Pith. sign in

REVIEW 2 major objections 2 minor 34 references

ViViD-5K: Vineyard vision dataset for field-based berry detection and segmentation and grape cluster closure estimation

T0 review · 2 major / 2 minor · reviewed 2026-06-30 · grok-4.3

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

Pith's one-line read A new dataset of 5,000 vineyard images with 648,000 berry annotations enables automated estimation of grape cluster closure.

desk verdict ViViD-5K supplies a sizable new berry-annotated vineyard dataset and a SAM-based pipeline, but missing annotation validation details undercuts the accuracy and generalization claims. read the letter →

arxiv 2605.24353 v1 pith:J6VCDZZA submitted 2026-05-23 cs.CV q-bio.OT

Xiangzhi Tong , Chengrui Zhang , Mac Flaherty , Andre Matteo Garcia , Dominic Gorman , Jonathan Jaramillo , Justine E. Vanden Heuvel , Yu Jiang This is my paper
classification cs.CVq-bio.OT
keywords vineyarddatasetberrydetectiongrapeclustersegmentationclosureestimationcomputervisionagriculturalimagingsegmentanythingmodel
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 establishes that fine-grained berry-level data has been missing for vineyard computer vision, which has blocked objective measurement of cluster closure. Cluster closure tracks how gaps between berries fill in and directly influences disease risk, yet current scoring relies on slow, subjective human judgment. By releasing ViViD-5K with dense centroid and mask labels across thirteen varieties, the work supplies the training resource needed for models that localize individual berries and delineate clusters. A two-stage pipeline then uses these labels to produce closure estimates from field images with little additional supervision. The reported counting and segmentation accuracy plus out-of-domain visualizations are presented as evidence that the approach can scale to practical vineyard use.

What carries the argument

ViViD-5K dataset paired with the GrapeSAM two-stage pipeline that links point-based berry localization to prompt-driven segmentation and transformer cluster delineation.

What would settle it

Models trained on ViViD-5K produce berry counts or closure scores that deviate substantially from human ground truth when tested on images from a vineyard using grape varieties or lighting conditions absent from the thirteen varieties in the released set.

Watch

Extended reading notes

Core claim

ViViD-5K supplies 5,000 field images containing over 648,000 manually marked berry centroids together with cluster segmentation masks spanning thirteen grape varieties; the accompanying GrapeSAM pipeline first locates berries from point prompts, then applies Segment Anything for mask generation and a transformer for cluster-level segmentation, thereby producing automated, in-field cluster-closure estimates that match or exceed manual methods across varied conditions.

Load-bearing premise

The dense manual berry-centroid and cluster-mask annotations are accurate enough and representative enough to train models that generalize to real vineyard conditions and unseen varieties.

Editorial extensions

If this is right

  • Automated closure estimates can replace labor-intensive visual scoring while increasing temporal resolution.
  • The same pipeline supports high-throughput phenotyping across multiple grape varieties with minimal new labeling.
  • Objective closure data can be generated at field scale to inform disease-risk decisions.
  • Robust performance on both in-domain and out-of-domain samples indicates the dataset covers sufficient visual diversity for initial deployment.

Reading between the lines

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

  • Integration with time-series imagery could turn closure estimates into growth-rate measurements rather than single-time snapshots.
  • The minimal-supervision design may allow rapid adaptation to new regions by adding only a few point labels per new cluster.
  • Closure outputs could be fused with weather or disease models to test whether automated scores improve spray-timing predictions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The paper presents ViViD-5K, a dataset of 5,000 in-field vineyard images with dense annotations comprising over 648,000 berry centroids and cluster segmentation masks across 13 grape varieties. It introduces the GrapeSAM two-stage pipeline combining point-based berry localization, prompt-based segmentation via Segment Anything, and transformer-based cluster segmentation to enable automated in-field cluster closure estimation with minimal supervision. The authors assert that quantitative results show strong segmentation and counting accuracy across diverse conditions and that visualizations confirm robustness on in-domain and out-of-domain samples.

Significance. A large-scale, densely annotated vineyard dataset focused on berry-level detail would address a clear gap in computer vision resources for agriculture, potentially supporting improved models for phenotyping and disease-risk assessment via cluster closure. The pipeline's use of existing foundation models like SAM with minimal supervision is a practical strength if the empirical claims are substantiated.

major comments (2)
  1. [Dataset construction / annotation protocol] The central empirical claims rest on the accuracy of the 648k manual berry centroid annotations, yet no validation of annotation quality (e.g., inter-annotator agreement, repeated labeling, or expert review) is described. Without this, reported segmentation and counting metrics may reflect label noise rather than model performance, directly undermining generalization claims to new varieties and real vineyard conditions.
  2. [Abstract / Results] The abstract states that 'quantitative results demonstrate strong segmentation and counting accuracy' but supplies no numerical values, error bars, validation splits, ablation studies, or baseline comparisons. This prevents assessment of whether the GrapeSAM pipeline's performance is load-bearing or merely consistent with the provided annotations.
minor comments (2)
  1. [Methods] Clarify the exact definition and computation of 'cluster closure' (e.g., as a percentage or ratio) and how it is derived from the cluster masks in the pipeline.
  2. [Experiments] The out-of-domain robustness claim would benefit from explicit description of the held-out varieties or imaging conditions used for testing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. Below we provide point-by-point responses to the major comments, indicating planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Dataset construction / annotation protocol] The central empirical claims rest on the accuracy of the 648k manual berry centroid annotations, yet no validation of annotation quality (e.g., inter-annotator agreement, repeated labeling, or expert review) is described. Without this, reported segmentation and counting metrics may reflect label noise rather than model performance, directly undermining generalization claims to new varieties and real vineyard conditions.

    Authors: We agree that explicit validation of annotation quality strengthens the empirical claims. The annotations were produced by trained annotators following a standardized protocol developed in consultation with viticulture experts; however, the current manuscript does not report quantitative measures such as inter-annotator agreement. We will expand the dataset section with a detailed description of the annotation protocol and quality-control steps. We will also report inter-annotator agreement computed on a held-out subset if the additional labeling effort can be completed within the revision timeline. revision: partial

  2. Referee: [Abstract / Results] The abstract states that 'quantitative results demonstrate strong segmentation and counting accuracy' but supplies no numerical values, error bars, validation splits, ablation studies, or baseline comparisons. This prevents assessment of whether the GrapeSAM pipeline's performance is load-bearing or merely consistent with the provided annotations.

    Authors: The abstract is intentionally concise and defers specific metrics to the results section, which contains the requested quantitative values, validation splits, ablations, and baseline comparisons. To improve standalone readability we will revise the abstract to include the principal numerical results (e.g., berry-counting MAE and cluster-segmentation IoU) while remaining within length limits. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical dataset release and applied pipeline

full rationale

The paper releases the ViViD-5K dataset with 5,000 images, 648k berry centroid annotations, and cluster masks across 13 varieties, then describes an applied two-stage pipeline (GrapeSAM) combining point-based localization, Segment Anything prompting, and transformer segmentation for cluster closure estimation. No equations, parameter fits, predictions derived from fitted inputs, or self-citation chains appear in the abstract or described structure. All quantitative claims rest on direct empirical evaluation against the released annotations rather than any reduction to prior self-referential results. The work is therefore self-contained as a dataset-plus-pipeline contribution with no load-bearing steps that collapse by construction.

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

Empirical computer-vision dataset and application paper; contains no mathematical derivations, fitted constants, or postulated entities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ViViD-5K: Vineyard vision dataset for field-based berry detection and segmentation and grape cluster closure estimation." pith.science (2026). https://pith.science/paper/J6VCDZZA

@misc{pith2026260524353,
  author       = {Pith},
  title        = {Pith review of: ViViD-5K: Vineyard vision dataset for field-based berry detection and segmentation and grape cluster closure estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J6VCDZZA}},
  note         = {Machine review of arXiv:2605.24353}
}
read the original abstract

Cluster closure, defined as the progressive filling of gaps between the berries in a grape bunch, is a key trait in vineyard management, impacting disease risk. However, traditional visual scoring methods are labor-intensive, subjective, and lack temporal resolution. Existing datasets rarely support fine-grained berry-level analysis, limiting the development of robust deep learning models. In this work, we present ViViD-5k, a large-scale in-field Vineyard Vision Dataset containing 5,000 images with dense annotations, including over 648,000 berry centroids and cluster segmentation masks spanning 13 grape varieties. Building on this dataset, we introduce GrapeSAM, a two-stage visual pipeline that combines point-based berry localization with prompt-based segmentation using Segment Anything, followed by transformer-based cluster segmentation. The pipeline enables automated, in-field estimation of cluster closure with minimal supervision. Quantitative results demonstrate strong segmentation and counting accuracy across diverse conditions, while visualizations confirm robustness on both in-domain and out-of-domain samples. This work provides a scalable and objective alternative to manual compactness scoring and supports high-throughput grape phenotyping with enhanced spatial detail.

Figures

Figures reproduced from arXiv: 2605.24353 by the authors.

Figure 1
Figure 1. Image source distribution of ViViD-5K. It is noteworthy that although majority of raw [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Visualization 1of example images containing cluster bounding box and mask and berry point labels in ViViD-5K. The distributions of grape cluster instances and berry points per image exhibit long tail patterns ( [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Grape and Berry Instance Frequency Count per File. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Diagram of GrapeSAM, the baseline approach for instance segmentation of grape clusters [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Representative inference results of in-domain images. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Representative inference results of out-of-domain images. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results of the effectiveness of IQR-filtering. Each row corresponds to a specific [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Boxplot of log-transformed instance area from individual berry masks segmented by the [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Basal Cluster Growth (??) Cluster Closure Temporal Variation. (??) Periodical Segmen￾tation Samples [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    A new methodology for estimating the grapevine-berry number per cluster using image analysis

    Arturo Aquino et al. “A new methodology for estimating the grapevine-berry number per cluster using image analysis”. In:Biosystems engineering156 (2017), pp. 80–95

    work page 2017

  2. [2]

    GrapesNet: Indian RGB & RGB-D vineyard image datasets for deep learning applications

    Dhanashree K. Barbole and Parul M. Jadhav. “GrapesNet: Indian RGB & RGB-D vineyard image datasets for deep learning applications”. In:Data in Brief48 (2023), p. 109100.issn: 2352-3409.doi:https : / / doi . org / 10 . 1016 / j . dib . 2023 . 109100.url:https : / / www . sciencedirect.com/science/article/pii/S2352340923002196

    work page 2023

  3. [3]

    A Grape Dataset for Instance Segmentation and Maturity Estimation

    Achilleas Blekos et al. “A Grape Dataset for Instance Segmentation and Maturity Estimation”. In:Agronomy13.8 (2023).issn: 2073-4395.doi:10.3390/agronomy13081995.url:https: //www.mdpi.com/2073-4395/13/8/1995

    work page doi:10.3390/agronomy13081995.url:https: 2023

  4. [4]

    Instance segmentation and number counting of grape berry images based on deep learning

    Yanmin Chen et al. “Instance segmentation and number counting of grape berry images based on deep learning”. In:Applied Sciences13.11 (2023), p. 6751

    work page 2023

  5. [5]

    Masked-attention mask transformer for universal image segmentation

    Bowen Cheng et al. “Masked-attention mask transformer for universal image segmentation”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 1290–1299

    work page 2022

  6. [6]

    Response of Thompson Seedless grapes to prebloom thinning

    ARIS Christodoulou, RJ Weaver, and RM Pool. “Response of Thompson Seedless grapes to prebloom thinning”. In:Vitis6.3 (1967), pp. 303–308

    work page 1967

  7. [7]

    Aust J Grape Wine R

    BG Coombe.Adoption of a system for identifying grapevine growth stages. Aust J Grape Wine R. 1995

    work page 1995

  8. [8]

    GBCNet: In-field grape berries counting for yield estimation by dilated CNNs

    Luca Coviello et al. “GBCNet: In-field grape berries counting for yield estimation by dilated CNNs”. In:Applied Sciences10.14 (2020), p. 4870

    work page 2020

Show all 34 references
  1. [9]

    Instance segmentation and berry counting of table grape before thinning based on AS-SwinT

    Wensheng Du and Ping Liu. “Instance segmentation and berry counting of table grape before thinning based on AS-SwinT”. In:Plant Phenomics5 (2023), p. 0085

    2023

  2. [10]

    Comparison of CNN architectures for single grape detection

    MR Gonz´ alez, ME Mart´ ınez-Rosas, and CA Brizuela. “Comparison of CNN architectures for single grape detection”. In:Computers and Electronics in Agriculture231 (2025), p. 109930

    2025

  3. [11]

    Relationship between cluster compactness and bunch rot in Vignoles grapes

    Bryan Hed, Henry K Ngugi, and James W Travis. “Relationship between cluster compactness and bunch rot in Vignoles grapes”. In:Plant disease93.11 (2009), pp. 1195–1201

    2009

  4. [12]

    Procedural grape bunch modeling

    Chun-Yen Huang et al. “Procedural grape bunch modeling”. In:Computers & graphics37.4 (2013), pp. 225–237

    2013

  5. [13]

    A Shine Muscat Grape Berry Detection and Grape Cluster Compactness Estimation for Assessment of Grape Quality Based on Instance Segmentation Methods

    EungChan Kim et al. “A Shine Muscat Grape Berry Detection and Grape Cluster Compactness Estimation for Assessment of Grape Quality Based on Instance Segmentation Methods”. In: Journal of the ASABE66.5 (2023), pp. 1175–1185

    2023

  6. [14]

    Segment anything

    Alexander Kirillov et al. “Segment anything”. In:Proceedings of the IEEE/CVF international conference on computer vision. 2023, pp. 4015–4026

    2023

  7. [15]

    Data set

    Josep Ramon Morros et al.AI4Agriculture Grape Dataset (1.0.0). Data set. Zenodo, 2021. doi:10.5281/zenodo.5660081.url:https://doi.org/10.5281/zenodo.5660081. 19

    2021 doi

  8. [16]

    Yield estimation in vineyards by visual grape detection

    Stephen Nuske et al. “Yield estimation in vineyards by visual grape detection”. In:2011 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE. 2011, pp. 2352– 2358

    2011

  9. [17]

    A non-invasive method based on computer vision for grapevine cluster compactness assessment using a mobile sensing platform under field conditions

    Fernando Palacios, Maria P Diago, and Javier Tardaguila. “A non-invasive method based on computer vision for grapevine cluster compactness assessment using a mobile sensing platform under field conditions”. In:Sensors19.17 (2019), p. 3799

    2019

  10. [18]

    Early leaf removal to improve vineyard efficiency: gas exchange, source-to-sink balance, and reserve storage responses

    Alberto Palliotti, Matteo Gatti, and Stefano Poni. “Early leaf removal to improve vineyard efficiency: gas exchange, source-to-sink balance, and reserve storage responses”. In:American Journal of Enology and Viticulture62.2 (2011), pp. 219–228

    2011

  11. [19]

    Zenodo, Feb

    Isabel Pinheiro.Grapevine Bunch Detection Dataset. Zenodo, Feb. 2023.doi:10 . 5281 / zenodo.7717055.url:https://doi.org/10.5281/zenodo.7717055

    2023 doi

  12. [20]

    Rubi Qui˜ nones, Syeda Mariah Banu, and Eren Gultepe. “GCNet: A Deep Learning Frame- work for Enhanced Grape Cluster Segmentation and Yield Estimation Incorporating Occluded Grape Detection with a Correction Factor for Indoor Experimentation”. In:Journal of Imaging 11.2 (2025)...

    2025 doi

  13. [21]

    Grape detection, segmentation, and tracking using deep neural net- works and three-dimensional association

    Thiago T Santos et al. “Grape detection, segmentation, and tracking using deep neural net- works and three-dimensional association”. In:Computers and Electronics in Agriculture170 (2020), p. 105247

    2020

  14. [22]

    Computer vision and machine learning for viticulture technology

    Kah Phooi Seng et al. “Computer vision and machine learning for viticulture technology”. In: IEEE Access6 (2018), pp. 67494–67510

    2018

  15. [23]

    Estimating duster compactness in Yaghouti grapes

    A Sepahi. “Estimating duster compactness in Yaghouti grapes”. In:VITIS-Journal of Grapevine Research19.2 (2016), p. 81

    2016

  16. [24]

    wGrapeUNIPD-DL: An open dataset for white grape bunch detection

    Marco Sozzi et al. “wGrapeUNIPD-DL: An open dataset for white grape bunch detection”. In:Data in Brief43 (2022), p. 108466.issn: 2352-3409.doi:https://doi.org/10.1016/ j . dib . 2022 . 108466.url:https : / / www . sciencedirect . com / science / article / pii / S2352340922006606

    2022

  17. [25]

    Mechanical yield regulation in winegrapes: comparison of early defo- liation and crop thinning

    Javier Tardaguila et al. “Mechanical yield regulation in winegrapes: comparison of early defo- liation and crop thinning”. In:Australian Journal of Grape and Wine Research18.3 (2012), pp. 344–352

    2012

  18. [26]

    Evaluation of indexes for the quantitative and objective estimation of grapevine bunch compactness

    Javier Tello and Javier Ib´ a˜ nez Marcos. “Evaluation of indexes for the quantitative and objective estimation of grapevine bunch compactness”. In: (2014)

    2014

  19. [27]

    A Preliminary Method for Tracking In-Season Grapevine Cluster Closure Using Image Segmentation and Image Thresholding

    Manushi Trivedi et al. “A Preliminary Method for Tracking In-Season Grapevine Cluster Closure Using Image Segmentation and Image Thresholding”. In:Australian Journal of Grape and Wine Research2023.1 (2023), p. 3923839

    2023

  20. [28]

    Evaluating and Mapping Grape Color Using Image-Based Phenotyping

    A. N. Underhill, C. D. Hirsch, and M. D. Clark. “Evaluating and Mapping Grape Color Using Image-Based Phenotyping”. In:Plant Phenomics2020 (2020).doi:10.34133/2020/8086309. url:https://spj.science.org/doi/abs/10.34133/2020/8086309. 20

    2020 doi

  21. [29]

    Image-based phenotyping identifies quan- titative trait loci for cluster compactness in grape

    Anna Underhill, Cory Hirsch, and Matthew Clark. “Image-based phenotyping identifies quan- titative trait loci for cluster compactness in grape”. In:Journal of the American Society for Horticultural Science145.6 (2020), pp. 363–373

    2020

  22. [30]

    Grape cluster architecture and the susceptibility of berries to Botrytis cinerea

    ME Vail and JJ Marois. “Grape cluster architecture and the susceptibility of berries to Botrytis cinerea.” In:Phytopathology81.2 (1991), pp. 188–191

    1991

  23. [31]

    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”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021, pp. 1974–1983

    2021

  24. [32]

    A 3D grape bunch reconstruction pipeline based on constraint- based optimisation and restricted reconstruction grammar

    Bolai Xin and Mark Whitty. “A 3D grape bunch reconstruction pipeline based on constraint- based optimisation and restricted reconstruction grammar”. In:Computers and Electronics in Agriculture196 (2022), p. 106840

    2022

  25. [33]

    Zabawa et al.Segmentation of wine berries

    L. Zabawa et al.Segmentation of wine berries. Data set. OpenAgrar Repository, 2021.doi: 10.5073/20210308-154150.url:https://doi.org/10.5073/20210308-154150

    2021 doi

  26. [34]

    Counting of grapevine berries in images via semantic segmentation using convolutional neural networks

    Laura Zabawa et al. “Counting of grapevine berries in images via semantic segmentation using convolutional neural networks”. In:ISPRS Journal of Photogrammetry and Remote Sensing 164 (2020), pp. 73–83. 21

    2020

Pith tools

Reviewed June 30, 2026 · model on record in the stance chip above.