Pith. sign in

REVIEW 4 major objections 5 minor 51 references

Multimodal Plant Root Phenotyping with Integration of 3D Skeleton Extraction and Language Analysis

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

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

Pith's one-line read One pipeline turns 3D root scans into explainable phenotype answers.

desk verdict Solid skeleton work undermined by circular LLM evaluation; deserves peer review but not acceptance as is. read the letter →

arxiv 2608.03109 v1 pith:NLEOOQJ5 submitted 2026-08-04 cs.CV cs.RO

Jiakai Lin , Zijun Li , Guoyu Lu This is my paper · ORCID
classification cs.CVcs.RO
keywords plantrootphenotyping3DskeletonextractionWeightedLaplacianContractionpointcloudsvisualquestionansweringGPTfine-tuninginterpretableAIrobotic
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes a robotic pipeline that takes dense 3D point clouds of plant roots, extracts a skeleton without any ground-truth supervision, computes quantitative morphological descriptors from that skeleton, and then fine-tunes GPT-4o to answer questions about the roots in natural language that cites those numbers as evidence. The authors claim the framework achieves accurate skeleton extraction across 12 plant species and outperforms existing skeletonization and VQA baselines, and that the Evidence-First fine-tuning makes the language outputs biologically coherent rather than merely fluent. A sympathetic reader would take the contribution to be a demonstration that unsupervised geometric perception plus structure-grounded language supervision is enough to give a robot an interpretable read on root architecture.

What carries the argument

The load-bearing mechanism is Weighted Laplacian-Based Contraction (W-LBC), an unsupervised skeletonization scheme in which a direction vector and a local density index predicted for each point by a PointTransformer v3 head re-weight the attraction and contraction terms of a Laplacian matrix, so the point cloud shrinks onto an intrinsically one-dimensional skeleton while preserving fine topology. The second mechanism is the Evidence-First template, which forces a fine-tuned GPT-4o to ground every answer in the quantitative descriptors computed from that skeleton graph.

What would settle it

Give a set of root point clouds with independent expert-annotated morphological labels and expert-written answers to the VQA questions, and check whether the automatically generated GPT-4o labels used in training match those expert labels for biologically meaningful traits such as lateral versus taproot classification or branching angle ranges; if expert and GPT-4o labels disagree substantially, the fine-tuned model's 'biologically consistent' answers are merely self-consistent.

Watch

Extended reading notes

Core claim

The paper's central claim is that root phenotyping can be fully automated by chaining an unsupervised 3D skeleton extractor to a language model fine-tuned on automatically generated evidence-first question–answer pairs. The skeleton extractor, Weighted Laplacian-Based Contraction (W-LBC), uses a PointTransformer v3 backbone to predict per-point direction and density cues, which modulate a Laplacian contraction so that fine root branches and dense crossings are preserved; from the resulting graph, descriptors such as root count, total length, branch angle and density are computed. These descriptors are placed into a four-field template (Evidence, Observation, Question, Answer), thousands of which are generated semi-automatically, and GPT-4o is fine-tuned on them so that its answers explicitly reference the quantitative evidence. The paper reports that the full system improves root count accuracy from 0.63 to 0.79 (for root systems with fewer than 100 roots) over Pc-Skeletor and lifts VQA direct-answer accuracy from roughly 45 percent for the general GPT-4o to 59.7 percent on the held-out test set, and that these gains persist in few-shot settings.

Load-bearing premise

The whole language reasoning module is trained and evaluated on answers that were generated and refined by GPT-4o itself, so the paper assumes those automatic answers are biologically correct and that agreeing with them measures understanding.

Editorial extensions

If this is right

  • With the same unsupervised skeletonization and fine-tuning, the pipeline works across species with very different root architectures, such as Sweet Potato and Grass, without per-species manual engineering.
  • The Evidence-First prompts cause the model to answer in terms of measured numbers (count, length, angle), making the reasoning auditable by a human.
  • Removing the skeleton extraction or the fine-tuning degrades accuracy, showing that both components carry the result.
  • Few-shot samples (one to five per species) improve generalization to unseen species and question types, suggesting the fine-tuned model transfers to scant data.

Reading between the lines

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

  • The fine-tuning uses labels generated by GPT-4o itself; if those labels contain biological errors, the evaluation rewards consistency rather than truth, and the paper does not validate labels against a human expert or an independent phenotyping tool.
  • The same 'measure first, explain later' template could generalize to other branching structures such as tree canopies or vasculature, provided a skeleton can be extracted.
  • A testable next step is to compare the fine-tuned GPT-4o answers against manual expert annotations on a held-out set to check whether 'biologically coherent' equals 'biologically correct'.
  • The current RCA/LRA metrics only check root count and longest root; a fuller phenotyping evaluation covering branching angle and density on annotated data would be needed to fully support the claim of robust quantitative characterization.
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 a multimodal framework for plant root phenotyping that combines an unsupervised 3D skeleton extraction module (W-LBC) with a fine-tuned GPT-4o language module. The geometric part uses PointTransformer v3 features and direction/density-aware weighted Laplacian contraction to extract skeletons, followed by topological reconstruction and computation of quantitative descriptors such as root count, length, branching angle, and density. The language part builds automatically generated 'Evidence-First' instruction-response pairs, fine-tunes GPT-4o on them, and evaluates on held-out prompts from the same pipeline. Experiments compare skeleton extraction against several baselines using root count accuracy and longest root accuracy, and compare VQA accuracy against general VQA models, with additional ablations and few-shot tests.

Significance. The W-LBC skeletonization contribution is potentially useful: it is unsupervised, the composite loss is plausible, and the qualitative examples suggest better preservation of fine roots compared with tested baselines. The paper also creates a new dataset with manual annotations for two traits, which is a useful resource for the community. However, the language reasoning claim is not empirically supported: the VQA training and evaluation labels are both generated and refined by GPT-4o, so the reported accuracy measures consistency with GPT-4o's own output rather than biological correctness. The skeleton evaluation also lacks error bars, per-species analysis, and independent ground truth for full skeleton topology. If the authors add independent human evaluation, statistical rigor, and parameter sensitivity analysis, the framework could become a valuable system paper; as it stands, the central 'interpretable reasoning' claim is unsubstantiated.

major comments (4)
  1. [Section IV.A, IV.B, V.C] The VQA supervision and evaluation are both derived from GPT-4o. Section IV.A states that GPT-4o processes images to generate the Observation field and refines terminology, and that the Answer field is produced by the same automatic pipeline. Section IV.B then fine-tunes GPT-4o on these pairs via supervised fine-tuning, and Section V.C evaluates on 'held-out structured prompts' drawn from the same automatically generated dataset. Consequently, the MC/DA scores in Table II measure the model's ability to reproduce GPT-4o's labeling style and evidence-citation format, not whether the explanations are biologically valid. No human expert labels or independent biological criteria are used for the VQA task. This directly undermines the abstract's claim of 'biologically consistent explanations' and 'interpretable reasoning.'
  2. [Section V.C, Table II] The baseline comparison in Table II is confounded by the statement in Section V.C that 'For models without built-in reasoning, we used GPT-4o to generate contextual inference, ensuring fair comparison.' This means the baseline models also receive GPT-4o-generated context that may contain the answer or reveal the expected reasoning. The reported gains of the fine-tuned model over these baselines could therefore stem from the fine-tuned model being adapted to the exact template rather than from superior reasoning ability. A fair comparison would require either identical inputs without GPT-generated context or a controlled human-annotated test set.
  3. [Section V.A, Table I] Table I reports only two metrics (RCA and LRA) aggregated over the entire dataset, with no error bars, significance tests, or per-species breakdown. Section V.A states that ten of the twelve species have only 1-5 samples each, while Sweet Potato and Apple Tree dominate (400 and 800 models, respectively). The abstract's claim of 'robust ... across 12 plant species' is therefore not substantiated: the aggregate numbers could be driven almost entirely by the two abundant species. In addition, the manual annotations cover only visible root count and longest root length, so branching angle, density, and topology accuracy of the extracted skeletons are never validated against ground truth.
  4. [Section III.A-B] The W-LBC formulation in Eqs. (1)-(8) introduces at least fourteen free parameters (sigma_a, sigma_c, kappa, mu, eta, nu, alpha1, alpha2, beta1, beta2, epsilon, tau_d, tau_c, tau_t) that control the contraction and reconstruction. The paper does not report any sensitivity analysis or ablation for these parameters, even though the quality of the skeleton is directly determined by them. The claim that the method is 'unsupervised' and generalizes across species requires evidence that performance is not the result of careful dataset-specific tuning. A parameter sensitivity study or a robustness table is needed.
minor comments (5)
  1. [References] Reference [29] is cited in the text for Pc-Skeletor, but the bibliography entry [29] is 'Communication between plant roots and the soil microbiome' by Molefe et al., which is unrelated to skeleton extraction; the correct Pc-Skeletor reference appears to be missing.
  2. [Figure 1 caption] The last sentence of the caption contains a typo: 'The framework is showe in Fig. 1' should be 'The framework is shown in Fig. 1.'
  3. [Eq. (6)] The text after Eq. (6) says L_w is constructed using 'the direction-density-aware weights defined in Eq. (7)-(8)', but Eqs. (7) and (8) define the smoothness and connectivity losses, not the weights; the weights are defined in Eqs. (3)-(4).
  4. [Section V.A] The sentence 'The same split was used for both the 3D skeleton extraction model and the fine-tuned GPT-4o model' is confusing because the skeleton extraction is unsupervised and does not use labels; please clarify why the split matters for that module.
  5. [Figure 3] The qualitative skeleton comparison in Figure 3 has no scale bar or quantitative caption, making it hard for the reader to assess fidelity differences across methods.

Circularity Check

2 steps flagged · score 7.0 of 10

The VQA half of the paper is circular: GPT-4o generates both the training labels and the held-out evaluation prompts, so the reported 'interpretable reasoning' largely measures reproduction of GPT-4o's own outputs; only the skeleton-extraction component has independent manual evaluation.

  1. fitted input called prediction [Section IV.A (Data Construction and Template Design) and Section V.C (Comparison of VQA Module Accuracy)]
    "During dataset construction, GPT-4o processes these images to generate the [Observation] field, summarizing traits such as branching density and curvature. ... Through this pipeline, we achieve large-scale generation of high-quality instruction–response pairs without manual annotation. ... Reasoning evaluation is performed on held-out structured prompts, measuring evidence consistency, reasoning quality, and linguistic coherence."

    The instruction–response pairs used for supervised fine-tuning are generated automatically by GPT-4o, and the held-out test prompts come from the same automatically generated dataset. Therefore high MC/DA/FS/ZS accuracy shows only that the fine-tuned model matches GPT-4o's label distribution and answer format; it does not establish biological correctness. No human expert labels or external biological ground truth are used for the VQA evaluation; the manual annotations in Section V.A cover only visible root count and longest root length for skeleton evaluation. The 'biologically consistent explanations' claim is thus untested and reduces in part to consistency with GPT-4o's own outputs.

  2. self definitional [Section IV.B (Fine-Tuning Procedure)]
    "We adopt standard Supervised Fine-Tuning (SFT) to align GPT's reasoning behavior with the proposed structure-aware dataset. ... The model is trained to maximize the likelihood of generating the target sequence conditioned on the prompt."

    Under SFT, the target completions are the automatically generated GPT-4o answers. The cross-entropy objective maximizes the likelihood of those exact answers, so the fine-tuned model is explicitly trained to reproduce GPT-4o's phrasing and evidence-citation conventions. Reporting 'biological consistency' from outputs generated by this fitting loop is self-definitional: GPT-4o-generated labels define the notion of correctness, and the model is scored against them. No independent biological validation intervenes, so the reasoning-evaluation scores are by construction measures of agreement with the label generator.

full rationale

The skeleton-extraction component is not circular: W-LBC is unsupervised, and Table I's RCA/LRA metrics are compared against methods on point clouds with manually annotated root count and longest root length. The circularity is confined to the language-reasoning component, which is load-bearing for the abstract's central claim of 'robust, interpretable reasoning across 12 plant species.' Both the supervision (Evidence-First instruction–response pairs) and the evaluation (held-out structured prompts) are generated by GPT-4o without manual annotation, so the VQA results demonstrate reproduction of GPT-4o's own labeling style rather than biological validity. Self-citations such as [25] and [43] are present but not the primary load-bearing mechanism. The resulting score reflects partial circularity: the geometric pipeline is independently grounded, while the interpretability claims reduce by construction to agreement with GPT-4o-generated outputs.

Assumptions & free parameters 14 free parameters · 6 assumptions · 0 invented entities

The central empirical claims rest on a private dataset, hand-set contraction hyperparameters, and a VQA training loop in which GPT-4o writes and refines the labels used to fine-tune itself. The morphological descriptors are only as good as the unsupervised skeleton and the graph reconstruction, neither of which is validated against independent measurements. No new physical or mathematical entities are introduced, so the invented-entity list is empty.

free parameters (14)
  • sigma_a
    Attraction kernel spatial radius in Eq. 3; hand-set, no value or sensitivity analysis reported.
  • sigma_c
    Contraction kernel spatial radius in Eq. 4; hand-set, no value reported.
  • kappa
    Directional alignment sensitivity in Eq. 3; no value reported.
  • mu
    Contraction weighting factor inside the Laplacian matrix in Eq. 1; no value reported.
  • eta
    Update learning rate for the spatial contraction step in Eq. 5; no value reported.
  • nu
    Contraction coefficient in the spatial update in Eq. 5; no value reported.
  • alpha1
    Loss weight for the Laplacian regularizer in Eq. 6; no value reported.
  • alpha2
    Loss weight for the squared Laplacian regularizer in Eq. 6; no value reported.
  • beta1
    Weight for the directional smoothness loss in the total objective; no value reported.
  • beta2
    Weight for the connectivity preservation loss in the total objective; no value reported.
  • epsilon
    Convergence threshold for the contraction iteration; no value reported.
  • tau_d
    Edge-length pruning threshold in the kNN graph construction in Section III.C; no value reported.
  • tau_c
    Spatial threshold for restoring refined edges in Section III.C; no value reported.
  • tau_t
    Directional threshold for restoring refined edges in Section III.C; no value reported.
assumptions (6)
  • domain assumption PointTransformer v3 features can predict directional vectors t_i and density indices rho_i that are accurate enough to guide skeleton contraction
    The W-LBC scheme in Section III.A depends on these predicted quantities as geometry-aware indicators, but no ground-truth evaluation of t_i or rho_i is provided.
  • domain assumption The unsupervised composite loss L_total drives contraction to the true root topology without supervision
    Section III.B defines the total objective but gives no proof or external validation that W-LBC converges to biologically correct skeletons; evaluation is only on the private dataset.
  • domain assumption kNN plus MST plus direction-aware edge refinement reconstructs a graph whose leaf count and longest path match true root morphology
    Section III.C defines root count, length, and depth through this reconstructed graph; if fine roots are missed or spurious loops are added, all morphological descriptors are wrong.
  • domain assumption Manual annotations of visible root count and longest root length are reliable ground truth for point clouds with overlapping and occluded roots
    Section V.A states that the dataset was manually annotated, but no inter-annotator agreement or error analysis is reported.
  • ad hoc to paper Answers generated and refined by GPT-4o are valid ground truth for supervised fine-tuning and evaluation
    Section IV.A uses GPT-4o to generate the Observation field and refines the Answer fields; correctness is assumed rather than verified, making the VQA evaluation self-referential.
  • ad hoc to paper Fine-tuning GPT-4o on 5,000 automatically generated pairs teaches causal biological reasoning rather than recitation of templates
    No human biological evaluation or independent experiment supports this; accuracy is measured only against GPT-4o-derived answers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multimodal Plant Root Phenotyping with Integration of 3D Skeleton Extraction and Language Analysis." pith.science (2026). https://pith.science/paper/NLEOOQJ5

@misc{pith2026260803109,
  author       = {Pith},
  title        = {Pith review of: Multimodal Plant Root Phenotyping with Integration of 3D Skeleton Extraction and Language Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NLEOOQJ5}},
  note         = {Machine review of arXiv:2608.03109}
}
read the original abstract

Plant root phenotyping is fundamental to understanding below-ground structures, optimizing crop management, and improving agricultural sustainability. This paper presents a multimodal robotic AI framework that integrates 3D skeleton extraction with language-guided reasoning for interpretable and data-efficient root analysis. We develop an unsupervised skeleton extraction network based on Weighted Laplacian Contraction (W-LBC) to generate high-fidelity structural representations from dense point clouds captured by robotic 3D sensing platforms. Quantitative morphological descriptors, including root count, length, branching angle, and density, are computed from the reconstructed skeleton graph to capture geometric and topological characteristics. Building on these features, we introduce an Evidence-First language modeling framework that fine-tunes GPT as an interactive analytical chatbot using automatically generated instruction--response pairs. Each training sample provides measurable evidence before natural-language reasoning, enabling the model to ground interpretation in quantitative morphology. Through supervised fine-tuning, GPT associates numerical structure with semantic meaning, producing biologically consistent explanations of growth patterns and adaptive traits. Experiments show that the structure-guided framework achieves robust, interpretable reasoning across 12 plant species with diverse root architectures. By integrating unsupervised 3D geometric perception with large-scale language understanding, our approach bridges quantitative analysis and semantic interpretation, establishing a unified paradigm for explainable robotic plant root phenotyping.

Figures

Figures reproduced from arXiv: 2608.03109 by the authors.

Figure 1
Figure 1. Overview of the proposed multimodal framework for plant root phenotyping. Point clouds are processed by Point Transformer [41] with direction [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Example of the Evidence-First template for fine-tuning, where [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Comparison of root skeletons generated from input point cloud using various methods. The point cloud input is shown on the left, followed by [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Our adaptive model can handle various types of test questions, with the image illustrating three example types and the corresponding responses [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 43 canonical work pages

  1. [1]

    Toporoot+: computing whorl and soil line traits of field-excavated maize roots from ct imaging.Plant Methods, 20(1):132, 2024

    work page 2024

  2. [2]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Al- tenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    arXiv 2023

  3. [3]

    Unlocking plant secrets: A systematic review of 3d imaging in plant phenotyping techniques.Computers and Electronics in Agriculture, 222:109033, 2024

    Muhammad Salman Akhtar, Zuhair Zafar, Raheel Nawaz, and Muham- mad Moazam Fraz. Unlocking plant secrets: A systematic review of 3d imaging in plant phenotyping techniques.Computers and Electronics in Agriculture, 222:109033, 2024

    work page 2024

  4. [4]

    Bottom-up and top-down attention for image captioning and visual question answering

    Peter Anderson, Xiaodong He, Chris Buehler, Damien Teney, Mark Johnson, Stephen Gould, and Lei Zhang. Bottom-up and top-down attention for image captioning and visual question answering. In CVPR, pages 6077–6086, 2018

    work page 2018

  5. [5]

    Hyperspectral imaging: a novel approach for plant root phenotyping.Plant methods, 14:1–17, 2018

    Gernot Bodner, Alireza Nakhforoosh, Thomas Arnold, and Daniel Leitner. Hyperspectral imaging: a novel approach for plant root phenotyping.Plant methods, 14:1–17, 2018

    work page 2018

  6. [6]

    Springer Science & Business Media, 2012

    Wolfgang B ¨ohm.Methods of studying root systems, volume 33. Springer Science & Business Media, 2012

    work page 2012

  7. [7]

    Language models are few-shot learners.arXiv preprint arXiv:2005.14165, 2020

    Tom B Brown. Language models are few-shot learners.arXiv preprint arXiv:2005.14165, 2020

    arXiv 2005

  8. [8]

    Does tone change the answer? evaluating prompt politeness effects on modern LLMs: GPT, Gemini, and LLaMA.arXiv preprint arXiv:2512.12812, 2025

    Hanyu Cai, Binqi Shen, Lier Jin, Lan Hu, and Xiaojing Fan. Does tone change the answer? evaluating prompt politeness effects on modern LLMs: GPT, Gemini, and LLaMA.arXiv preprint arXiv:2512.12812, 2025

    arXiv 2025

Show all 51 references
  1. [9]

    Hyperpri: a dataset of hyperspectral images for underground plant root study.Computers and Electronics in Agriculture, 225:109307, 2024

    Spencer J Chang, Ritesh Chowdhry, Yangyang Song, Tomas Mejia, Anna Hampton, Shelby Kucharski, TM Sazzad, Yuxuan Zhang, San- jeev J Koppal, Chris H Wilson, et al. Hyperpri: a dataset of hyperspectral images for underground plant root study.Computers and Electronics in Agricultu...

    2024

  2. [10]

    Vqa therapy: Exploring answer differences by visually grounding answers

    Chongyan Chen, Samreen Anjum, and Danna Gurari. Vqa therapy: Exploring answer differences by visually grounding answers. InICCV, pages 15315–15325, 2023

    2023

  3. [11]

    Uniter: Universal image-text representation learning

    Yen-Chun Chen, Linjie Li, Licheng Yu, Ahmed El Kholy, Faisal Ahmed, Zhe Gan, Yu Cheng, and Jingjing Liu. Uniter: Universal image-text representation learning. InECCV, pages 104–120, 2020

    2020

  4. [12]

    Smart-tree: neural medial axis approximation of point clouds for 3d tree skeletonization

    Harry Dobbs, Oliver Batchelor, Richard Green, and James Atlas. Smart-tree: neural medial axis approximation of point clouds for 3d tree skeletonization. InIberian Conference on Pattern Recognition and Image Analysis, pages 351–362, 2023

    2023

  5. [13]

    Adtree: Accurate, detailed, and automatic modelling of laser-scanned trees.Remote Sensing, 11(18):2074, 2019

    Shenglan Du, Roderik Lindenbergh, Hugo Ledoux, Jantien Stoter, and Liangliang Nan. Adtree: Accurate, detailed, and automatic modelling of laser-scanned trees.Remote Sensing, 11(18):2074, 2019

    2019

  6. [14]

    Boosting transferability in vision-language attacks via diversification along the intersection region of adversarial trajectory

    Sensen Gao, Xiaojun Jia, Xuhong Ren, Ivor Tsang, and Qing Guo. Boosting transferability in vision-language attacks via diversification along the intersection region of adversarial trajectory. InECCV, pages 442–460. Springer, 2025

    2025

  7. [15]

    CoDES: A context-efficient framework for enhancing small language models via domain-specific adaptation and model ensembling.Preprints, Mar

    Lan Hu, Yuting Xin, Binqi Shen, Hanyu Cai, and Lier Jin. CoDES: A context-efficient framework for enhancing small language models via domain-specific adaptation and model ensembling.Preprints, Mar. 2026

    2026

  8. [16]

    Promptcap: Prompt-guided image captioning for vqa with gpt-3

    Yushi Hu, Hang Hua, Zhengyuan Yang, Weijia Shi, Noah A Smith, and Jiebo Luo. Promptcap: Prompt-guided image captioning for vqa with gpt-3. InICCV, pages 2963–2975, 2023

    2023

  9. [17]

    Clip2point: Transfer clip to point cloud classification with image-depth pre- training

    Tianyu Huang, Bowen Dong, Yunhan Yang, Xiaoshui Huang, Ryn- son WH Lau, Wanli Ouyang, and Wangmeng Zuo. Clip2point: Transfer clip to point cloud classification with image-depth pre- training. InCVPR, pages 22157–22167, 2023

    2023

  10. [18]

    Visual prompt tuning

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Visual prompt tuning. InECCV, pages 709–727. Springer, 2022

    2022

  11. [19]

    Skeleton extraction from point clouds of trees with complex branches via graph contraction.The Visual Computer, 37:2235–2251, 2021

    Anling Jiang, Ji Liu, Jianling Zhou, and Min Zhang. Skeleton extraction from point clouds of trees with complex branches via graph contraction.The Visual Computer, 37:2235–2251, 2021

    2021

  12. [20]

    Webly supervised concept expan- sion for general purpose vision models

    Amita Kamath, Christopher Clark, Tanmay Gupta, Eric Kolve, Derek Hoiem, and Aniruddha Kembhavi. Webly supervised concept expan- sion for general purpose vision models. InECCV, 2022

    2022

  13. [21]

    Vilt: Vision-and-language transformer without convolution or region supervision

    Wonjae Kim, Bokyung Son, and Ildoo Kim. Vilt: Vision-and-language transformer without convolution or region supervision. InICML, pages 5583–5594. PMLR, 2021

    2021

  14. [22]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. InICML, pages 12888–12900, 2022

    2022

  15. [23]

    Automatic organ-level point cloud segmentation of maize shoots by integrating high-throughput data acquisition and deep learning.CEA, 2022

    Yinglun Li, Weiliang Wen, Teng Miao, Sheng Wu, Zetao Yu, Xiaodong Wang, Xinyu Guo, and Chunjiang Zhao. Automatic organ-level point cloud segmentation of maize shoots by integrating high-throughput data acquisition and deep learning.CEA, 2022

    2022

  16. [24]

    Chad Bolding, J

    Jiakai Lin, M. Chad Bolding, J. L. Conrad IV , Alicia Peduzzi, and Guoyu Lu. 3d wood reconstruction and volume estimation for determining log truck payloads using consumer-grade cameras.Smart Agricultural Technology, page 102142, 2026

    2026

  17. [25]

    3d plant root skeleton detection and extraction

    Jiakai Lin, Jinchang Zhang, Ge Jin, Wenzhan Song, Tianming Liu, and Guoyu Lu. 3d plant root skeleton detection and extraction. In IROS. IEEE, 2025

    2025

  18. [26]

    Keypoint detection and description for raw bayer images

    Jiakai Lin, Jinchang Zhang, and Guoyu Lu. Keypoint detection and description for raw bayer images. IEEE, 2025

    2025

  19. [27]

    Krisp: Integrating implicit and symbolic knowledge for open-domain knowledge-based vqa

    Kenneth Marino, Xinlei Chen, Devi Parikh, Abhinav Gupta, and Marcus Rohrbach. Krisp: Integrating implicit and symbolic knowledge for open-domain knowledge-based vqa. InCVPR, 2021

    2021

  20. [28]

    Clipcap: Clip prefix for image captioning.arXiv preprint arXiv:2111.09734, 2021

    Ron Mokady, Amir Hertz, and Amit H Bermano. Clipcap: Clip prefix for image captioning.arXiv preprint arXiv:2111.09734, 2021

    2021 arXiv

  21. [29]

    Communication between plant roots and the soil microbiome; involve- ment in plant growth and development.Symbiosis, 90(3), 2023

    Rebaona R Molefe, Adenike E Amoo, and Olubukola O Babalola. Communication between plant roots and the soil microbiome; involve- ment in plant growth and development.Symbiosis, 90(3), 2023

    2023

  22. [30]

    Root traits and phenotyping strategies for plant improvement.Plants, 4(2):334–355, 2015

    Ana Paez-Garcia, Christy M Motes, Wolf-R ¨udiger Scheible, Rujin Chen, Elison B Blancaflor, and Maria J Monteros. Root traits and phenotyping strategies for plant improvement.Plants, 4(2):334–355, 2015

    2015

  23. [31]

    Iterative image segmentation of plant roots for high-throughput phenotyping.Scientific Reports, 12(1):16563, 2022

    Kyle Seidenthal, Karim Panjvani, Rahul Chandnani, Leon Kochian, and Mark Eramian. Iterative image segmentation of plant roots for high-throughput phenotyping.Scientific Reports, 12(1):16563, 2022

    2022

  24. [32]

    Prompting large language models with answer heuristics for knowledge-based visual question answering

    Zhenwei Shao, Zhou Yu, Meng Wang, and Jun Yu. Prompting large language models with answer heuristics for knowledge-based visual question answering. InCVPR, pages 14974–14983, 2023

    2023

  25. [33]

    How much can clip benefit vision-and-language tasks?arXiv preprint arXiv:2107.06383, 2021

    Sheng Shen, Liunian Harold Li, Hao Tan, Mohit Bansal, Anna Rohrbach, Kai-Wei Chang, Zhewei Yao, and Kurt Keutzer. How much can clip benefit vision-and-language tasks?arXiv preprint arXiv:2107.06383, 2021

    2021 arXiv

  26. [34]

    Deepskeleton: Learning multi-task scale-associated deep side outputs for object skeleton extraction in natural images.IEEE Transactions on Image Processing, 26(11):5298–5311, 2017

    Wei Shen, Kai Zhao, Yuan Jiang, Yan Wang, Xiang Bai, and Alan Yuille. Deepskeleton: Learning multi-task scale-associated deep side outputs for object skeleton extraction in natural images.IEEE Transactions on Image Processing, 26(11):5298–5311, 2017

    2017

  27. [35]

    Alpha-clip: A clip model focusing on wherever you want

    Zeyi Sun, Ye Fang, Tong Wu, Pan Zhang, Yuhang Zang, Shu Kong, Yuanjun Xiong, Dahua Lin, and Jiaqi Wang. Alpha-clip: A clip model focusing on wherever you want. InCVPR, pages 13019–13029, 2024

    2024

  28. [36]

    Lxmert: Learning cross-modality encoder representations from transformers.arXiv preprint arXiv:1908.07490, 2019

    Hao Tan and Mohit Bansal. Lxmert: Learning cross-modality encoder representations from transformers.arXiv preprint arXiv:1908.07490, 2019

    1908 arXiv

  29. [37]

    Curvature-adaptive consistency flow matching: Autonomous trajectory optimization via reinforcement learning, 2026

    Songtao Tian, Guhan Chen, Bohan Li, Jingyi Ma, and Zixiong Yu. Curvature-adaptive consistency flow matching: Autonomous trajectory optimization via reinforcement learning, 2026

    2026

  30. [38]

    Multimodal few-shot learning with frozen language models.NeurIPS, 2021

    Maria Tsimpoukelli, Jacob L Menick, Serkan Cabi, SM Eslami, Oriol Vinyals, and Felix Hill. Multimodal few-shot learning with frozen language models.NeurIPS, 2021

    2021

  31. [39]

    Vqa-gnn: Reasoning with multimodal knowledge via graph neural networks for visual question answering

    Yanan Wang, Michihiro Yasunaga, Hongyu Ren, Shinya Wada, and Jure Leskovec. Vqa-gnn: Reasoning with multimodal knowledge via graph neural networks for visual question answering. InICCV, 2023

    2023

  32. [40]

    Learnable skeleton- aware 3d point cloud sampling

    Cheng Wen, Baosheng Yu, and Dacheng Tao. Learnable skeleton- aware 3d point cloud sampling. InCVPR, pages 17671–17681, 2023

    2023

  33. [41]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. InCVPR, 2024

    2024

  34. [42]

    Dettoolchain: A new prompting paradigm to unleash detection ability of mllm.arXiv preprint arXiv:2403.12488, 2024

    Yixuan Wu, Yizhou Wang, Shixiang Tang, Wenhao Wu, Tong He, Wanli Ouyang, Philip Torr, and Jian Wu. Dettoolchain: A new prompting paradigm to unleash detection ability of mllm.arXiv preprint arXiv:2403.12488, 2024

    2024 arXiv

  35. [43]

    Interpretable traffic responsibility from dashcam video via legal multi-agent reasoning.arXiv preprint arXiv:2603.17930, 2026

    Jingchun Yang and Jinchang Zhang. Interpretable traffic responsibility from dashcam video via legal multi-agent reasoning.arXiv preprint arXiv:2603.17930, 2026

    2026

  36. [44]

    Phenotyping root system architecture, anatomy, and physiology to understand soil foraging.High-Throughput Crop Phenotyping, pages 209–221, 2021

    Larry M York. Phenotyping root system architecture, anatomy, and physiology to understand soil foraging.High-Throughput Crop Phenotyping, pages 209–221, 2021

    2021

  37. [45]

    Long-clip: Unlocking the long-text capability of clip

    Beichen Zhang, Pan Zhang, Xiaoyi Dong, Yuhang Zang, and Jiaqi Wang. Long-clip: Unlocking the long-text capability of clip. InECCV, pages 310–325, 2024

    2024

  38. [46]

    Adaptive event stream slicing for open-vocabulary event-based object de- tection via vision-language knowledge distillation.arXiv preprint arXiv:2510.00681, 2025

    Jinchang Zhang, Zijun Li, Jiakai Lin, and Guoyu Lu. Adaptive event stream slicing for open-vocabulary event-based object de- tection via vision-language knowledge distillation.arXiv preprint arXiv:2510.00681, 2025

    2025

  39. [47]

    Underground mapping and localiza- tion based on ground-penetrating radar

    Jinchang Zhang and Guoyu Lu. Underground mapping and localiza- tion based on ground-penetrating radar. InACCV, 2024

    2024

  40. [48]

    Vision-language embodiment for monocular depth estimation

    Jinchang Zhang and Guoyu Lu. Vision-language embodiment for monocular depth estimation. InCVPR, 2025

    2025

  41. [49]

    Pointclip: Point cloud understanding by clip

    Renrui Zhang, Ziyu Guo, Wei Zhang, Kunchang Li, Xupeng Miao, Bin Cui, Yu Qiao, Peng Gao, and Hongsheng Li. Pointclip: Point cloud understanding by clip. InCVPR, pages 8552–8562, 2022

    2022

  42. [50]

    Unified vision-language pre-training for image captioning and vqa

    Luowei Zhou, Hamid Palangi, Lei Zhang, Houdong Hu, Jason Corso, and Jianfeng Gao. Unified vision-language pre-training for image captioning and vqa. InAAAI, volume 34, pages 13041–13049, 2020

    2020

  43. [51]

    Underground plant exploration: Non- destructive 3d root assessment with gpr based on point graph neural network

    Yuwei Zhou and Guoyu Lu. Underground plant exploration: Non- destructive 3d root assessment with gpr based on point graph neural network. InCVPR, 2026

    2026

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.