REVIEW 3 major objections 5 minor 33 references
ProgRoCC: A Progressive Approach to Rough Crowd Counting
T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash
Pith's one-line read Rough count labels—approximate per-image totals instead of point annotations—suffice to train crowd counting that beats semi-supervised methods and rivals fully supervised ones.
desk verdict A genuinely new task framing and a clever CLIP digit-wise counting scheme, but the core claim rests on synthetic ±5% labels standing in for human rough estimates—worth reviewing, not yet settled. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central mechanism is progressive estimation learning (PEL), which turns count prediction into three classification steps—hundreds, tens, and units digits—each over ten text prompts. This reduces inference to 30 contrastive matches, down from up to 1,000 for direct count-matching, and lets the model leverage CLIP's pretrained visual-text alignment. A second component, the visual-language matching adapter (VlMA), maintains $M$ key-value pairs; during training it updates keys and values based on the distance between queried values and text embeddings, and during inference it blends the stored value with the visual embedding before matching. Together, PEL provides the coarse-to-fine supervision and VlMA refines the visual features for more reliable matching.
What would settle it
Collect rough count labels from human annotators on SHA, QNRF, and JHU++ without revealing true counts, train ProgRoCC on these real rough labels, and compare its MAE/MSE against the results obtained with synthetic ±5% labels. If performance degrades toward or below the semi-supervised baselines (for instance, SSL-FT), the equivalence between simulated and real rough labels—and with it the central claim—is not supported.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is that a CLIP model can learn to count crowds from approximate per-image totals if the count is decomposed into digit-wise classification tasks. Instead of matching an image against prompts for every possible count, ProgRoCC predicts the hundreds, tens, and units digits sequentially, using only 30 image-text matches to estimate counts up to 999. A visual-language matching adapter stores effective key-value alignments between visual and textual features and refines the visual embedding at inference. The paper reports that this approach outperforms current semi-supervised and unsupervised counting methods on three benchmarks, and that it transfers across datasets better than several fully supervised methods.
Load-bearing premise
The method's reported advantage depends on the assumption that real human rough estimates vary like uniform random samples within ±5% of the true count; if human estimates are biased, heavy-tailed, or image-dependent, the training signal would differ and the observed gains over semi-supervised methods could shrink.
Editorial extensions
If this is right
- Rough labels could replace point annotations for building large crowd-counting datasets, because annotators only need to give an approximate total per image instead of locating every person.
- Digit-wise progressive prediction lets a CLIP-based counter scale to larger counts with a linear increase in the number of matches: roughly 30 matches for counts under 1,000 and about 300 for counts under 10,000.
- The reported performance on SHA (70.0 MAE) is within 1.8 MAE of fully supervised CSRNet, suggesting that cheap approximate supervision can substitute for exact supervision in some density regimes.
- Cross-dataset experiments show that ProgRoCC trained on SHA transfers to QNRF and JHU++ with lower MAE than fully supervised CSRNet and HMoDE+REL, indicating that rough-label training may generalize better to unseen scenes.
- The method's robustness to label noise—performance stays relatively stable across error sampling ranges from ±15% to ±50%—implies it could tolerate imprecise human estimates in practice.
Reading between the lines
- Editorial inference: the paper's simulation of rough labels as uniform random samples within ±5% of the true count has not been validated against real human rough estimates; human errors are likely biased toward underestimation in dense crowds and may be scene-dependent, so real-world performance could differ.
- Editorial inference: the digit-wise decomposition is a general strategy that could be applied to other vision-language quantity estimation tasks, such as object counting in open-vocabulary settings or estimating lengths, areas, and other scalar attributes from images.
- Editorial inference: if rough labels prove sufficient in practice, crowd counting datasets could be created at web scale by crowdsourcing approximate totals, enabling rapid retraining for new camera geometries, weather conditions, or object categories without per-target annotation.
- Editorial inference: a testable extension is to combine rough count labels with a small number of point annotations, which might close the remaining gap to fully supervised methods while keeping annotation cost low.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces ProgRoCC, a CLIP-based method for 'rough crowd counting,' in which training images are annotated with approximate count labels rather than point-level or exact-count labels. The method predicts a count digit-by-digit (hundreds, tens, units) through a progressive estimation learning strategy, which reduces the number of image-text similarity computations at inference to 30 comparisons for counts up to 999. A visual-language matching adapter is then used to refine visual features using stored key-value pairs. Experiments on SHA, QNRF, and JHU++ report MAEs of 70.0, 114.8, and 92.5, respectively, outperforming the compared semi-supervised and unsupervised methods, and cross-dataset experiments show favorable transferability relative to the same baselines. The central claim is that rough labels, which are claimed to be easier to acquire than point annotations, can provide better accuracy than semi- and weakly-supervised alternatives.
Significance. If validated, the paper would make a useful contribution: it proposes a new annotation paradigm for crowd counting, demonstrates that a CLIP-based progressive digit-classification scheme can learn from noisy count-level labels, and reports consistent gains over several semi-supervised and unsupervised baselines across three datasets. The ablation study isolates the contributions of the progressive strategy and the matching adapter, and the cross-dataset evaluation is a valuable addition. The efficiency gain over direct CLIP matching is also clearly demonstrated. However, the significance is currently tempered by the fact that all rough labels are generated synthetically with a specific uniform noise model, so the paper's motivating claim that rough labels are easier to acquire and behave like human estimates is not directly tested.
major comments (3)
- [Section 4.2 and Table 3] The central claim that rough labels are 'easier to acquire' and yield better accuracy rests on an unvalidated equivalence between synthetic and real rough labels. All experiments use labels drawn uniformly from a ±5% interval around the true count, with n=10 simulated 'expert' estimates and then a training label sampled between the min and max of those estimates. No experiment uses labels produced by human annotators, and the paper's own motivation in Section 1 states that enumeration-based estimates in dense crowds are biased toward underestimation. A realistic error process is therefore likely to be biased, heavy-tailed, or density-dependent rather than symmetric and uniform. If real human rough labels deviate from this model, the progressive digit targets in Section 3.1 become miscalibrated and the reported margins over SSL-FT and SEEM could shrink. Please validate the noise model with a human-annotation study on a subset of images, or at minimum stress-test the method under alternative error models such as multiplicative log-normal noise, systematic undercounting, and density-dependent bias, and report the resulting MAE/MSE.
- [Section 3.2 and Section 4.2] The visual-language matching adapter is a key component of the method (Table 2 shows a 2.1 MAE improvement on SHA), but the manuscript does not specify how the M key-value pairs are initialized, when the adapter is updated relative to the contrastive training in Eq. (5), whether the update rules in Eqs. (6)-(8) are applied per batch or per epoch, or how the text features u_i are obtained for the value update in Eq. (8). The adapter appears to be updated by non-differentiable nearest-neighbor rules, yet it is used to refine visual features during inference, so the training procedure must be described precisely. Without these details or released code, the adapter component is not reproducible and the reported gains cannot be independently verified.
- [Section 3.1 and Table 5] The method hard-limits the counting range to 0-999, and Table 5 shows that ProgRoCC's MSE degrades sharply in the [800,∞) interval (161.1 MSE versus 141.7 for CCTrans), which is consistent with the model being unable to predict counts above 999. It is unclear whether the model operates on whole images resized to 224×224 or on image patches; if whole images are used, high-density images are systematically underestimated by construction. Since QNRF contains many images with more than 1,000 people, the headline QNRF and JHU++ numbers in Table 1 partially reflect this truncation. Please clarify the whole-image versus patch protocol and quantify the effect of the 999 cap, or qualify the claim of robustness on very dense crowds.
minor comments (5)
- [Section 4.4.2] There are several typos, including 'nosiy lables' and 'highly nosiy lables', and 'board knowledge' in Section 1 should be 'broad knowledge'.
- [Section 4.4] In the text of Section 4.4, TransCrowd is cited as reference [27], but TransCrowd is reference [12] in the bibliography; CCTrans is [27]. Please correct the citations.
- [Section 3.1] The text prompt template used for the digit labels is never stated. Please provide the exact template, for example 'There are approximately {number} people in the photo', since the prompt wording directly affects CLIP matching.
- [Section 3.2 and Figure 2] The figure caption says that green straight arrows and black straight arrows denote the training stages of PEL and VlMA, but the figure does not appear to include a legend; please clarify the arrow colors in the figure itself.
- [Section 4.4.2] The statement that performance is 'relatively stable' at error sampling rates between ±15% and ±50% is only weakly supported by Table 3, since the MAE varies between 119.8 and 133.6 in that range; please either quantify this variability or soften the wording.
Circularity Check
No circularity found: the rough-label training signal is a perturbed version of the true count, and the final prediction is a digit-wise CLIP match that is not algebraically forced to equal the training label.
full rationale
ProgRoCC trains a CLIP image encoder using rough labels that are generated by uniformly sampling within ±5% of the true count (Section 4.2), and it evaluates on the same true counts. The inference procedure (Algorithm 1) selects each digit by argmax CLIP similarity over text prompts, producing a count P = 100*P_h + 10*P_t + P_u. Neither the contrastive objective in Eq. (5) nor the inference rule in Eq. (4) has the training rough label as a direct algebraic input; the training label is a noisy approximation, not a deterministic transformation of the test target. No self-citation is load-bearing: the paper cites standard CLIP and crowd-counting baselines but does not rest its central claim on any prior work by its own authors. The only substantive concern is external validity, namely that the ±5% uniform synthetic noise has not been validated against real human rough estimates, so the claimed advantage may not transfer if real estimates are biased or heavy-tailed. That concern is an empirical-validation gap, not a circular derivation, and it does not make the reported MAEs forced by construction. The ablation study selecting ±5% after comparing several ranges is hyperparameter tuning, not a fitted parameter renamed as a prediction. Under the requested standard, the derivation is self-contained and not circular.
Assumptions & free parameters
free parameters (5)
- Error sampling range =
±5%
- Threshold delta =
0.14
- Scaling factor lambda =
0.1
- Number of key-value pairs M =
3000
- Number of sampled experts n =
10
assumptions (3)
- domain assumption CLIP's pretrained image-text alignment transfers to the crowd counting domain.
- domain assumption Every image patch contains at most 999 people, so the digits hundreds, tens, and units cover all counts.
- ad hoc to paper Rough labels uniformly sampled within a fixed percentage of the true count mimic human estimation behavior.
Cite this review
Pith. "Pith review of ProgRoCC: A Progressive Approach to Rough Crowd Counting." pith.science (2026). https://pith.science/paper/6ULVLYDI
@misc{pith2026250413405,
author = {Pith},
title = {Pith review of: ProgRoCC: A Progressive Approach to Rough Crowd Counting},
year = {2026},
howpublished = {\url{https://pith.science/paper/6ULVLYDI}},
note = {Machine review of arXiv:2504.13405}
}
read the original abstract
As the number of individuals in a crowd grows, enumeration-based techniques become increasingly infeasible and their estimates increasingly unreliable. We propose instead an estimation-based version of the problem: we label Rough Crowd Counting that delivers better accuracy on the basis of training data that is easier to acquire. Rough crowd counting requires only rough annotations of the number of targets in an image, instead of the more traditional, and far more expensive, per-target annotations. We propose an approach to the rough crowd counting problem based on CLIP, termed ProgRoCC. Specifically, we introduce a progressive estimation learning strategy that determines the object count through a coarse-to-fine approach. This approach delivers answers quickly, outperforms the state-of-the-art in semi- and weakly-supervised crowd counting. In addition, we design a vision-language matching adapter that optimizes key-value pairs by mining effective matches of two modalities to refine the visual features, thereby improving the final performance. Extensive experimental results on three widely adopted crowd counting datasets demonstrate the effectiveness of our method.
Figures
Reference graph
Works this paper leans on
-
[1]
Completely self-supervised crowd counting via dis- tribution matching
Deepak Babu Sam, Abhinav Agarwalla, Jimmy Joseph, Vishwanath A Sindagi, R Venkatesh Babu, and Vishal M Patel. Completely self-supervised crowd counting via dis- tribution matching. In European Conference on Computer Vision, pages 186–204. Springer, 2022. 2, 3, 7, 8
work page 2022
-
[2]
Multi-task semi-supervised crowd counting via global to local self-correction
Jiwei Chen and Zengfu Wang. Multi-task semi-supervised crowd counting via global to local self-correction. Pattern Recognition, 140:109506, 2023. 2, 3
work page 2023
-
[3]
A primarily se- rial, foveal accumulator underlies approximate numerical es- timation
Samuel J Cheyette and Steven T Piantadosi. A primarily se- rial, foveal accumulator underlies approximate numerical es- timation. Proceedings of the National Academy of Sciences, 116(36):17729–17734, 2019. 2
work page 2019
-
[4]
Redesigning multi-scale neural network for crowd counting
Zhipeng Du, Miaojing Shi, Jiankang Deng, and Stefanos Zafeiriou. Redesigning multi-scale neural network for crowd counting. IEEE Transactions on Image Processing , 32: 3664–3678, 2023. 1, 7, 8
work page 2023
-
[5]
Densitytoken: Weakly-supervised crowd counting with density classifica- tion
Zaiyi Hu, Binglu Wang, and Xuelong Li. Densitytoken: Weakly-supervised crowd counting with density classifica- tion. In IEEE International Conference on Acoustics, Speech and Signal Processing, pages 1–5. IEEE, 2023. 7
work page 2023
-
[6]
Counting crowds in bad weather
Zhi-Kai Huang, Wei-Ting Chen, Yuan-Chun Chiang, Sy- Yen Kuo, and Ming-Hsuan Yang. Counting crowds in bad weather. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 23308–23319, 2023. 7
work page 2023
-
[7]
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. In Proceedings of the European Conference on Computer Vision, pages 532–546,
-
[8]
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 ACM International Conference on Multimedia, pages 4535– 4545, 2023. 3
work page 2023
Show all 33 references
-
[9]
Knowledge-aware prompt tun- ing for generalizable vision-language models
Baoshuo Kan, Teng Wang, Wenpeng Lu, Xiantong Zhen, Weili Guan, and Feng Zheng. Knowledge-aware prompt tun- ing for generalizable vision-language models. In Proceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 15670–15680, 2023. 3
2023
-
[10]
To- wards using count-level weak supervision for crowd count- ing
Yinjie Lei, Yan Liu, Pingping Zhang, and Lingqiao Liu. To- wards using count-level weak supervision for crowd count- ing. Pattern Recognition, 109:107616, 2021. 7
2021
-
[11]
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. In Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 1091–1100, 2018. 1, 6, 7, 8
2018
-
[12]
Transcrowd: Weakly-supervised crowd counting with transformers
Dingkang Liang, Xiwu Chen, Wei Xu, Yu Zhou, and Xiang Bai. Transcrowd: Weakly-supervised crowd counting with transformers. Science China Information Sciences , 65(6): 160104, 2022. 7, 8
2022
-
[13]
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. In European Con- ference on Computer Vision , pages 38–54. Springer, 2022. 3
2022
-
[14]
Crowdclip: Unsupervised crowd counting via vision-language model
Dingkang Liang, Jiahao Xie, Zhikang Zou, Xiaoqing Ye, Wei Xu, and Xiang Bai. Crowdclip: Unsupervised crowd counting via vision-language model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2893–2903, 2023. 2, 3, 7, 8
2023
-
[15]
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. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 1676–1685, 2023. 3
2023
-
[16]
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. In European Conference on Computer Vision, pages 242–259. Springer, 2020. 7, 8
2020
-
[17]
Bayesian loss for crowd count estimation with point super- vision
Zhiheng Ma, Xing Wei, Xiaopeng Hong, and Yihong Gong. Bayesian loss for crowd count estimation with point super- vision. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 6142–6151, 2019. 2
2019
-
[18]
S-clip: Semi-supervised vision-language learning us- ing few specialist captions
Sangwoo Mo, Minkyu Kim, Kyungmin Lee, and Jinwoo Shin. S-clip: Semi-supervised vision-language learning us- ing few specialist captions. Advances in Neural Information Processing Systems, 36:61187–61212, 2023. 3
2023
-
[19]
Crowd counting with decomposed uncertainty
Min-hwan Oh, Peder Olsen, and Karthikeyan Natesan Ra- mamurthy. Crowd counting with decomposed uncertainty. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 11799–11806, 2020. 7
2020
-
[20]
Teaching clip to count to ten
Roni Paiss, Ariel Ephrat, Omer Tov, Shiran Zada, Inbar Mosseri, Michal Irani, and Tali Dekel. Teaching clip to count to ten. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 3170–3180, 2023. 3
2023
-
[21]
Semi-supervised crowd counting with contextual modeling: facilitating holistic un- derstanding of crowd scenes
Yifei Qian, Xiaopeng Hong, Zhongliang Guo, Ognjen Arandjelovi´c, and Carl R Donovan. Semi-supervised crowd counting with contextual modeling: facilitating holistic un- derstanding of crowd scenes. IEEE Transactions on Circuits and Systems for Video Technology, 2024. 3
2024
-
[22]
Learn- ing transferable visual models from natural language super- vision
Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In International Conference on Machine Learning...
2021
-
[23]
Crowd- diff: Multi-hypothesis crowd density estimation using dif- fusion models
Yasiru Ranasinghe, Nithin Gopalakrishnan Nair, Wele Gedara Chaminda Bandara, and Vishal M Patel. Crowd- diff: Multi-hypothesis crowd density estimation using dif- fusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12809– ...
2024
-
[24]
Almost unsupervised learning for dense crowd counting
Deepak Babu Sam, Neeraj N Sajjan, Himanshu Maurya, and R Venkatesh Babu. Almost unsupervised learning for dense crowd counting. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 8868–8875, 2019. 3, 7
2019
-
[25]
Locate, size, and count: accurately re- solving people in dense crowds via detection
Deepak Babu Sam, Skand Vishwanath Peri, Mukun- tha Narayanan Sundararaman, Amogh Kamath, and R Venkatesh Babu. Locate, size, and count: accurately re- solving people in dense crowds via detection. IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 43(8): 2739–27...
2021
-
[26]
Jhu-crowd++: Large-scale crowd counting dataset and a benchmark method
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. 2, 5, 6
2020
-
[27]
Cctrans: Simplifying and improving crowd counting with transformer
Ye Tian, Xiangxiang Chu, and Hongpeng Wang. Cctrans: Simplifying and improving crowd counting with transformer. arXiv preprint arXiv:2109.14483, 2021. 7, 8
2021 arXiv
-
[28]
Robust unsupervised crowd counting and localization with adaptive resolution sam
Jia Wan, Qiangqiang Wu, Wei Lin, and Antoni B Chan. Robust unsupervised crowd counting and localization with adaptive resolution sam. arXiv preprint arXiv:2402.17514,
-
[29]
Distribution matching for crowd counting
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. 1, 3
2020
-
[30]
Self-supervised learning with data-efficient su- pervised fine-tuning for crowd counting
Rui Wang, Yixue Hao, Long Hu, Jincai Chen, Min Chen, and Di Wu. Self-supervised learning with data-efficient su- pervised fine-tuning for crowd counting. IEEE Transactions on Multimedia, 25:1538–1546, 2023. 2, 3, 6, 7
2023
-
[31]
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. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 589–597, 2016. 2, 5, 6, 7
2016
-
[32]
Contrastive learning of medical visual representations from paired images and text
Yuhao Zhang, Hang Jiang, Yasuhide Miura, Christopher D Manning, and Curtis P Langlotz. Contrastive learning of medical visual representations from paired images and text. In Machine Learning for Healthcare Conference , pages 2–
-
[33]
Locality-aware crowd counting
Joey Tianyi Zhou, Le Zhang, Jiawei Du, Xi Peng, Zhiwen Fang, Zhe Xiao, and Hongyuan Zhu. Locality-aware crowd counting. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 44(7):3602–3613, 2022. 6, 7
2022
Reviewed August 16, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.