s to 0.0064s per image.
  • The same loss transfers to unsupervised domain adaptation, where P2R surpasses prior UDA methods on four cross-domain protocols (e.g., B->Q MAE 139.5 vs 194.9 for the best tabulated UDA baseline).
  • PSAM comparisons show the P2R-trained model keeps activations concentrated around true foreground pixels instead of spreading them onto neighbors, which is the mechanism behind the stable training.

Reading between the lines

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

  • Beyond the paper: the reliable-background rule means any person missed by the teacher (score below the 0.5 extraction threshold) is explicitly supervised as background, so P2R's performance should degrade as teacher recall drops; the paper does not measure this sensitivity.
  • Beyond the paper: because the ablation shows MAE is flat as the radius $\mu$ grows from 32 to 256, P2R should transfer to other point-detection tasks such as cell or animal counting with little hyperparameter tuning.
  • Beyond the paper: PSAM is a general diagnostic for dense localization; it could be used in other self-training pipelines to detect when background supervision is being lost, not just in crowd counting.
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 a point-to-region (P2R) loss for semi-supervised point-based crowd counting. It first observes that the standard point-to-point (P2P) matching loss on unlabeled data has an identically zero background term (Eq. 9), which makes training ill-posed. It then introduces PSAM, a gradient-based visualization tool, to show that the unlabeled P2P loss causes over-activation around foreground points. Based on this observation, the paper replaces P2P with P2R: each pseudo-point supervises a local circular region, and pixels farther than a radius \mu from all pseudo-points are treated as confident background. Experiments on ShTech A/B, UCF-QNRF, and JHU++ under 5%, 10%, and 40% labeled protocols, plus unsupervised domain adaptation, report state-of-the-art results and a large speed-up from removing the Hungarian algorithm.

Significance. If the results hold, the paper makes a useful contribution: it identifies a concrete failure mode of pseudo-labeling for point-based counters, provides a correct mathematical diagnosis of the missing background gradient, and offers a simple and faster alternative. The method is clearly specified, the code is released, and the experiments cover four datasets and UDA. The derivation in Supplement Sec. 11 is sound. The main weakness is that the confident-background term in Eq. (23) relies on an untested assumption about pseudo-label recall; the reported gains could be partially due to suppressing missed people as background. The absence of error bars or multiple-seed runs also tempers the state-of-the-art claims.

major comments (2)
  1. [Sec. 5.2, Eq. (23)] The term (1_n - \beta) assigns confidence 1 and pseudo-label 0 to every pixel farther than \mu from all extracted pseudo-points. This is valid only if the teacher's pseudo-labels extracted with the 0.5 threshold in Eq. (6) have near-perfect recall outside those radii. In sparse or semi-dense crowd regions, a person whose teacher score is below 0.5 and who is not within \mu of a detected person is silently converted into a confident background example. Because the teacher is an EMA of the student, this error can be self-reinforcing across iterations. The paper does not quantify teacher recall or sensitivity to it; the \mu ablation in Fig. 6(c) varies the radius with real pseudo-labels and cannot isolate the effect of missing pseudo-points. I request an explicit experiment in which known fractions of ground-truth points are removed before constructing pseudo-labels, with P2R MAE reported as a function of the removal rate, or an alternative formulation that gates the background confidence term on a teacher background-probability estimate.
  2. [Sec. 6, Tables 1-2 and Fig. 6] All reported numbers are single runs without error bars or multiple seeds, and the key hyperparameters \tau, \eta, \mu, and \alpha are selected on the same datasets and protocols used for the final comparisons. This creates a selection-bias risk for the state-of-the-art claims (e.g., 5% ShTech A MAE 69.9 vs. 83.7 for OT-M). I recommend reporting mean and standard deviation over at least three seeds and stating how hyperparameters are chosen, ideally on a held-out validation split rather than the test protocol.
minor comments (5)
  1. [Table 1] In the 10% protocol, the MT [59] row lists MAE 319.3 and MSE 94.5 for ShTech A; this ordering is impossible for non-negative errors and appears to be a typo or table-formatting error. Please correct and re-check all numbers in that row.
  2. [Eq. (7) and Algorithm 2] The sentence 'map to \hat p_t by M_st' is hard to parse; please define the vector z = M_st \zeta explicitly and make the dimensions of \zeta and \beta clear in Eq. (23) and Algorithm 2.
  3. [Algorithms 1-2] The algorithms do not specify behavior when the pseudo-label set is empty (m=0) or when a matched region contains no pixel with positive score; please add a guard or discuss how such cases are handled in practice.
  4. [Sec. 6.1, efficiency comparison] The 68x speed-up of P2R over P2P for loss computation is reported without specifying hardware or whether the timing includes GPU transfer and Hungarian matching; please add these details.
  5. [Table 3] The inverse-sigmoid versus identity ablation is reported on ShTech A only; if space permits, adding a second dataset would make the choice of S(\cdot) more convincing.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: P2R is a training loss motivated by PSAM diagnostics and validated on held-out benchmarks; no prediction reduces to a fitted input or to a self-citation chain.

full rationale

The paper's central claim is that replacing point-to-point matching with point-to-region matching prevents training breakdown in semi-supervised point-based counting and improves held-out MAE/MSE. This is an empirical claim evaluated on ShTech A/B, UCF-QNRF, and JHU++ against external baselines (OT-M, DAC, etc.), not a derived constant that equals its own inputs. The PSAM analysis is a visualization/diagnostic tool, not a theorem smuggled in by citation; the P2R loss is a construction motivated by the observed over-activation, and its success is demonstrated by benchmark numbers. Equation (23)'s assignment of reliable background to pixels farther than mu from every pseudo-point is a modeling assumption about teacher recall, and the paper does not quantify sensitivity to false negatives, but that is a limitation or correctness risk, not circularity: the final predictions are not defined to match the pseudo-labels, and the reported gains are not identities. Hyperparameters (tau, eta, mu, alpha) are tuned on validation sets, which is standard practice and does not make the central benchmark comparison circular. Self-citations (ODAM, OT-M) are prior work by the same authors but are not load-bearing: ODAM is cited as inspiration for gradient-based visual explanation, and OT-M is used as a baseline, not as the justification for P2R. The supplementary proof in Sec. 11 that the P2P unlabeled loss has zero background gradient is a mathematical derivation with stated assumptions, independent of the paper's empirical claims. No step reduces any reported prediction to a fitted value or to an author-imported uniqueness constraint.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on four tuned hyperparameters (mu, tau, eta, alpha) and several domain assumptions about pseudo-label reliability and local-region supervision. No new physical entities are introduced.

free parameters (5)
  • mu (region radius) = not reported; swept 32-256 in Fig. 6(c)
    Defines foreground neighborhoods and reliable background in Eqs. (20) and (23).
  • tau (distance-score balance) = 8 (best in Fig. 6(a))
    Balances l2 distance and inverse-sigmoid score in Eq. (21).
  • eta (confidence threshold) = 0.7 (best in Fig. 6(b))
    Filters pseudo-points in Eq. (7).
  • alpha (unlabeled loss weight) = 2/3 (alpha/(1-alpha)=2, Fig. 6(d))
    Balances labeled and unlabeled losses in Eq. (17).
  • lambda (re-weighting factor) = inherited from P2PNet [57], not reported
    Re-weights matched pixels in BCE loss Eq. (5).
assumptions (4)
  • domain assumption Pseudo-points with score > eta are reliable; pixels farther than mu from all pseudo-points are reliable background.
    Eqs. (7) and (23); if false, the loss suppresses true detections.
  • domain assumption Each person corresponds to a local region around a point, so supervising surrounding pixels as background is beneficial.
    Section 5, paragraph 'Supervising the region around the positive pixel...'; inferred from PSAM, not proven.
  • domain assumption The decoder's output at a pixel depends only on features in its receptive field, allowing PSAM to be computed per local block.
    Section 4.1, Eqs. (12)-(16); valid for standard CNNs.
  • ad hoc to paper Inverse sigmoid scoring S(p) in the cost matrix is a better choice than identity.
    Eq. (21), validated empirically in Table 3; no derivation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting." pith.science (2026). https://pith.science/paper/C6MK65ZN

@misc{pith2026250521943,
  author       = {Pith},
  title        = {Pith review of: Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C6MK65ZN}},
  note         = {Machine review of arXiv:2505.21943}
}
read the original abstract

Point detection has been developed to locate pedestrians in crowded scenes by training a counter through a point-to-point (P2P) supervision scheme. Despite its excellent localization and counting performance, training a point-based counter still faces challenges concerning annotation labor: hundreds to thousands of points are required to annotate a single sample capturing a dense crowd. In this paper, we integrate point-based methods into a semi-supervised counting framework based on pseudo-labeling, enabling the training of a counter with only a few annotated samples supplemented by a large volume of pseudo-labeled data. However, during implementation, the training encounters issues as the confidence for pseudo-labels fails to be propagated to background pixels via the P2P. To tackle this challenge, we devise a point-specific activation map (PSAM) to visually interpret the phenomena occurring during the ill-posed training. Observations from the PSAM suggest that the feature map is excessively activated by the loss for unlabeled data, causing the decoder to misinterpret these over-activations as pedestrians. To mitigate this issue, we propose a point-to-region (P2R) scheme to substitute P2P, which segments out local regions rather than detects a point corresponding to a pedestrian for supervision. Consequently, pixels in the local region can share the same confidence with the corresponding pseudo points. Experimental results in both semi-supervised counting and unsupervised domain adaptation highlight the advantages of our method, illustrating P2R can resolve issues identified in PSAM. The code is available at https://github.com/Elin24/P2RLoss.

Figures

Figures reproduced from arXiv: 2505.21943 by the authors.

Figure 1
Figure 1. The workflow of semi-supervised point-based count [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The generation process of PSAM. In this implementation, the forward process begins with extracting sliding local blocks F˜ according to the receptive field r of D, and then decoding F˜ through (2) in parallel. The final prediction p is generated by stacking pixels lo￾cated at the center of r ˆ r patches denoted as in D˜: F P R hˆwˆc Ñ F˜ P R phwqˆprˆrˆcq , (12) DpF , ˜ Θdq Ñ D˜ P R nˆprˆrq center of patches ÝÝÝÝÝÝÝÝ… view at source ↗
Figure 3
Figure 3. Observations in PSAM. (a) The training process, where [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Difference between P2P and P2R matching. (a) and (d) [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: qualitative comparison with other models. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 7
Figure 7. Figure 7: The comparison of PSAMs among different training [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Computation of Lu in P2R. (a) Input image (b) The 200th epoch (c) The last epoch [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: More visualization of the pseudo-labels. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: Visualization of P2R’s Prediction (# 1). [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Visualization of P2R’s Prediction (#2) [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

86 extracted references · 77 canonical work pages

  1. [1]

    Localization in the crowd with topological con- straints

    Shahira Abousamra, Minh Hoai, Dimitris Samaras, and Chao Chen. Localization in the crowd with topological con- straints. InProceedings of the AAAI Conference on Artificial Intelligence, pages 872–881, 2021. 2

    2021

  2. [2]

    Anomalous event detection and localization in dense crowd scenes.Mul- timedia Tools and Applications, 82(10):15673–15694, 2023

    Areej Alhothali, Amal Balabid, Reem Alharthi, Bander Alzahrani, Reem Alotaibi, and Ahmed Barnawi. Anomalous event detection and localization in dense crowd scenes.Mul- timedia Tools and Applications, 82(10):15673–15694, 2023. 1

    2023

  3. [3]

    Switching convolutional neural network for crowd counting

    Deepak Babu Sam, Shiv Surya, and R Venkatesh Babu. Switching convolutional neural network for crowd counting. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 5744–5752, 2017. 2

    2017

  4. [4]

    Bayesian poisson regression for crowd counting

    Antoni B Chan and Nuno Vasconcelos. Bayesian poisson regression for crowd counting. In2009 IEEE 12th interna- tional conference on computer vision, pages 545–551. IEEE,

  5. [5]

    Counting people with low-level features and bayesian regression.IEEE Trans- actions on image processing, 21(4):2160–2177, 2011

    Antoni B Chan and Nuno Vasconcelos. Counting people with low-level features and bayesian regression.IEEE Trans- actions on image processing, 21(4):2160–2177, 2011. 1

    2011

  6. [6]

    Privacy preserving crowd monitoring: Counting peo- ple without people models or tracking

    Antoni B Chan, Zhang-Sheng John Liang, and Nuno Vascon- celos. Privacy preserving crowd monitoring: Counting peo- ple without people models or tracking. In2008 IEEE confer- ence on computer vision and pattern recognition, pages 1–7. IEEE, 2008. 2

    2008

  7. [7]

    Anchor-based group detection in crowd scenes

    Mulin Chen, Qi Wang, and Xuelong Li. Anchor-based group detection in crowd scenes. In2017 IEEE Interna- tional Conference on Acoustics, Speech and Signal Process- ing (ICASSP), pages 1378–1382. IEEE, 2017. 1

    work page 2017

  8. [8]

    Rethinking spatial invariance of convolutional networks for object counting

    Zhi-Qi Cheng, Qi Dai, Hong Li, Jingkuan Song, Xiao Wu, and Alexander G Hauptmann. Rethinking spatial invariance of convolutional networks for object counting. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 19638–19648, 2022. 2

    work page 2022

Show all 86 references
  1. [9]

    Learning independent instance maps for crowd localization

    Junyu Gao, Tao Han, Qi Wang, Yuan Yuan, and Xuelong Li. Learning independent instance maps for crowd localization. arXiv preprint arXiv:2012.04164, 2020. 2

    2012 arXiv

  2. [10]

    Steerer: Resolving scale variations for counting and localization via selective inheritance learning

    Tao Han, Lei Bai, Lingbo Liu, and Wanli Ouyang. Steerer: Resolving scale variations for counting and localization via selective inheritance learning. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 21848–21859, 2023. 1, 3

    2023

  3. [11]

    Error-aware density isomorphism re- construction for unsupervised cross-domain crowd counting

    Yuhang He, Zhiheng Ma, Xing Wei, Xiaopeng Hong, Wei Ke, and Yihong Gong. Error-aware density isomorphism re- construction for unsupervised cross-domain crowd counting. InProceedings of the AAAI conference on artificial intelli- gence, pages 1540–1548, 2021. 7

    2021

  4. [12]

    Composition loss for counting, density map estima- tion and localization in dense crowds

    Haroon Idrees, Muhmmad Tayyab, Kishan Athrey, Dong Zhang, Somaya Al-Maadeed, Nasir Rajpoot, and Mubarak Shah. Composition loss for counting, density map estima- tion and localization in dense crowds. InProceedings of the European conference on computer vision (ECCV), pages 532...

    2018

  5. [13]

    Ex- plaining convolutional neural networks using softmax gra- dient layer-wise relevance propagation

    Brian Kenji Iwana, Ryohei Kuroki, and Seiichi Uchida. Ex- plaining convolutional neural networks using softmax gra- dient layer-wise relevance propagation. In2019 IEEE/CVF International Conference on Computer Vision Workshop (IC- CVW), pages 4176–4185. IEEE, 2019. 3

    2019

  6. [14]

    Clip- count: Towards text-guided zero-shot object counting

    Ruixiang Jiang, Lingbo Liu, and Changwen Chen. Clip- count: Towards text-guided zero-shot object counting. In Proceedings of the 31st ACM International Conference on Multimedia, pages 4535–4545, 2023. 2

    2023

  7. [15]

    Pseudo-label: The simple and effi- cient semi-supervised learning method for deep neural net- works

    Dong-Hyun Lee et al. Pseudo-label: The simple and effi- cient semi-supervised learning method for deep neural net- works. InWorkshop on challenges in representation learn- ing, ICML, page 896. Atlanta, 2013. 2, 3, 5

    2013

  8. [16]

    Pedes- trian detection in crowded scenes

    Bastian Leibe, Edgar Seemann, and Bernt Schiele. Pedes- trian detection in crowded scenes. In2005 IEEE computer society conference on computer vision and pattern recogni- tion (CVPR’05), pages 878–885. IEEE, 2005. 2

    2005

  9. [17]

    Learning to count objects in images.Advances in neural information process- ing systems, 23, 2010

    Victor Lempitsky and Andrew Zisserman. Learning to count objects in images.Advances in neural information process- ing systems, 23, 2010. 2

    2010

  10. [18]

    Calibrating uncertainty for semi-supervised crowd counting

    Chen Li, Xiaoling Hu, Shahira Abousamra, and Chao Chen. Calibrating uncertainty for semi-supervised crowd counting. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 16685–16695. IEEE, 2023. 2, 7

    2023

  11. [19]

    Semi- supervised crowd counting based on hard pseudo-labels

    Hanxiao Li, Yonghong Song, and Tong Geng. Semi- supervised crowd counting based on hard pseudo-labels. In 2024 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE, 2024. 2

    2024

  12. [20]

    Csrnet: Di- lated convolutional neural networks for understanding the highly congested scenes

    Yuhong Li, Xiaofan Zhang, and Deming Chen. Csrnet: Di- lated convolutional neural networks for understanding the highly congested scenes. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 1091–1100, 2018. 2

    2018

  13. [21]

    An end-to-end transformer model for crowd localization

    Dingkang Liang, Wei Xu, and Xiang Bai. An end-to-end transformer model for crowd localization. InECCV, pages 38–54. Springer, 2022. 1, 2

    2022

  14. [22]

    Direct measure matching for crowd counting

    Hui Lin, Xiaopeng Hong, Zhiheng Ma, Xing Wei, Yun- feng Qiu, Yaowei Wang, and Yihong Gong. Direct measure matching for crowd counting. InProceedings of the Thirti- eth International Joint Conference on Artificial Intelligence, pages = 837–844, publisher = ijcai.org, year = 2021, . 2

    2021

  15. [23]

    Semi-supervised crowd counting via density agency

    Hui Lin, Zhiheng Ma, Xiaopeng Hong, Yaowei Wang, and Zhou Su. Semi-supervised crowd counting via density agency. InACM Multimedia, 2022. 2, 7

    2022

  16. [24]

    Boosting crowd counting via multifaceted attention

    Hui Lin, Zhiheng Ma, Rongrong Ji, Yaowei Wang, and Xi- aopeng Hong. Boosting crowd counting via multifaceted attention. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19628– 19637, 2022. 1, 2

    2022

  17. [25]

    Semi-supervised count- ing via pixel-by-pixel density distribution modelling.arXiv preprint arXiv:2402.15297, 2024

    Hui Lin, Zhiheng Ma, Rongrong Ji, Yaowei Wang, Zhou Su, Xiaopeng Hong, and Deyu Meng. Semi-supervised count- ing via pixel-by-pixel density distribution modelling.arXiv preprint arXiv:2402.15297, 2024. 2

    2024 arXiv

  18. [26]

    Optimal transport mini- mization: Crowd localization on density maps for semi- supervised counting

    Wei Lin and Antoni B Chan. Optimal transport mini- mization: Crowd localization on density maps for semi- supervised counting. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 21663–21673, 2023. 1, 2, 3, 6, 7

    2023

  19. [27]

    A fixed-point approach to unified prompt-based counting

    Wei Lin and Antoni B Chan. A fixed-point approach to unified prompt-based counting. InProceedings of the AAAI Conference on Artificial Intelligence, pages 3468– 3476, 2024. 2

    2024

  20. [28]

    Proximal mapping loss: Understanding loss functions in crowd counting & lo- calization

    Wei Lin, Jia Wan, and Antoni B Chan. Proximal mapping loss: Understanding loss functions in crowd counting & lo- calization. InThe Thirteenth International Conference on Learning Representations, . 2

  21. [29]

    Learning to detect anomaly events in crowd scenes from synthetic data

    Wei Lin, Junyu Gao, Qi Wang, and Xuelong Li. Learning to detect anomaly events in crowd scenes from synthetic data. Neurocomputing, 436:248–259, 2021. 1, 7

    2021

  22. [30]

    Scale- prior deformable convolution for exemplar-guided class- agnostic counting

    Wei Lin, Kunlin Yang, Xinzhu Ma, Junyu Gao, Lingbo Liu, Shinan Liu, Jun Hou, Shuai Yi, and Antoni B Chan. Scale- prior deformable convolution for exemplar-guided class- agnostic counting. InBMVC, page 313, 2022. 2

    2022

  23. [31]

    Point-query quadtree for crowd counting, localization, and more

    Chengxin Liu, Hao Lu, Zhiguo Cao, and Tongliang Liu. Point-query quadtree for crowd counting, localization, and more. InCVPR, pages 1676–1685, 2023. 1, 2

    2023

  24. [32]

    Leveraging unlabeled data for crowd counting by learning to rank

    Xialei Liu, Joost Van De Weijer, and Andrew D Bagdanov. Leveraging unlabeled data for crowd counting by learning to rank. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7661–7669, 2018. 2, 7

    2018

  25. [33]

    Exploiting unlabeled data in cnns by self-supervised learning to rank.IEEE transactions on pattern analysis and machine intelligence, 41(8):1862–1878, 2019

    Xialei Liu, Joost Van De Weijer, and Andrew D Bagdanov. Exploiting unlabeled data in cnns by self-supervised learning to rank.IEEE transactions on pattern analysis and machine intelligence, 41(8):1862–1878, 2019. 2

    2019

  26. [34]

    Semi-supervised crowd counting via self-training on surrogate tasks

    Yan Liu, Lingqiao Liu, Peng Wang, Pingping Zhang, and Yinjie Lei. Semi-supervised crowd counting via self-training on surrogate tasks. InEuropean Conference on Computer Vision, pages 242–259. Springer, 2020. 2, 7

    2020

  27. [35]

    Towards unsupervised crowd counting via regression-detection bi-knowledge transfer

    Yuting Liu, Zheng Wang, Miaojing Shi, Shin’ichi Satoh, Qi- jun Zhao, and Hongyu Yang. Towards unsupervised crowd counting via regression-detection bi-knowledge transfer. In Proceedings of the 28th ACM International Conference on Multimedia, pages 129–137, 2020. 7

    2020

  28. [36]

    Semi-supervised crowd counting via multi-task pseudo-label self-correction strategy.IEEE Transactions on Circuits and Systems for Video Technology, 2024

    Yanbo Liu, Yingxiang Hu, Guo Cao, and Yanfeng Shang. Semi-supervised crowd counting via multi-task pseudo-label self-correction strategy.IEEE Transactions on Circuits and Systems for Video Technology, 2024. 2

    2024

  29. [37]

    Counting people crossing a line using integer programming and local features.IEEE Transactions on Circuits and Systems for Video Technology, 26(10):1955–1969, 2015

    Zheng Ma and Antoni B Chan. Counting people crossing a line using integer programming and local features.IEEE Transactions on Circuits and Systems for Video Technology, 26(10):1955–1969, 2015. 1

    1955

  30. [38]

    Bayesian loss for crowd count estimation with point supervi- sion

    Zhiheng Ma, Xing Wei, Xiaopeng Hong, and Yihong Gong. Bayesian loss for crowd count estimation with point supervi- sion. InProceedings of the IEEE International Conference on Computer Vision, pages 6142–6151, 2019. 1, 2, 7, 13

    2019

  31. [39]

    Learning to count via unbalanced optimal transport

    Zhiheng Ma, Xing Wei, Xiaopeng Hong, Hui Lin, Yunfeng Qiu, and Yihong Gong. Learning to count via unbalanced optimal transport. InProceedings of the AAAI Conference on Artificial Intelligence, pages 2319–2327, 2021. 2

    2021

  32. [40]

    Domain generalization via gradient surgery

    Lucas Mansilla, Rodrigo Echeveste, Diego H Milone, and Enzo Ferrante. Domain generalization via gradient surgery. InProceedings of the IEEE/CVF international conference on computer vision, pages 6630–6638, 2021. 7

    2021

  33. [41]

    De- tection and tracking of groups in crowd

    Riccardo Mazzon, Fabio Poiesi, and Andrea Cavallaro. De- tection and tracking of groups in crowd. In2013 10th IEEE International conference on advanced video and sig- nal based surveillance, pages 202–207. IEEE, 2013. 1

    2013

  34. [42]

    Spa- tial uncertainty-aware semi-supervised crowd counting

    Yanda Meng, Hongrun Zhang, Yitian Zhao, Xiaoyun Yang, Xuesheng Qian, Xiaowei Huang, and Yalin Zheng. Spa- tial uncertainty-aware semi-supervised crowd counting. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15549–15559, 2021. 2, 3, 7

    2021

  35. [43]

    Single domain general- ization for crowd counting

    Zhuoxuan Peng and S-H Gary Chan. Single domain general- ization for crowd counting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 28025–28034, 2024. 7

    2024

  36. [44]

    Ablation-cam: Visual explanations for deep convolutional network via gradient- free localization

    Harish Guruprasad Ramaswamy et al. Ablation-cam: Visual explanations for deep convolutional network via gradient- free localization. Inproceedings of the IEEE/CVF winter conference on applications of computer vision, pages 983– 991, 2020. 3

    2020

  37. [45]

    Learning to count everything

    Viresh Ranjan, Udbhav Sharma, Thu Nguyen, and Minh Hoai. Learning to count everything. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3394–3403, 2021. 2

    2021

  38. [46]

    Tracking-by-counting: Using network flows on crowd density maps for tracking multiple targets.IEEE Transactions on Image Processing, 30:1439– 1452, 2020

    Weihong Ren, Xinchao Wang, Jiandong Tian, Yandong Tang, and Antoni B Chan. Tracking-by-counting: Using network flows on crowd density maps for tracking multiple targets.IEEE Transactions on Image Processing, 30:1439– 1452, 2020. 1

    2020

  39. [47]

    Crowd counting and indi- vidual localization using pseudo square label.IEEE Access,

    Jihye Ryu and Kwangho Song. Crowd counting and indi- vidual localization using pseudo square label.IEEE Access,

  40. [48]

    Grad-cam: Visual explanations from deep networks via gradient-based localization

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization. InProceedings of the IEEE in- ternational conference on computer vision, pages 618–626,

  41. [49]

    Training- free object counting with prompts

    Zenglin Shi, Ying Sun, and Mengmi Zhang. Training- free object counting with prompts. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 323–331, 2024. 2

    2024

  42. [50]

    Crowd counting in the frequency domain

    Weibo Shu, Jia Wan, Kay Chen Tan, Sam Kwong, and An- toni B Chan. Crowd counting in the frequency domain. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 19618–19627, 2022. 2

    2022

  43. [51]

    Crowd counting in the frequency domain

    Weibo Shu, Jia Wan, Kay Chen Tan, Sam Kwong, and An- toni B Chan. Crowd counting in the frequency domain. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 19618–19627, 2022. 2

    2022

  44. [52]

    Generalized char- acteristic function loss for crowd analysis in the frequency domain.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2023

    Weibo Shu, Jia Wan, and Antoni B Chan. Generalized char- acteristic function loss for crowd analysis in the frequency domain.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2023. 2

    2023

  45. [53]

    Deep inside convolutional networks: Visu- alising image classification models and saliency maps.arXiv preprint arXiv:1312.6034, 2013

    Karen Simonyan. Deep inside convolutional networks: Visu- alising image classification models and saliency maps.arXiv preprint arXiv:1312.6034, 2013. 4

    2013 arXiv

  46. [54]

    Learning to count in the crowd from limited labeled data

    Vishwanath A Sindagi, Rajeev Yasarla, Deepak Sam Babu, R Venkatesh Babu, and Vishal M Patel. Learning to count in the crowd from limited labeled data. InEuropean Confer- ence on Computer Vision, pages 212–229. Springer, 2020. 2, 7

    2020

  47. [55]

    Jhu-crowd++: Large-scale crowd counting dataset and a benchmark method.IEEE transactions on pattern analysis and machine intelligence, 44(5):2594–2609, 2020

    Vishwanath A Sindagi, Rajeev Yasarla, and Vishal M Pa- tel. Jhu-crowd++: Large-scale crowd counting dataset and a benchmark method.IEEE transactions on pattern analysis and machine intelligence, 44(5):2594–2609, 2020. 6, 7

    2020

  48. [56]

    Fixmatch: Simplifying semi-supervised learning with consistency and confidence

    Kihyuk Sohn, David Berthelot, Nicholas Carlini, Zizhao Zhang, Han Zhang, Colin A Raffel, Ekin Dogus Cubuk, Alexey Kurakin, and Chun-Liang Li. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. Advances in neural information processing systems, 33:5...

    2020

  49. [57]

    Rethinking counting and localization in crowds: A purely point-based framework

    Qingyu Song, Changan Wang, Zhengkai Jiang, Yabiao Wang, Ying Tai, Chengjie Wang, Jilin Li, Feiyue Huang, and Yang Wu. Rethinking counting and localization in crowds: A purely point-based framework. InCVPR, pages 3365–3374,

  50. [58]

    Riedmiller

    Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, and Martin A. Riedmiller. Striving for simplicity: The all convolutional net. In3rd International Conference on Learning Representations, ICLR 2015, Workshop Track Pro- ceedings, 2015. 4

    2015

  51. [59]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results.Advances in neural information processing systems, 30, 2017

    Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results.Advances in neural information processing systems, 30, 2017. 1, 3, 7

    2017

  52. [60]

    Kernel- based density map generation for dense object counting

    Jia Wan, Qingzhong Wang, and Antoni B Chan. Kernel- based density map generation for dense object counting. IEEE Transactions on Pattern Analysis and Machine Intel- ligence, 44(3):1357–1370, 2020. 2

    2020

  53. [61]

    A generalized loss function for crowd counting and localization

    Jia Wan, Ziquan Liu, and Antoni B Chan. A generalized loss function for crowd counting and localization. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1974–1983, 2021. 1, 2, 3, 13

    1974

  54. [62]

    Robust zero-shot crowd counting and localization with adaptive res- olution sam

    Jia Wan, Qiangqiang Wu, Wei Lin, and Antoni Chan. Robust zero-shot crowd counting and localization with adaptive res- olution sam. InEuropean Conference on Computer Vision, pages 478–495. Springer, 2024. 1

    2024

  55. [63]

    Distribution matching for crowd counting.Ad- vances in neural information processing systems, 33:1595– 1607, 2020

    Boyu Wang, Huidong Liu, Dimitris Samaras, and Minh Hoai Nguyen. Distribution matching for crowd counting.Ad- vances in neural information processing systems, 33:1595– 1607, 2020. 2, 13

    2020

  56. [64]

    Score-cam: Score-weighted visual explanations for convolutional neural networks

    Haofan Wang, Zifan Wang, Mengnan Du, Fan Yang, Zijian Zhang, Sirui Ding, Piotr Mardziel, and Xia Hu. Score-cam: Score-weighted visual explanations for convolutional neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, ...

    2020

  57. [65]

    Learning from synthetic data for crowd counting in the wild

    Qi Wang, Junyu Gao, Wei Lin, and Yuan Yuan. Learning from synthetic data for crowd counting in the wild. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8198–8207, 2019. 2, 7

    2019

  58. [66]

    Nwpu- crowd: A large-scale benchmark for crowd counting and lo- calization.IEEE transactions on pattern analysis and ma- chine intelligence, 43(6):2141–2149, 2020

    Qi Wang, Junyu Gao, Wei Lin, and Xuelong Li. Nwpu- crowd: A large-scale benchmark for crowd counting and lo- calization.IEEE transactions on pattern analysis and ma- chine intelligence, 43(6):2141–2149, 2020. 1

    2020

  59. [67]

    Density- aware curriculum learning for crowd counting.IEEE Trans- actions on Cybernetics, 52(6):4675–4687, 2020

    Qi Wang, Wei Lin, Junyu Gao, and Xuelong Li. Density- aware curriculum learning for crowd counting.IEEE Trans- actions on Cybernetics, 52(6):4675–4687, 2020. 2, 6

    2020

  60. [68]

    Density- aware curriculum learning for crowd counting.IEEE Trans- actions on Cybernetics, 52(6):4675–4687, 2020

    Qi Wang, Wei Lin, Junyu Gao, and Xuelong Li. Density- aware curriculum learning for crowd counting.IEEE Trans- actions on Cybernetics, 52(6):4675–4687, 2020. 1

    2020

  61. [69]

    Pixel-wise crowd understanding via synthetic data.International Jour- nal of Computer Vision, 129(1):225–245, 2021

    Qi Wang, Junyu Gao, Wei Lin, and Yuan Yuan. Pixel-wise crowd understanding via synthetic data.International Jour- nal of Computer Vision, 129(1):225–245, 2021. 2

    2021

  62. [70]

    Dynamic mo- mentum adaptation for zero-shot cross-domain crowd count- ing

    Qiangqiang Wu, Jia Wan, and Antoni B Chan. Dynamic mo- mentum adaptation for zero-shot cross-domain crowd count- ing. InProceedings of the 29th ACM International Confer- ence on Multimedia, pages 658–666, 2021. 7

    2021

  63. [71]

    Zero-shot object counting

    Jingyi Xu, Hieu Le, Vu Nguyen, Viresh Ranjan, and Dim- itris Samaras. Zero-shot object counting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15548–15557, 2023. 2

    2023

  64. [72]

    Cross-view cross-scene multi-view crowd counting

    Qi Zhang, Wei Lin, and Antoni B Chan. Cross-view cross-scene multi-view crowd counting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 557–567, 2021. 7

    2021

  65. [73]

    Single-image crowd counting via multi-column convolutional neural network

    Yingying Zhang, Desen Zhou, Siqin Chen, Shenghua Gao, and Yi Ma. Single-image crowd counting via multi-column convolutional neural network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 589–597, 2016. 1, 2, 6, 7, 13

    2016

  66. [74]

    Chenyang Zhao and Antoni B. Chan. ODAM: gradient- based instance-specific visual explanations for object detec- tion. InICLR, 2023. 2, 3, 4

    2023

  67. [75]

    Crowd anomaly event de- tection in surveillance video based on the evolution of the spatial position relationship feature

    Chunhui Zhao, Zhiyuan Zhang, Jinwen Hu, Dong Wang, Bin Fan, Quan Pan, and Qiang He. Crowd anomaly event de- tection in surveillance video based on the evolution of the spatial position relationship feature. In2018 IEEE 14th In- ternational Conference on Control and Automation ...

    2018

  68. [76]

    Gradient-based instance-specific visual explanations for ob- ject specification and object discrimination.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 2024

    Chenyang Zhao, Janet H Hsiao, and Antoni B Chan. Gradient-based instance-specific visual explanations for ob- ject specification and object discrimination.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 2024. 2, 3

    2024

  69. [77]

    Gradient-based visual explanation for transformer-based clip

    Chenyang Zhao, Kun Wang, Xingyu Zeng, Rui Zhao, and Antoni B Chan. Gradient-based visual explanation for transformer-based clip. InInternational Conference on Ma- chine Learning, pages 61072–61091. PMLR, 2024

    2024

  70. [78]

    Learning deep features for discrimina- tive localization

    Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning deep features for discrimina- tive localization. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2921–2929,

  71. [79]

    Fine-grained fragment diffusion for cross domain crowd counting

    Huilin Zhu, Jingling Yuan, Zhengwei Yang, Xian Zhong, and Zheng Wang. Fine-grained fragment diffusion for cross domain crowd counting. InProceedings of the 30th ACM International Conference on Multimedia, pages 5659–5668,

  72. [80]

    Find gold in sand: Fine-grained similarity min- ing for domain-adaptive crowd counting.IEEE Transactions on Multimedia, 2023

    Huilin Zhu, Jingling Yuan, Xian Zhong, Liang Liao, and Zheng Wang. Find gold in sand: Fine-grained similarity min- ing for domain-adaptive crowd counting.IEEE Transactions on Multimedia, 2023. 7 Point-to-Region Loss for Semi-Supervised Point-Based Crowd Counting Supplementary ...

    2023

  73. [81]

    1 and Algo

    Pseudo Code of P2R In Algo. 1 and Algo. 2, we present the pseudo-code to com- pute the loss for data with ground truth (GT) labels and pseudo-labels, respectively. In Algo. 2, the parts that dif- Method Identity Inv-Sigmoid MAE MSE MAE MSE P2P [57] 52.74 85.60 52.50 82.94 P2R ...

  74. [82]

    For labeled images, we apply horizontal flips to each cropped sample with a probability of 0.5 and randomly resize the images with a scale factor be- tween 0.7 and 1.3

    Implementation Details Data pre-processing.Images in the datasets are cropped into256ˆ256for training. For labeled images, we apply horizontal flips to each cropped sample with a probability of 0.5 and randomly resize the images with a scale factor be- tween 0.7 and 1.3. For u...

  75. [83]

    We present the empirical results in Table 3 to demonstrate its advantage

    Ablation Study on Cost Function In (4), we use the inverse sigmoid function, Sppq“´log ˆ 1 p´1 ˙ ,(24) rather than the identity operator used in vanilla P2PNet [57] for better performance. We present the empirical results in Table 3 to demonstrate its advantage. It shows that ...

  76. [84]

    Comparison of counting losses (100% Label Pct.) since the second term for the background part is set to 0, as shown in (9)

    The Proof of Ill-Posed Unlabeled Loss Under the P2P framework, we demonstrate that the loss function for unlabeled data, formulated as (8), is ill-posed, Loss Point-based ShTech B QNRF FPSfunction counting modelMAE MSE MAE MSE L2 [73]✗ 7.6 13.0 102.0 171.4 1503.8 BL [38]✗ 7.7 ...

  77. [85]

    2:Related Works

    Comparison with Other Losses A theoretical overview about current counting losses is briefly introduced in the first part of Sec. 2:Related Works. Tab. 4 presents the empirical comparison, and P2R achieves better performance. Besides, the main paper provides a brief comparison...

  78. [86]

    Pseudo-Labels In Fig

    Visualization 13.1. Pseudo-Labels In Fig. 8, we illustrate the pipeline of loss computation for P2R. Given the teacher’s predictionPt, the tensorsP1 t andζ are generated by filtering out pixels with values greater than 0.5 andη, respectively. Subsequently, the region-to-point ...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.