Pith. sign in

REVIEW 3 major objections 6 minor 41 references

Expanding Zero-Shot Object Counting with Rich Prompts

T0 review · 3 major / 6 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 RichCount claims that zero-shot object counting improves when simple category labels are replaced with rich, image-specific descriptions and the text and image encoders are aligned in a dedicated first training stage.

desk verdict Real but narrower than it looks: the 15.78 headline uses ChatGPT-4 descriptions generated from the test image itself, yet the text-only variant still beats baselines — a protocol concern on top of a solid paper that deserves review. read the letter →

arxiv 2505.15398 v2 pith:W4R5LQNX submitted 2025-05-21 cs.CV

Huilin Zhu , Senyao Li , Jingling Yuan , Zhengwei Yang , Yu Guo , Wenxuan Liu , Xian Zhong , Shengfeng He This is my paper
classification cs.CV
keywords zero-shotobjectcountingtext-guidedvision-languagealignmentCLIPdensitymapclass-agnosticmultimodallanguagemodelpromptenrichment
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 argues that zero-shot object counting fails with simple category labels because text and image features are misaligned and category names carry too little semantic content. It proposes RichCount, a two-stage training strategy: first enrich category labels into detailed, image-specific descriptions and align those descriptions to visual features with a feed-forward network and an adapter trained under a contrastive margin loss; then freeze those encoders and train a text-conditioned density-map counter. On FSC-147 the method reports a test-set mean absolute error of 15.78, better than the reproduced CLIP-Count baseline at 17.19, and it reports improved transfer to car and crowd datasets. The paper's message is that prompt enrichment alone is not enough; the richer text has to be actively aligned with visual features before it helps counting.

What carries the argument

The central mechanism is a two-stage alignment-and-count pipeline built on CLIP, a pre-trained vision-language encoder. Stage one augments the frozen visual encoder with a feed-forward network and the frozen text encoder with an adapter, and trains them with a contrastive margin loss so that images, category labels, rich descriptions, and negative descriptions separate properly in a shared embedding space. Stage two freezes those aligned encoders and trains an interaction module that treats image features as queries and text features as keys and values, followed by a decoder that regresses density maps; a consistency loss requires predictions from category labels, rich descriptions, and generic 'object' descriptions to agree. This machinery converts a one-step text-to-count problem into a two-step text-alignment-then-count problem.

What would settle it

Take the FSC-147 test split, strip the image-specific portions from the descriptions or replace them with generic category descriptions, and re-run RichCount; if the mean absolute error rises to the level of the category-label baseline, the reported gain is caused by information leaked from the test image into the prompt rather than by the alignment modules.

Watch

Extended reading notes

Core claim

On the paper's own terms, RichCount establishes that the bottleneck in zero-shot counting is cross-modal alignment rather than prompt vocabulary. Replacing a category name like 'bread rolls' with a description such as 'golden-brown bread rolls on two metal baking trays on a stove top' only helps if the text encoder and image encoder are first tuned so that such descriptions sit close to the corresponding image regions. The paper trains a feed-forward network on the visual side and an adapter on the text side using a contrastive loss with a margin, then freezes both encoders and trains an interaction module and decoder to output density maps from category labels, detailed descriptions, and generic 'object' descriptions, with a loss that forces all three prediction modes to agree. The reported result is state-of-the-art zero-shot counting on FSC-147, with the best configuration using image-specific descriptions.

Load-bearing premise

The method's best results rely on an external multimodal language model looking at each test image and producing an accurate, image-specific description of the objects to be counted, which goes beyond the usual zero-shot setting where only a category label is available.

Editorial extensions

If this is right

  • Zero-shot counting can move beyond fixed category labels to arbitrary free-text prompts such as attributes, questions, and spatial descriptions during inference.
  • The same aligned text encoder is shared between the alignment stage and the counting stage, so the counting model does not need to relearn text-vision associations for every new category.
  • Cross-dataset transfer improves: trained on FSC-147, RichCount reports an MAE of 9.91 on CARPK, below the 13.59 baseline, and lower error than CLIP-based baselines on the ShanghaiTech crowd datasets.
  • The consistency loss among category, description, and generic-description predictions makes the model less sensitive to the exact wording of the prompt at test time.

Reading between the lines

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

  • Because the alignment stage is trained only on image-text similarity, it should carry over to other text-conditioned dense prediction tasks such as referring segmentation or open-vocabulary detection; a test would be to take the aligned encoders and swap the counting head for a segmentation head.
  • The reported advantage may partly come from descriptions that reference the test image itself, including its colors, layout, and object locations; a cleaner zero-shot evaluation would hold out such image-specific descriptions or generate them only during training.
  • If a smaller model can generate descriptions of comparable accuracy, the method's dependence on an expensive external multimodal language model could be removed; the paper's own text-only variant is weaker, so the cost-quality trade-off is the decisive open question.
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

3 major / 6 minor

Summary. The paper proposes RichCount, a two-stage framework for zero-shot object counting. In the first stage, a feed-forward network (FFN) and a text adapter are trained with a contrastive loss to align CLIP visual and textual features, using both category labels and MLLM-generated image descriptions as text inputs. In the second stage, the encoders are frozen and an interaction module plus a decoder are trained to produce density maps from diverse textual prompts. Experiments on FSC-147, CARPK, ShanghaiTech, and COUNTBENCH report state-of-the-art zero-shot counting results, with a headline test MAE of 15.78 on FSC-147 obtained with ChatGPT-4 generated descriptions.

Significance. The two-stage alignment recipe (FFN on the visual encoder, adapter on the text encoder, contrastive loss with margin) is a plausible and potentially useful contribution, and the paper's own text-only variant (FSC-147 test MAE 16.37) still beats the reproduced CLIP-Count baseline (17.19), suggesting that the alignment mechanism has some value independent of description augmentation. The paper also provides ablations, cross-dataset transfer experiments, and an analysis of different MLLMs, which are informative. However, the headline result is obtained with test-image-derived descriptions, which compromises the zero-shot claim; the lack of error bars, released prompts, and code limits reproducibility; and the ablation does not cleanly isolate the alignment contribution from the description contribution. These issues are load-bearing for the central claim of state-of-the-art zero-shot counting.

major comments (3)
  1. [Section 3.1, Eq. (4); Table 1] The headline FSC-147 test MAE of 15.78 reported in Table 1 for "RichCount (Ours) Description" is obtained with descriptions T_d = G(I, P_t) generated by ChatGPT-4 from the test image I itself (Eq. (4)). Because the description can encode the target objects' color, location, and layout, the textual prompt carries information that is unavailable to the user-supplied category label in a conventional zero-shot setting. The comparison to prior zero-shot methods (CLIP-Count, CounTX, etc.) that use only category names is therefore not a like-for-like zero-shot evaluation, and the paper's claim of "state-of-the-art performance in zero-shot counting" (Abstract; Sec. 4.2) is overstated. The paper's own "RichCount (Ours) Text" row (test MAE 16.37) is the appropriate zero-shot comparison and still improves on the reproduced CLIP-Count baseline (17.19), but the 0.59 MAE difference between the Description and Text rows is potentially attributable to test-image information rather than to the proposed text-image alignment. Please present the Text-only result as the primary zero-shot claim, or explicitly redefine the protocol to allow image-derived text and compare against methods that use similar image-derived prompts.
  2. [Section 3.1; Table 5] The experimental protocol is not auditable. The exact prompt template P_t for the MLLM is not specified (only a single example appears in Fig. 2), no generated descriptions are released, no code or seeds are provided, and no error bars or multiple runs are reported for any result in Tables 1-5. Since T_d is the output of a proprietary, non-deterministic model (ChatGPT-4), the reported test MAE of 15.78 is a single sample of a random quantity; the 0.59 MAE advantage over the Text variant may be within run-to-run variation. The authors should release the prompt templates and generated descriptions, specify the API version and sampling parameters (e.g., temperature), and report means and variances over multiple runs for the headline comparisons.
  3. [Section 4.3, Table 4] The ablation study in Table 4 does not separate the contribution of the proposed alignment modules (FFN, adapter, contrastive loss) from the contribution of the image-derived descriptions. The text states that "the Adapter ... and the contrastive loss were the most influential factors," but no condition combines these components with text-only descriptions (e.g., category names) during training and inference. Without such a condition, the observed improvement can be attributed either to the alignment mechanism or to the extra information in the descriptions. Please add an ablation with the full alignment pipeline but without Des (and with text-only inference) to isolate the effect of the proposed modules.
minor comments (6)
  1. [Section 4.2-4.3; Table captions] Table captions and cross-references: Sec. 4.2 refers to "Tab. 6" for the CARPK results, which is Table 2; Sec. 4.3 refers to "Tab. 7" and "Tab. 8" for the component and MLLM ablations, which are Tables 4 and 5; the supplementary text has similar mismatches. Please renumber or fix the cross-references.
  2. [Equation (14)] The prose after Eq. (14) introduces an index x that does not appear in the formula; the sums are indexed by a and (a,b). Please clarify the notation or revise the explanatory sentence.
  3. [Figure 5 caption] The right panel is labeled "EACount(Ours)" but the paper's model is RichCount, and "B/L" is not defined. Please correct the label and define the abbreviation.
  4. [Section 4.1] The phrase "Following ClipViT" uses an undefined term; presumably it means CLIP ViT. Also, the statement that the image and text encoders are "fixed" is confusing because the FFN and adapter are trainable; please rephrase.
  5. [Section 4.1; References] FSC-147 is cited as [7] (the RCC paper by Hobley and Prisacariu), but the dataset was introduced in the FamNet paper [22]. Please correct the dataset citation in Sec. 4.1 and in the supplementary material.
  6. [Abstract; Table 4] The abstract's claim of "the first framework to address these limitations" is too strong given prior text-description-based counting work such as CounTX; please soften or substantiate the novelty claim. Additionally, in Table 4, the meaning of the "#" symbols is not defined: the first row has four "#"s and reports baseline-level numbers, while the last row has no "#" and reports the best numbers, making it impossible to know which configuration is the full model. Please define the notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the training/evaluation chain is self-contained, and the ChatGPT-4 description protocol is a benchmark-leakage concern rather than a circular derivation.

full rationale

Walking the paper's derivation chain: Stage 1 trains an FFN and adapter with a contrastive loss (Eqs. 5-6) on text-image similarity; Stage 2 freezes the encoders and trains the interaction module and decoder with density-map MSE plus consistency losses (Eqs. 13-14). Hyperparameters such as the margin and FFN/adapter depth are chosen on validation (Table 8, Table 9), and no parameter is fitted to the test set. The reported test MAE of 15.78 is obtained by evaluating the trained model on unseen FSC-147 classes using ChatGPT-4 descriptions T_d = G(I, P_t) generated from the test image (Eq. 4). This is not a circular step in the mathematical sense: the description is an input to a frozen model, not a fitted quantity, and the paper's own Text-only variant (16.37 MAE) still improves over the reproduced CLIP-Count baseline (17.19). The image-derived description does encode test-image information, which is a legitimate external-validity and benchmark-protocol concern, but it does not make the prediction equivalent to its inputs by construction. The only self-citation ([35], VA-Count) is in related work and is not load-bearing for any claim. Thus, no self-definitional, fitted-as-prediction, or self-citation-reduction pattern is present.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard neural-network training assumptions plus the availability and accuracy of an external MLLM at both training and inference. The only fitted hyperparameters are the contrastive margin, the depth of the alignment modules, and the training schedule. No new theoretical entities or forces are introduced.

free parameters (3)
  • margin m in contrastive loss (Eq. 6) = 1.0
    Selected via validation experiments (Supplementary Table 8); margin of 1.0 gave best FSC-147 val MAE.
  • FFN and adapter depth = 5 layers each
    Chosen via ablation (Supplementary Table 9); Ada-5 + FFN-5 gave best val MAE.
  • Number of training epochs = 200
    Reported in implementation details; not ablated.
assumptions (4)
  • domain assumption CLIP ViT-B/16 embeddings are a sufficient foundation for zero-shot counting
    The whole method builds on frozen CLIP encoders; if CLIP features lack fine-grained discriminability for objects, the alignment cannot recover it. Used throughout Section 3.
  • domain assumption FSC-147 dot annotations are accurate and the density-map sum equals the object count
    Training uses MSE between predicted density maps and ground-truth density maps built from dot annotations (Section 4.1).
  • domain assumption MLLM-generated descriptions are accurate and useful for counting
    Stage 1 and inference assume ChatGPT-4 descriptions correctly identify the target category and add discriminative attributes (Section 3.1).
  • domain assumption Class-disjoint evaluation on FSC-147 remains valid when descriptions are generated from test images
    The zero-shot claim depends on this; but generating descriptions from test images captures image-specific information, weakening the disjointness guarantee (Section 3.1, Eq. 4).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Expanding Zero-Shot Object Counting with Rich Prompts." pith.science (2026). https://pith.science/paper/W4R5LQNX

@misc{pith2026250515398,
  author       = {Pith},
  title        = {Pith review of: Expanding Zero-Shot Object Counting with Rich Prompts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W4R5LQNX}},
  note         = {Machine review of arXiv:2505.15398}
}
read the original abstract

Expanding pre-trained zero-shot counting models to handle unseen categories requires more than simply adding new prompts, as this approach does not achieve the necessary alignment between text and visual features for accurate counting. We introduce RichCount, the first framework to address these limitations, employing a two-stage training strategy that enhances text encoding and strengthens the model's association with objects in images. RichCount improves zero-shot counting for unseen categories through two key objectives: (1) enriching text features with a feed-forward network and adapter trained on text-image similarity, thereby creating robust, aligned representations; and (2) applying this refined encoder to counting tasks, enabling effective generalization across diverse prompts and complex images. In this manner, RichCount goes beyond simple prompt expansion to establish meaningful feature alignment that supports accurate counting across novel categories. Extensive experiments on three benchmark datasets demonstrate the effectiveness of RichCount, achieving state-of-the-art performance in zero-shot counting and significantly enhancing generalization to unseen categories in open-world scenarios.

Figures

Figures reproduced from arXiv: 2505.15398 by the authors.

Figure 1
Figure 1. Illustration of Text-Visual Association. (a) Text-based counting methods (ClipCount [11]) often result in non-specific category estimations, whereas visual prompts (T-Rex [10]) mitigate this issue. A natural modality and semantic gap exists between text and visual prompts. (b) Our method addresses this misalignment, enabling the use of diverse text prompts as inputs. counting. It leverages a small number of annotate… view at source ↗
Figure 2
Figure 2. Overview of the Proposed Method. The framework consists of two training stages: (1) Visual-Text Alignment, which utilizes ChatGPT to generate descriptive text for image categories. To align features, an FFN is added to the CLIP visual encoder, and an adapter is integrated into the CLIP text encoder; (2) Text-Based Counting, which freezes the encoders and trains the interaction module and decoder to ensure consistenc… view at source ↗
Figure 3
Figure 3. Illustration of the Alignment Strategy. An adapter refines text embeddings, and an FFN processes visual features aligned via contrastive loss for cross-modal understanding. Adapter Training for Textual Feature Refinement Sub￾sequently, an adapter is integrated into the text encoder after enhancing the visual features: ft(·) = Adapter(Ct(·)). (9) During this phase, both the FFN and CLIP encoder are frozen, focusing t… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Illustration of Descriptions Generated by ChatGPT-4. (a) Clipcount(B/L) (b) EACount(Ours) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visualization of t-SNE Clusters Before and After Alignment. Misaligned clusters are circled in yellow. their respective clusters, with several image-text pairs failing to establish a correspondence. This suggests that, even when the object described by the text prompt …
Figure 6
Figure 6. Figure 6: Zero-Shot Density Maps on FSC-147. Errors are highlighted in orange. B/L+FFN: FFN in the image encoder; B/L+Adapter: adapter in the text encoder. Class: tested with class labels; Description: tested with image descriptions. Description Question Phrase Green grapes are …
Figure 7
Figure 7. Figure 7: Illustration of Density Maps Generated from Various Texts. “yellow finger foods on the plate”, even when multiple in￾stances of the same category are present. It also shows robustness in identifying objects with distinct visual charac￾teristics, such as sharp edges. An…
Figure 8
Figure 8. Figure 8: Zero-Shot Density Estimation on COUNTBENCH Using Models Trained on FSC-147. Text inputs are sourced from COUNTBENCH test set, with prediction errors highlighted in orange. Predicted values are displayed in white, and ground truth values are indicated in orange. 2 [PIT…
Figure 9
Figure 9. Figure 9: Illustration of CARPK and SHANGHAITECH [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Illustration of FSC-147. margin mitigates boundary ambiguity between positive and negative samples, reducing confusion among visually similar but semantically distinct categories (e.g., green grapes vs. green peas). 3 [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Visualization of Textual Descriptions for FSC-147 Images. The descriptions are generated by ChatGPT-4, ChatGPT-4-turbo, and Claude, with count-related categories highlighted in red. Margin Validation Set Test Set Epoch(Similarity) MAE RMSE MAE RMSE 40 80 100 0.2 18.12…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 31 canonical work pages

  1. [1]

    Open-world text-specified object count- ing

    Niki Amini-Naieni, Kiana Amini-Naieni, Tengda Han, and Andrew Zisserman. Open-world text-specified object count- ing. arXiv:2306.01851, 2023. 2, 5, 6

    arXiv 2023

  2. [2]

    Countgd: Multi-modal open-world counting

    Niki Amini-Naieni, Tengda Han, and Andrew Zisser- man. Countgd: Multi-modal open-world counting. arXiv:2407.04619, 2024. 2, 5

    arXiv 2024

  3. [3]

    The claude 3 model family: Opus, sonnet, haiku,

    Anthropic. The claude 3 model family: Opus, sonnet, haiku,

    work page

  4. [4]

    Lempitsky, and Andrew Zisserman

    Carlos Arteta, Victor S. Lempitsky, and Andrew Zisserman. Counting in the wild. In Proc. Eur. Conf. Comput. Vis., pages 483–498, 2016. 1

    work page 2016

  5. [5]

    Shikra: Unleashing multimodal llm’s referential dialogue magic

    Keqin Chen, Zhao Zhang, Weili Zeng, Richong Zhang, Feng Zhu, and Rui Zhao. Shikra: Unleashing multimodal llm’s referential dialogue magic. arXiv:2306.15195, 2023. 3

    arXiv 2023

  6. [6]

    A low-shot object counting network with iterative prototype adaptation

    Nikola Djukic, Alan Lukezic, Vitjan Zavrtanik, and Matej Kristan. A low-shot object counting network with iterative prototype adaptation. In Proc. IEEE/CVF Int. Conf. Comput. Vis., pages 18826–18835, 2023. 1, 2, 5

    work page 2023

  7. [7]

    Hobley and Victor Prisacariu

    Michael A. Hobley and Victor Prisacariu. Learning to count anything: Reference-less class-agnostic counting with weak supervision. arXiv:2205.10203, 2022. 5, 6, 1

    arXiv 2022

  8. [8]

    Meng-Ru Hsieh, Yen-Liang Lin, and Winston H. Hsu. Drone- based object counting by spatially regularized regional pro- posal network. In Proc. IEEE/CVF Int. Conf. Comput. Vis., pages 4165–4173, 2017. 5, 1

    work page 2017

Show all 41 references
  1. [9]

    Point, segment and count: A general- ized framework for object counting

    Zhizhong Huang, Mingliang Dai, Yi Zhang, Junping Zhang, and Hongming Shan. Point, segment and count: A general- ized framework for object counting. In Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pages 17067–17076, 2024. 2, 5

    2024

  2. [10]

    T-rex: Counting by visual prompting

    Qing Jiang, Feng Li, Tianhe Ren, Shilong Liu, Zhaoyang Zeng, Kent Yu, and Lei Zhang. T-rex: Counting by visual prompting. arXiv:2311.13596, 2023. 1

    2023 arXiv

  3. [11]

    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 Proc. ACM Multimedia, pages 4535–4545, 2023. 1, 2, 5, 6

    2023

  4. [12]

    Vlcounter: Text-aware visual representation for zero- shot object counting

    Seunggu Kang, WonJun Moon, Euiyeon Kim, and Jae-Pil Heo. Vlcounter: Text-aware visual representation for zero- shot object counting. In Proc. AAAI Conf. Artif. Intell., pages 2714–2722, 2024. 2, 5

    2024

  5. [13]

    LISA: reasoning segmentation via large language model

    Xin Lai, Zhuotao Tian, Yukang Chen, Yanwei Li, Yuhui Yuan, Shu Liu, and Jiaya Jia. LISA: reasoning segmentation via large language model. In Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pages 9579–9589, 2024. 3

    2024

  6. [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 Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pages 2893–2903, 2023. 6

    2023

  7. [15]

    Countr: Transformer-based generalised visual counting

    Chang Liu, Yujie Zhong, Andrew Zisserman, and Weidi Xie. Countr: Transformer-based generalised visual counting. In Proc. Brit. Mach. Vis. Conf., page 370, 2022. 1, 2, 5

    2022

  8. [16]

    Grounding DINO: marrying DINO with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, and Lei Zhang. Grounding DINO: marrying DINO with grounded pre-training for open-set object detection. arXiv:2303.05499, 2023. 2

    2023 arXiv

  9. [17]

    Class-agnostic counting

    Erika Lu, Weidi Xie, and Andrew Zisserman. Class-agnostic counting. In Proc. Asian Conf. Comput. Vis., 2018. 1, 2

    2018

  10. [18]

    Nathan Mundhenk, Goran Konjevod, Wesam A

    T. Nathan Mundhenk, Goran Konjevod, Wesam A. Sakla, and Kofi Boakye. A large contextual dataset for classification, detection and counting of cars with deep learning. In Proc. Eur. Conf. Comput. Vis., pages 785–800, 2016. 1

    2016

  11. [19]

    GPT-4 technical report

    OpenAI. GPT-4 technical report. arXiv:2303.08774, 2023. 3, 7, 1

    2023 arXiv

  12. [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 Proc. IEEE/CVF Int. Conf. Comput. Vis. , pages 3147–3157, 2023. 1

    2023

  13. [21]

    Kosmos-2: Grounding multimodal large language models to the world

    Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shao- han Huang, Shuming Ma, and Furu Wei. Kosmos-2: Grounding multimodal large language models to the world. arXiv:2306.14824, 2023. 3

    2023 arXiv

  14. [22]

    Learning to count everything

    Viresh Ranjan, Udbhav Sharma, Thu Nguyen, and Minh Hoai. Learning to count everything. In Proc. IEEE/CVF Conf. Com- put. Vis. Pattern Recognit., pages 3394–3403, 2021. 1, 2, 5, 6

    2021

  15. [23]

    Shaker, Salman H

    Hanoona Abdul Rasheed, Muhammad Maaz, Sahal Shaji Mul- lappilly, Abdelrahman M. Shaker, Salman H. Khan, Hisham Cholakkal, Rao Muhammad Anwer, Eric P. Xing, Ming- Hsuan Yang, and Fahad Shahbaz Khan. Glamm: Pixel ground- ing large multimodal model. In Proc. IEEE/CVF Conf. Com- p...

    2024

  16. [24]

    Pixellm: Pixel reasoning with large multimodal model

    Zhongwei Ren, Zhicheng Huang, Yunchao Wei, Yao Zhao, Dongmei Fu, Jiashi Feng, and Xiaojie Jin. Pixellm: Pixel reasoning with large multimodal model. In Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pages 26364–26373,

  17. [25]

    Sindagi, R

    Deepak Babu Sam, Abhinav Agarwalla, Jimmy Joseph, Vish- wanath A. Sindagi, R. Venkatesh Babu, and Vishal M. Patel. Completely self-supervised crowd counting via distribution matching. In Proc. Eur. Conf. Comput. Vis., pages 186–204,

  18. [26]

    Represent, compare, and learn: A similarity-aware framework for class-agnostic counting

    Min Shi, Hao Lu, Chen Feng, Chengxin Liu, and Zhiguo Cao. Represent, compare, and learn: A similarity-aware framework for class-agnostic counting. In Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pages 9519–9528, 2022. 2, 6

    2022

  19. [27]

    Degpr: Deep guided posterior regularization for multi-class cell detection and counting

    Aayush Kumar Tyagi, Chirag Mohapatra, Prasenjit Das, Govind Makharia, Lalita Mehra, Prathosh AP, and Mausam. Degpr: Deep guided posterior regularization for multi-class cell detection and counting. In Proc. IEEE/CVF Conf. Com- put. Vis. Pattern Recognit., pages 23913–23923, 2023. 1

    2023

  20. [28]

    Visionllm: Large language model is also an open-ended decoder for vision-centric tasks

    Wenhai Wang, Zhe Chen, Xiaokang Chen, Jiannan Wu, Xizhou Zhu, Gang Zeng, Ping Luo, Tong Lu, Jie Zhou, Yu 9 Qiao, and Jifeng Dai. Visionllm: Large language model is also an open-ended decoder for vision-centric tasks. In Adv. Neural Inf. Process. Syst., 2023. 3

    2023

  21. [29]

    Vi- sion transformer off-the-shelf: A surprising baseline for few- shot class-agnostic counting

    Zhicheng Wang, Liwen Xiao, Zhiguo Cao, and Hao Lu. Vi- sion transformer off-the-shelf: A surprising baseline for few- shot class-agnostic counting. In Proc. AAAI Conf. Artif. Intell., pages 5832–5840, 2024. 1, 2, 5

    2024

  22. [30]

    Zero-shot object counting

    Jingyi Xu, Hieu Le, Vu Nguyen, Viresh Ranjan, and Dimitris Samaras. Zero-shot object counting. In Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., pages 15548–15557,

  23. [31]

    Zero-shot object counting with language-vision models

    Jingyi Xu, Hieu Le, and Dimitris Samaras. Zero-shot object counting with language-vision models. arXiv:2309.13097,

    arXiv

  24. [32]

    Hsu, and Wen- Chin Chen

    Shuo-Diao Yang, Hung-Ting Su, Winston H. Hsu, and Wen- Chin Chen. Class-agnostic few-shot object counting. In Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis., pages 869–877,

  25. [33]

    Gpt4roi: Instruction tuning large language model on region-of-interest

    Shilong Zhang, Peize Sun, Shoufa Chen, Min Xiao, Wenqi Shao, Wenwei Zhang, Kai Chen, and Ping Luo. Gpt4roi: Instruction tuning large language model on region-of-interest. arXiv:2307.03601, 2023. 3

    2023 arXiv

  26. [34]

    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 Proc. IEEE/CVF Conf. Com- put. Vis. Pattern Recognit., pages 589–597, 2016. 5, 6, 1

    2016

  27. [35]

    Zero-shot object counting with good exemplars

    Huilin Zhu, Jingling Yuan, Zhengwei Yang, Yu Guo, Zheng Wang, Xian Zhong, and Shengfeng He. Zero-shot object counting with good exemplars. In Proc. Eur. Conf. Comput. Vis., 2024. 2, 5 10 Expanding Zero-Shot Object Counting with Rich Prompts Supplementary Material

    2024

  28. [36]

    2) • Extended visualizations of density maps (Sec

    Overview • Evaluation of performance on C OUNT BENCH (Sec. 2) • Extended visualizations of density maps (Sec. 5) • Analysis of various descriptions (Sec. 4) • Analysis of different margins (Sec. 5) • Analysis of different FFNs and adapters (Sec. 6)

  29. [37]

    Evaluation of performance on COUNT- BENCH Tab. 6 demonstrates the superior performance of the RichCount model compared to CLIP-Count on COUNT- BENCH [20], particularly in its enhanced ability to interpret textual descriptions for counting tasks. RichCount achieves significantl...

  30. [38]

    9 illustrates RichCount’s performance on CARPK [8] and SHANGHAI TECH [34]

    Extended visualizations of density maps Fig. 9 illustrates RichCount’s performance on CARPK [8] and SHANGHAI TECH [34]. The predicted counts (Pre) closely align with the ground truth (Gt) across parking lots on CARPK , demonstrating robustness in structured environ- ments. On ...

  31. [39]

    7 presents an ablation study on FSC-147 , comparing the use of basic category labels ( Class), detailed descrip- tions (Des), and generic terms ( Des-f)

    Analysis of various descriptions Tab. 7 presents an ablation study on FSC-147 , comparing the use of basic category labels ( Class), detailed descrip- tions (Des), and generic terms ( Des-f). Models utilizing detailed descriptions consistently outperform those with sim- pler p...

  32. [40]

    8 illustrates the impact of various margin values on image-text alignment performance during training

    Analysis of different margins Tab. 8 illustrates the impact of various margin values on image-text alignment performance during training. We con- ducted a series of experiments on FSC-147 , testing margin values of 0.2, 0.4, 0.6, 0.8, 1.0, and 1.2 over 100 epochs. Using an FFN...

    1968

  33. [41]

    9 illustrates the impact of various FFN structures on the expressiveness of image and text features

    Analysis of different FFNs and adapters Tab. 9 illustrates the impact of various FFN structures on the expressiveness of image and text features. Deeper or wider FFNs are capable of capturing complex feature rela- tionships, while adapters facilitate fine-grained adjustments t...

Pith tools

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