Pith. sign in

REVIEW 4 major objections 5 minor 44 references

Count2Density: Crowd Density Estimation without Location-level Annotations

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Crowd density maps from counts alone, no point labels

desk verdict A genuinely new count-only density-estimation pipeline, but the spatial-quality claims rest on ablations that never measure spatial error — worth refereeing with that gap fixed. read the letter →

arxiv 2509.03170 v1 pith:PT2CMZ3R submitted 2025-09-03 cs.CV cs.LG

Mattia Litrico , Feng Chen , Michael Pound , Sotirios A Tsaftaris , Sebastiano Battiato , Mario Valerio Giuffrida This is my paper · ORCID
classification cs.CVcs.LG
keywords crowddensityestimationcount-levelsupervisionpseudo-densitymapshistoricalmapbankcontrastivelearningweaklysupervisedsubregioncountingsaliencyinitialization
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

The paper claims that a training pipeline can produce spatially meaningful crowd density maps using only the total number of people per image as supervision. It does so by maintaining a Historical Map Bank of past density predictions, initialized with an unsupervised saliency map, and sampling as many pseudo-points as the count label from that bank via a hypergeometric distribution. A contrastive regularizer pushes the model to separate crowd from background features. If correct, this removes the main annotation bottleneck in density estimation and enables subregion counting without location-level labels. The paper reports large error reductions over cross-domain adaptation baselines, for example from 198.3 to 149.2 Mean Absolute Error on UCF-QNRF.

What carries the argument

Historical Map Bank: a per-image exponential moving average of the model's own past density predictions, initialized with the BASNet saliency estimator, which serves as a probability prior for sampling y_i locations with a hypergeometric distribution to generate the pseudo-density map used for supervision. The contrastive spatial regularizer selects crowded versus background patches from the predicted density map and optimizes a contrastive loss on feature representations, making the learned features spatially consistent.

What would settle it

Take a crowd dataset where people are not visually salient (low contrast, heavy occlusion, or cluttered backgrounds), train Count2Density with only count labels, and compare its density maps against a count-regression baseline and against the same method with the bank seeded by ground-truth point locations; if MAE and subregion-counting error do not improve over regression, the central claim that spatial information can be recovered from counts fails in that regime.

Watch

Extended reading notes

Core claim

On its own terms, Count2Density establishes that count-level supervision is sufficient to learn density maps with quantitative spatial information. The key move is to convert each image's total count into a pseudo-density map: sample exactly y_i locations from a normalized historical average of past predictions, itself initialized by unsupervised saliency, place point masses at those locations, and train the density estimator against this pseudo-map. The bank's exponential moving average acts as an ensemble that dampens confirmation bias. A self-supervised contrastive term encourages the feature extractor to treat crowded areas and background as distinct, so spatial structure survives in the

Load-bearing premise

The whole training loop leans on the assumption that the saliency-seeded Historical Map Bank points at pixels where people actually are; if that spatial prior is wrong in a new domain, the self-training loop could reinforce the error instead of correcting it.

Editorial extensions

If this is right

  • Density estimation models can be trained on datasets that only have total counts, removing the point-annotation bottleneck.
  • Predicted density maps from count-only training support subregion counting by integrating over image tiles.
  • The pseudo-label bank approach works across multiple density-estimation backbones, including NCC, Bayesian loss, generalized loss, and MAN.
  • In a semi-supervised variant with 5–10% location labels, the method outperforms several recent semi-supervised approaches on benchmark datasets.
  • Cross-domain adaptation baselines that use location labels in a source domain are outperformed on dense-scene benchmarks such as UCF-QNRF.

Reading between the lines

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

  • Editorial inference: the sampling-from-bank recipe is generic; any task where total mass is known but locations are not, such as counting cells, animals, or vehicles from aggregate counts, could use the same EMA bank plus hypergeometric sampling.
  • Editorial inference: if saliency is a poor prior for the target domain, the self-training loop could amplify the mismatch; the paper's own ablations suggest the bank still helps with trivial initialization, so the loop itself does substantial work.
  • Editorial inference: storing one map per training image limits scalability, so a compressed or sampled bank, for example storing prototypes or features rather than full maps, is a natural extension.
  • Editorial inference: the stronger performance on dense scenes suggests the hypergeometric sampling prior is better matched to crowded imagery; sparse scenes may need a different sampling distribution.
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

4 major / 5 minor

Summary. The paper proposes Count2Density, a crowd density estimation pipeline trained only with image-level count annotations. It generates pseudo-density maps by sampling a number of points equal to the count label from a Historical Map Bank that stores an EMA of past predictions, initialized with an unsupervised saliency estimator. A contrastive spatial regularizer is added to separate crowd and background features. The method is evaluated on UCF-QNRF, ShanghaiTech-A/B, JHU-Crowd++, and NWPU-Crowd using multiple backbones, and compared with cross-domain and semi-supervised baselines; ablations and qualitative pseudo-map evolution are also provided. The main claim is that count-level supervision, combined with the bank and regularizer, is sufficient to recover quantitative spatial density information, enabling subregion counting and localization.

Significance. If the central claim is substantiated, the contribution is significant: it would reduce annotation cost to count-level labels while retaining density-map outputs, a step beyond regression-based weakly supervised counting that does not support subregion counting. The paper's strengths are its extensive empirical scope (four datasets, multiple backbones), the clear integration of a historical bank and contrastive regularizer, the inclusion of subregion counting and localization evaluation (Tables 3-4), and honest ablation/initialization studies (Tables 5-6). However, the central claim about recovering spatial information is currently supported only by full-pipeline spatial metrics; the component ablations do not measure spatial accuracy, and several headline comparisons are not controlled for the supervision actually used. The method is novel and plausible, but the evidence as presented is not yet sufficient to establish the advertised superiority.

major comments (4)
  1. [Section 4.2, Tables 5 and 6] The central claim is that Count2Density 'infers detailed spatial density information from count-level supervision.' The ablations that attribute this to the Historical Map Bank, saliency initialization, and contrastive regularizer are evaluated only with MAE/MSE (Tables 5 and 6), which are global count errors and are insensitive to the spatial placement of density. A configuration could achieve the same MAE while producing a centered blob instead of the true crowd distribution. Spatial metrics (SSIM/PSNR, subregion counting, localization F1) are reported only for the full pipeline (Tables 3-4), not for ablated variants. This is especially consequential because Table 6 shows that replacing BASNet with a centered blob changes MAE by only ~1.5%, so the experiments cannot discriminate whether the initial spatial prior actually contributes to spatial accuracy. Please report subregion counting
  2. [Section 4.1, Table 2, 'Comparisons with Cross-Domain Approaches'] The comparison with cd-cc and bla is confounded by annotation budget: Count2Density is trained with target-domain count labels for every image, whereas cd-cc and bla use no target-domain labels at all. The advertised 49.1 MAE reduction on UCF-QNRF (198.3 vs 149.2) is therefore not an equivalent-setting comparison. The text's remark that cross-domain methods use source location labels does not remove this confound, because target count labels are still a form of target supervision. Moreover, on ShT-B the cross-domain baselines are better (cd-cc 11.4, bla 11.9 vs Count2Density 15.5), which is acknowledged in the text but not in the abstract. Please either add a target-unsupervised variant of Count2Density or recast the comparison explicitly as 'using count-level target annotations versus no target annotations.'
  3. [Section 4.1, Table 2, 'Comparisons with Semi-Supervised Approaches'] The claim that Count2Density 'achieves better results than recent state-of-the-art approaches in semi-supervised settings' is not supported for the count-only configuration. In Table 2, Count2Density(man) has MAE 149.2 on UCF-QNRF, while irast achieves 135.6 and pal 128.1; on ShT-A several semi-supervised methods (l2r 90.3, irast 86.9, matt 80.1, pal 72.7, cu 70.7) beat Count2Density's 91.5; on ShT-B, cu (9.7) beats 15.5. The outperformance appears only for the Count2Density(Semi-sup.) variant that additionally receives 5-10% location-level labels. Please separate the claims: one for count-only training and one for the semi-supervised variant with a subset of location labels.
  4. [Section 3.2, Eq. (2) and Figure 3] The pseudo-label generation is a self-training loop: the model's own predictions (averaged in the Historical Map Bank) generate the training targets, and the only external spatial grounding is the saliency initialization. Count labels constrain only the total mass, not the location of the density. The paper does not analyze the confirmation-bias risk beyond invoking EMA and the bank. Given Table 6's near-insensitivity of MAE to the initialization, it remains unclear what prevents the loop from converging to a count-accurate but spatially generic prediction. I suggest an additional experiment in which the bank is frozen at the BASNet initialization (no online updates) to isolate the contribution of the self-training dynamics, and reporting spatial metrics for that variant.
minor comments (5)
  1. [Eq. (1)] The summation bounds use W for the horizontal index and H for the vertical index; please make the notation consistent with the map dimensions (H x W) and clarify the coordinate convention.
  2. [Table 4 description] The text says 'identifying local minima' when detecting people from density maps; this should presumably be local maxima, since people correspond to peaks in density.
  3. [Section 3.2] The sampling procedure is described as a 'hypergeometric distribution' and 'equivalent to performing yi Bernoulli samplings without replacement.' Weighted sampling without replacement is the more standard and precise description; the hypergeometric distribution typically refers to the number of successes in draws without replacement, not to the location-sampling operation itself.
  4. [Section 4.2] Minor grammatical issues: 'the such improvement is marginal' should be 'such improvement is marginal'; 'corrispond' should be 'correspond'; 'BASNET' should be 'BASNet' for consistency.
  5. [Table 2] Some rows have inconsistent symbol usage (e.g., the † marker is defined only in the caption for irast). Please ensure all markers are explained and that the 'Semi-supervised' vs 'Cross-Domain' categories are visually distinct for readers.

Circularity Check

1 steps flagged · score 5.0 of 10

Spatial supervision is a closed loop: pseudo-density targets are generated from an EMA of the model's own predictions, so the claimed inference of spatial structure from counts is partially self-referential.

  1. self definitional [Section 3.2, Eq. (2); Section 3.4, Eq. (4)]
    "Specifically, each entry Hi in the historical map bank is updated at each epoch t by calculating an exponential moving average of the predicted density map M̂i as follows: H_i^(t) = α M̂_i^(t) + (1−α) H_i^(t−1) ... to generate the pseudo-density maps, we sample ... number of locations equal to the count label yi ... The sampling is performed using a hypergeometric distribution M(y_i, dH_i^(t)) ... The generated pseudo-density map is then used to train the model in a self-supervised fashion."

    The pseudo-density target is sampled from the historical map bank H_i, which is defined as an EMA of the model's own predicted density maps M̂_i (Eq. 2). The total loss L_map then trains M̂_i to match this pseudo-density map, so the spatial supervision is a function of the model's own output. The count label only fixes the number of sampled points; the spatial locations are drawn from a probability map whose structure originates from the BASNet saliency initialisation and the model's own past predictions. Thus the claim that the method 'retrieves quantitative spatial information from count-level annotations' is not a derivation from counts: the spatial pattern is, by construction, a self-confirmed version of the model's output, seeded by saliency. This is a self-training loop, so the centr

full rationale

The main circularity concern is the self-referential training loop: the pseudo-density maps that supervise the density head are generated from the Historical Map Bank, which is updated as an EMA of the model's own predictions (Eq. 2). Consequently, the spatial information in the training targets is not derived from the count-level labels (which only determine the number of sampled points), but is inherited from the BASNet initialisation and the model's own outputs. This makes the central claim of 'inferring detailed spatial density information from count-level supervision' partially circular by construction. However, the method is not fully circular: it is grounded by external count annotations and an unsupervised saliency prior, and it is evaluated on held-out data. I found no load-bearing self-citation: the two self-citations (refs [24] and [35]) are used for related-work context and a hyperparameter, not to justify the central result. The self-training loop is explicit and standard, but it does undermine the attribution of spatial information to count labels. Score 5 reflects partial circularity rather than complete equivalence.

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

The central claim rests on the assumption that an unsupervised saliency map can provide a reliable spatial prior, and that iterative sampling from the model's own predictions (with EMA stabilization) yields useful pseudo-labels. The free parameters are few, but the unspecified Gaussian sigma and contrastive threshold add hidden degrees of freedom.

free parameters (4)
  • EMA update rate alpha = 0.7
    Chosen via grid search (Section 4, Implementation Details).
  • Contrastive temperature tau = 0.07
    Taken from reference [35], not tuned here.
  • Gaussian smoothing applied to historical map after normalization = not specified
    Section 3.2 mentions applying a Gaussian filter but does not report sigma; this affects pseudo-label noise.
  • Threshold for crowd vs background in contrastive pairs = not specified
    Section 3.3 says 'apply a threshold' but no value is given.
assumptions (4)
  • domain assumption Unsupervised saliency from BASNet correlates with crowd locations
    Used to initialize the historical map bank (Section 3.2); if wrong, pseudo-labels are mislocated.
  • domain assumption Sampling y_i points from the normalized historical map without replacement yields a pseudo-density map whose spatial distribution approximates the true crowd layout
    Core of the pseudo-label generation (Section 3.2).
  • domain assumption EMA of past predictions reduces confirmation bias in self-training
    Claimed in Section 3.2, motivated by DivideMix [28].
  • standard math Standard EMA and hypergeometric distribution formulas
    Used in eq. (2) and sampling without replacement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Count2Density: Crowd Density Estimation without Location-level Annotations." pith.science (2026). https://pith.science/paper/PT2CMZ3R

@misc{pith2026250903170,
  author       = {Pith},
  title        = {Pith review of: Count2Density: Crowd Density Estimation without Location-level Annotations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PT2CMZ3R}},
  note         = {Machine review of arXiv:2509.03170}
}
read the original abstract

Crowd density estimation is a well-known computer vision task aimed at estimating the density distribution of people in an image. The main challenge in this domain is the reliance on fine-grained location-level annotations, (i.e. points placed on top of each individual) to train deep networks. Collecting such detailed annotations is both tedious, time-consuming, and poses a significant barrier to scalability for real-world applications. To alleviate this burden, we present Count2Density: a novel pipeline designed to predict meaningful density maps containing quantitative spatial information using only count-level annotations (i.e., total number of people) during training. To achieve this, Count2Density generates pseudo-density maps leveraging past predictions stored in a Historical Map Bank, thereby reducing confirmation bias. This bank is initialised using an unsupervised saliency estimator to provide an initial spatial prior and is iteratively updated with an EMA of predicted density maps. These pseudo-density maps are obtained by sampling locations from estimated crowd areas using a hypergeometric distribution, with the number of samplings determined by the count-level annotations. To further enhance the spatial awareness of the model, we add a self-supervised contrastive spatial regulariser to encourage similar feature representations within crowded regions while maximising dissimilarity with background regions. Experimental results demonstrate that our approach significantly outperforms cross-domain adaptation methods and achieves better results than recent state-of-the-art approaches in semi-supervised settings across several datasets. Additional analyses validate the effectiveness of each individual component of our pipeline, confirming the ability of Count2Density to effectively retrieve spatial information from count-level annotations and enabling accurate subregion counting.

Figures

Figures reproduced from arXiv: 2509.03170 by the authors.

Figure 1
Figure 1. Overview of Count2Density. (a) The input image xi is provided to the backbone to predict a density map Mˆ i . (b) xi is provided to an unsupervised saliency estimator to initialise the Historical Map Bank H (Section 3.2). The historical map bank is updated at each epoch, using an Exponential Moving Average (Equation (2)) of the predicted density map Mˆ i . For each image xi , the information in H is retrieved and Hi… view at source ↗
Figure 2
Figure 2. Density maps predicted by Count2Density compared to ground-truth density, showing both global and subregion counting. semi-supervised methods. Unlike Count2Density, these methods require a subset of the training set labelled with location-level annotations. Additionally, some semi-supervised approaches [22] augment the dataset by cropping images into 14 [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Pseudo-density maps generated during training. At the beginning, the pseudo-map is generated [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 42 canonical work pages

  1. [1]

    Lempitsky, A

    V . Lempitsky, A. Zisserman, Learning to count objects in images, Advances in neural information processing systems 23 (2010). 2, 8

    work page 2010

  2. [2]

    Y . Lei, Y . Liu, P. Zhang, L. Liu, Towards using count-level weak supervision for crowd counting, Pattern Recognition 109 (2021) 107616. 2, 3, 5, 13

    work page 2021

  3. [3]

    Zhang, D

    Y . Zhang, D. Zhou, S. Chen, S. Gao, Y . Ma, Single-image crowd counting via 19 multi-column convolutional neural network, in: 2016 IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), 2016, pp. 589–597. 2, 3, 12, 13

    work page 2016

  4. [4]

    Jaiswal, A

    S. Jaiswal, A. S. Gadgil, A. M. Kaslikar, K. S. Kothari, Comprehensive study of various methods for estimating crowd density, in: S. D. P. Ragavendiran, V . D. Pavaloaia, M. S. Mekala, A. S. Cabezuelo (Eds.), Innovations and Advances in Cognitive Systems, Springer Nature Switzerland, Cham, 2024, pp. 383–400. 2, 3

    work page 2024

  5. [5]

    X. Liu, J. van de Weijer, A. D. Bagdanov, Leveraging unlabeled data for crowd counting by learning to rank, 2018 IEEE /CVF Conference on Computer Vision and Pattern Recognition (2018) 7661–7669. 2, 5, 13

    work page 2018

  6. [6]

    W. Liu, N. Durasov, P. Fua, Leveraging self-supervision for cross-domain crowd counting, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR) (2021) 5331–5342. 2, 4, 13, 15

    work page 2022

  7. [7]

    Liang, J

    D. Liang, J. Xie, Z. Zou, X. Ye, W. Xu, X. Bai, Crowdclip: Unsupervised crowd counting via vision-language model, in: 2023 IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE Computer Society, Los Alamitos, CA, USA, 2023, pp. 2893–2903. 2, 5

    work page 2023

  8. [8]

    Xiong, L

    Z. Xiong, L. Chai, W. Liu, Y . Liu, S. Ren, S. He, Glance to count: Learning to rank with anchors for weakly-supervised crowd counting, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 343–352. 2, 4, 6

    work page 2024

Show all 44 references
  1. [9]

    Z. Ma, X. Wei, X. Hong, Y . Gong, Bayesian loss for crowd count estimation with point supervision, 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019) 6141–6150. 4, 11, 12, 13

    2019

  2. [10]

    H. Lin, Z. Ma, R. Ji, Y . Wang, X. Hong, Boosting crowd counting via multi- faceted attention, 2022 IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022) 19596–19605. 4, 11, 12, 13, 15

    2022

  3. [11]

    Y . Liu, L. Liu, P. Wang, P. Zhang, Y . Lei, Semi-supervised crowd counting via self-training on surrogate tasks, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm 20 (Eds.), Computer Vision – ECCV 2020, Springer International Publishing, Cham, 2020, pp. 242–259. 4, 5, 12, 13, 15, 16

    2020

  4. [12]

    V . A. Sindagi, R. Yasarla, D. S. Babu, R. V . Babu, V . M. Patel, Learning to count in the crowd from limited labeled data, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Computer Vision – ECCV 2020, Springer International Publishing, Cham, 2020, pp. 212–229. 4, 5, 13

    2020

  5. [13]

    S. Gong, S. Zhang, J. Yang, D. Dai, B. Schiele, Bi-level alignment for cross- domain crowd counting, in: Proceedings of the IEEE /CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2022, pp. 7542–7550. 4, 6, 13, 15

    2022

  6. [14]

    Wang et al., Joint CNN and transformer network via weakly supervised learning for efficient crowd counting, arXiv:2203.06388 (2022). 4

    2022 arXiv

  7. [15]

    Idrees, M

    H. Idrees, M. Tayyab, K. Athrey, D. Zhang, S. A. Al-Maadeed, N. M. Rajpoot, M. Shah, Composition loss for counting, density map estimation and localization in dense crowds, in: European Conference on Computer Vision, 2018. 3, 12, 13

    2018

  8. [16]

    V . A. Sindagi, R. Yasarla, V . M. Patel, JHU-CROWD ++: Large-scale crowd counting dataset and a benchmark method, IEEE Transactions on Pattern Analy- sis and Machine Intelligence 44 (2020) 2594–2609. 3, 12, 13

    2020

  9. [17]

    Z. Yan, R. Zhang, H. Zhang, Q. Zhang, W. Zuo, Crowd counting via perspective- guided fractional-dilation convolution, IEEE Transactions on Multimedia 24 (2021) 2633–2647. 5

    2021

  10. [18]

    Zhang, D

    Y . Zhang, D. Zhou, S. Chen, S. Gao, Y . Ma, Single-image crowd counting via multi-column convolutional neural network, in: Proceedings of the IEEE Confer- ence on Computer Vision and Pattern Recognition, 2016, pp. 589–597. 5

    2016

  11. [19]

    W. Liu, M. Salzmann, P. Fua, Context-aware crowd counting, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 5099–5108. 5 21

    2019

  12. [20]

    S. Yang, W. Guo, Y . Ren, Crowdformer: An overlap patching vision transformer for top-down crowd counting, in: International Joint Conference on Artificial Intelligence, 2022. 5

    2022

  13. [21]

    Ranasinghe, N

    Y . Ranasinghe, N. G. Nair, W. G. C. Bandara, V . M. Patel, CrowdDi ff: Multi-Hypothesis Crowd Density Estimation Using Di ffusion Models, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 12809–12819. doi:10.1109/CVPR52733.2024.01217. 5

    2024

  14. [22]

    Z. Zhao, M. Shi, X. Zhao, L. Li, Active crowd counting with limited supervision, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Computer Vision – ECCV 2020, Springer International Publishing, Cham, 2020, pp. 565–581. 6, 13, 14

    2020

  15. [23]

    Y . Xu, Z. Zhong, D. Lian, J. Li, Z. Li, X. Xu, S. Gao, Crowd counting with partial annotations in an image, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 15550–15559. 6, 13

    2021

  16. [24]

    Litrico, S

    M. Litrico, S. Battiato, S. A. Tsaftaris, M. V . Giuffrida, Semi-supervised domain adaptation for holistic counting under label gap, Journal of Imaging 7 (10) (2021). 6

    2021

  17. [25]

    Y . Yang, G. Li, Z. Wu, L. Su, Q. Huang, N. Sebe, Weakly-supervised crowd counting learns from sorting rather than locations, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Computer Vision – ECCV 2020, Springer Interna- tional Publishing, Cham, 2020, pp. 1–17. 6, 13

    2020

  18. [26]

    Kumagai, K

    S. Kumagai, K. Hotta, T. Kurita, Mixture of counting CNNs: Adaptive integration of CNNs specialized to specific appearance for crowd counting, arXiv preprint arXiv:1703.09393 (2017). 6

    2017 arXiv

  19. [27]

    Liang, X

    D. Liang, X. Chen, W. Xu, Y . Zhou, X. Bai, Transcrowd: weakly-supervised crowd counting with transformers, Science China Information Sciences 65 (6) (2022) 160104. 6 22

    2022

  20. [28]

    J. Li, R. Socher, S. C. H. Hoi, Dividemix: Learning with noisy labels as semi- supervised learning, ArXiv abs/2002.07394 (2020). 9

    2002 arXiv

  21. [29]

    Cheng, N

    M.-M. Cheng, N. J. Mitra, X. Huang, P. H. S. Torr, S.-M. Hu, Global contrast based salient region detection, IEEE Transactions on Pattern Analysis and Ma- chine Intelligence 37 (3) (2015) 569–582. 10

    2015

  22. [30]

    Zhang, D

    P. Zhang, D. Wang, H. Lu, H. Wang, X. Ruan, Amulet: Aggregating multi-level convolutional features for salient object detection, 2017 IEEE International Con- ference on Computer Vision (ICCV) (2017) 202–211. 10

    2017

  23. [31]

    X. Qin, Z. Zhang, C. Huang, C. Gao, M. Dehghan, M. Jagersand, Basnet: Boundary-aware salient object detection, in: 2019 IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 7471–7481. 10, 18, 19

    2019

  24. [32]

    van den Oord, Y

    A. van den Oord, Y . Li, O. Vinyals, Representation learning with contrastive pre- dictive coding, ArXiv abs/1807.03748 (2018). 10

    2018 arXiv

  25. [33]

    J. Wan, A. Chan, Modeling noisy annotations for crowd counting, in: H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, H. Lin (Eds.), Advances in Neural Information Processing Systems, V ol. 33, Curran Associates, Inc., 2020, pp. 3386–3396. 11, 12, 13, 15

    2020

  26. [34]

    J. Wan, Z. Liu, A. B. Chan, A generalized loss function for crowd counting and localization, in: 2021 IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 1974–1983. 11, 12, 13, 15

    2021

  27. [35]

    Litrico, A

    M. Litrico, A. D. Bue, P. Morerio, Guiding pseudo-labels with uncertainty es- timation for source-free unsupervised domain adaptation, in: 2023 IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE Com- puter Society, Los Alamitos, CA, USA, 2023, pp. 764...

    2023

  28. [36]

    Q. Wang, J. Gao, W. Lin, X. Li, Nwpu-crowd: A large-scale benchmark for crowd counting and localization, IEEE Transactions on Pattern Analysis and Machine Intelligence (2020). doi:10.1109/TPAMI.2020.3013269. 12, 13 23

    2020

  29. [37]

    Q. Xie, Z. Dai, E. Hovy, M.-T. Luong, Q. V . Le, Unsupervised data augmentation for consistency training, in: Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS’20, Curran Associates Inc., Red Hook, NY , USA, 2020. 13

    2020

  30. [38]

    Tarvainen, H

    A. Tarvainen, H. Valpola, Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results, in: Pro- ceedings of the 31st International Conference on Neural Information Process- ing Systems, NIPS’17, Curran Associates Inc....

    2017

  31. [39]

    Verma, A

    V . Verma, A. Lamb, J. Kannala, Y . Bengio, D. Lopez-Paz, Interpolation consis- tency training for semi-supervised learning, in: Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19, International Joint Conferences on Artificial ...

    2019

  32. [40]

    C. LI, X. Hu, S. Abousamra, C. Chen, Calibrating uncertainty for semi-supervised crowd counting, in: Proceedings of the IEEE /CVF International Conference on Computer Vision (ICCV), 2023, pp. 16731–16741. 13

    2023

  33. [41]

    T. Han, J. Gao, Y . Yuan, Q. Wang, Focus on semantic consistency for cross- domain crowd understanding, ICASSP 2020 - 2020 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP) (2020) 1848–1852. 13

    2020

  34. [42]

    Han, et al., Steerer: Resolving scale variations for counting and localization via selective inheritance learning, in: ICCV , 2023. 16

    2023

  35. [43]

    Wan, et al., Robust unsupervised crowd counting and localization with adaptive resolution sam, ArXiv abs/2402.17514 (2024)

    J. Wan, et al., Robust unsupervised crowd counting and localization with adaptive resolution sam, ArXiv abs/2402.17514 (2024). 16

    2024 arXiv

  36. [44]

    18, 19 24

    Zhou et al., Texture-guided saliency distilling for unsupervised salient object de- tection, CVPR (2023). 18, 19 24

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.