Pith. sign in

REVIEW 3 major objections 5 minor 148 references

Fine-tuning a general-purpose segmentation model on procedurally generated synthetic tomato greenhouses lifts its real-world fruit segmentation IoU by 12.9 percentage points.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 14:57 UTC pith:3OE7DQN2

load-bearing objection The sim-to-real transfer claim holds up on the evidence: fine-tuning SAM 3 on procedural synthetic greenhouse data improves real tomato fruit segmentation by 12.9 pp macro IoU, but the result is a single-run best-of-eight with no variance estimates. the 3 major comments →

arxiv 2607.18576 v1 pith:3OE7DQN2 submitted 2026-07-20 cs.CV

Text-conditioned Segmentation for Tomato Phenotyping via Procedural Synthetic Data

Samy Mounir , Mikolaj Cieslak , Najmeddine Dhieb , Hakim Ghazzai , Jonathan Klein , Katja Froehlich , Soeren Pirk , Wojciech Palubicki
show 3 more authors
Gianluca Setti Ahmed M. Eltawil Dominik L. Michels
This is my paper
classification cs.CV
keywords sim-to-real transfersynthetic datatomato phenotypingsemantic segmentationSegment Anything ModelL-systemsfine-tuningagricultural computer vision
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to show that you can train a general-purpose image segmentation model to recognize tomato organs in real greenhouses without any real annotated images. It builds a procedurally generated 3D model of a commercial cherry-tomato greenhouse, renders 68,000 labeled synthetic images, and fine-tunes the Segment Anything Model 3 (SAM 3) on them. On three real tomato datasets, the best fine-tuned variant raises fruit segmentation IoU from 52.7% to 65.6% for the zero-shot model, a gain of 12.9 percentage points. The authors also find that a model's score on synthetic validation data does not predict how well it transfers to real data, so real-domain evaluation is needed for model selection. If right, the result means foundation models can be specialized for agriculture using only synthetic data, reducing annotation labor.

Core claim

The central claim is that combining procedurally generated synthetic data with foundation-model fine-tuning substantially improves text-conditioned segmentation on real greenhouse tomato imagery. Using an L-system model of cherry tomato with sympodial growth, tropisms, and the leaning-and-lowering management practice, the authors generate a multimodal synthetic dataset (RGB, depth, instance, semantic) and use it to fine-tune SAM 3 in three ways: LoRA, full fine-tuning, and weight interpolation with the zero-shot checkpoint. Full fine-tuning yields the best aggregate transfer (macro fruit IoU 0.6562 vs 0.5269 zero-shot; image-weighted 0.7372), while WiSE-FT LoRA gives the best individual data

What carries the argument

The load-bearing object is a procedural L-system model of Solanum lycopersicum var. cerasiform that encodes sympodial branching, stochastic morphology, phototropism/gravitropism, and greenhouse management (pruning, leaning-and-lowering) and is rendered in Unreal Engine 5 to produce 68k images with pixel-perfect semantic/instance/depth labels. The second half of the machinery is fine-tuning SAM 3, a promptable segmentation foundation model, via full fine-tuning, LoRA on the last 12 vision-transformer attention qkv layers, and WiSE-FT weight interpolation (alpha=0.75) to preserve the general visual prior while specializing for leaf/stem/flower/fruit text prompts.

Load-bearing premise

The whole sim-to-real transfer rests on the assumption that the L-system plus Unreal Engine 5 greenhouse faithfully represents real commercial tomato greenhouses in morphology, lighting, occlusion, and management; the model was manually calibrated to one anonymous greenhouse and not quantitatively validated across diverse cultivars, trellis systems, or climates (Sections 3.3, 5.2, Appendices A.3/A.7).

What would settle it

Run the released fine-tuned SAM 3 on a tomato greenhouse with a different cultivar or trellising system than the simulated cherry-tomato setup (e.g., beefsteak tomatoes on a high-wire system with different pruning) and compute fruit IoU; if the macro fruit IoU drops back to the zero-shot level (~52.7%) while the model still works on the original three datasets, the sim-to-real generalization is specific to the simulated greenhouse type, not to tomato greenhouses in general.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Fine-tuning a foundation model on synthetic data alone improves real tomato fruit segmentation by 12.9 percentage points macro IoU over zero-shot SAM 3.
  • Full fine-tuning gives the best aggregate transfer; LoRA and WiSE-FT LoRA also transfer strongly, with WiSE-FT LoRA best on LaboroTomato.
  • Supervised models trained from scratch on the same synthetic data fail to transfer (macro IoU below 0.23), so the general visual prior of the foundation model is what makes synthetic fine-tuning work.
  • Synthetic validation scores do not predict real-world transfer, so model selection must use real-domain evaluation (or label-free confidence metrics).
  • Because two real datasets have non-exhaustive fruit labels, the reported IoU may understate segmentation quality for models that detect all visible fruits.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same procedural pipeline should transfer to other trellised crops (cucumber, pepper) or additional organs, since the L-system parameters are explicit and only the rendering and management rules need changing.
  • A direct head-to-head against fine-tuning on a small real annotated set would establish whether synthetic-only fine-tuning is cost-competitive; the paper does not attempt this comparison.
  • The synthetic–real inversion suggests that a small but representative real validation set, or label-free confidence and TTA-agreement metrics like those used in the paper, should be part of any sim-to-real training pipeline.
  • The paper's own future-work caveat that semantic specialization may trade off against SAM 3's native detection behavior implies that multi-task benchmarks (segmentation plus detection) are needed before deployment in harvest robots.

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

3 major / 5 minor

Summary. The paper presents a sim-to-real pipeline for text-conditioned tomato organ segmentation. It combines a procedural L-system model of a commercial cherry tomato greenhouse, rendered in Unreal Engine 5, with fine-tuning of the Segment Anything Model 3 (SAM 3). The authors generate a 68k-frame synthetic dataset with pixel-perfect semantic/instance/depth annotations and train four SAM 3 variants (Full FT, LoRA, WiSE-FT, WiSE-FT LoRA), comparing them against the zero-shot SAM 3 baseline and three supervised baselines trained on the same synthetic data. Evaluation on three real-world mask datasets shows that all four fine-tuned SAM 3 variants improve macro fruit IoU over zero-shot, with the best variant (Full FT) raising it from 52.7% to 65.6%. Label-free confidence metrics and qualitative overlays are also reported on an unlabeled commercial greenhouse dataset. The procedural model, synthetic dataset, and fine-tuned weights are publicly released.

Significance. If the results hold, the paper would demonstrate that procedural synthetic data can specialize a foundation model for a dense agricultural segmentation task without real annotated training data, and would contribute a reusable virtual greenhousse and dataset. The work has several genuine strengths: it evaluates on three external real-world datasets, it releases the synthetic corpus, procedural model, and model weights, and it openly discusses label non-exhaustiveness and the synthetic-real performance inversion. The main weaknesses are statistical: all results are single-run point estimates, the headline number is the best of eight variants evaluated on the same test sets, and the macro IoU metric gives equal weight to a small, noisy dataset. These issues affect the magnitude and the 'significant improvement' claim, though the qualitative pattern that fine-tuned SAM 3 variants outperform zero-shot appears consistent.

major comments (3)
  1. [§4.2, Table 3] The reported 12.9 pp improvement is a single-run point estimate, and Full FT is selected as the best of eight model variants evaluated on the same three test sets. No confidence intervals, standard deviations across random seeds, or significance tests are reported. Since the paper acknowledges annotation noise (§5.2), the 4.4 pp margin between Full FT (0.6562) and WiSE-FT LoRA (0.6118) could plausibly be within run-to-run noise. The central claim 'significantly improves segmentation performance' (Abstract, §4.2) therefore needs support: please report mean ± std over at least three fine-tuning seeds for each variant, and a paired or permutation significance test across datasets (or justify why the fine-tuning is deterministic and noise-free).
  2. [Table 3, §5.2, §4.3] Rob2Pheno has only 105 images and non-exhaustive fruit annotations; the paper itself notes that comprehensive segmenters are penalized with false positives on it. The zero-shot model's under-segmentation (foreground fraction 0.3385 vs. 0.6574 for WiSE-FT in Table 4) may make it appear artificially better on non-exhaustive labels, while Fine-tuned models are penalized for producing additional true masks. Because macro fruit IoU is an unweighted mean, Rob2Pheno contributes one-third of the headline number and is the smallest dataset. Please add an evaluation protocol that is fair under non-exhaustive annotation, e.g., masked IoU restricted to regions around annotated fruits or a tolerant matching metric, and report per-dataset results with an estimate of how much the reported 12.9 pp gain could change under alternative annotation-fairness assumptions.
  3. [§3.3–3.4, §4.2] The manuscript attributes the transfer gain to the combination of synthetic data and fine-tuning, but does not include a control that decomposes the contribution of procedural greenhouse realism from the mere availability of many fruit examples. A control fine-tuning on a simpler synthetic fruit dataset (e.g., fruit collages without greenhouse structure) or on a small real tomato training set would clarify whether the gain is specific to the proposed procedural model or is largely a generic effect of seeing more fruit instances. Without such a control, the claim that the procedural greenhouse model is the enabling component remains under-supported, although the supervised baselines trained on the same synthetic data do indicate that the foundation-model prior is also essential.
minor comments (5)
  1. [§4.2, Table 3] The text states that WiSE-FT SAM 3 improved over zero-shot on LaboroTomato and Rob2Pheno but not TomatoMAP-Seg, yet Table 3 shows TomatoMAP-Seg IoU of 0.6192 for WiSE-FT versus 0.5938 for zero-shot, i.e., an improvement. The sentence contradicts the table and should be corrected.
  2. [Table 3] The rows in Table 3 are not ordered by any stated criterion. Sorting by macro IoU (or by family) would make the comparison easier to read.
  3. [Table 4] The entropy values for WiSE-FT and zero-shot are identical (0.1374), and the text says 'Entropy alone tied.' This is fine, but it is worth noting that the entropy metric as defined may be insensitive to the under-coverage/over-coverage difference; consider adding a brief comment in the text.
  4. [Table 1 / References] The author name 'Zhen wei Wu' appears split in the table and reference; this should be formatted consistently (e.g., 'Zhenwei Wu').
  5. [General] The term 'SAM 3' is used throughout, but the manuscript does not identify the exact checkpoint/version or provide a link to the model card; adding this would support reproducibility.

Circularity Check

0 steps flagged

No circular derivation: synthetic training and real test data are disjoint, and the headline IoU is a genuine out-of-distribution transfer measurement.

full rationale

The paper's central claim is that fine-tuning SAM 3 on procedurally generated synthetic greenhouse data improves text-conditioned fruit segmentation on real greenhouse datasets. The derivation chain is: (1) an L-system/UE5 procedural greenhouse model is built, (2) a 68k-frame synthetic dataset with pixel-perfect labels is rendered from it, (3) SAM 3 is fine-tuned on that synthetic corpus, and (4) the resulting models are evaluated on three external real-world mask datasets (TomatoMAP-Seg, LaboroTomato, Rob2Pheno) plus a qualitative unlabeled greenhouse set. The evaluation data are disjoint from the training data, and no parameter or constant is fitted to the external test sets. The reported macro fruit IoU improvement (0.5269 to 0.6562) is therefore an out-of-distribution measurement, not an identity or a fitted-value restatement. The paper itself flags the main evaluation caveat: 'LaboroTomato and Rob2Pheno are fruit-mask datasets with non-exhaustive annotations' (Section 4.2), so standard IoU can penalize comprehensive segmenters. That is an annotation-quality bias affecting how the metric should be read, not a circular derivation of the result from its inputs. Similarly, the label-free confidence metrics on the anonymous greenhouse are explicitly stated as 'purely indicative' (Section 4.3) and are secondary to the external IoU numbers; they use the model's own outputs, but they are not claimed as the quantitative proof of sim-to-real transfer. The cited prior work by overlapping authors (e.g., Cieslak et al. 2021 for the interactive L-system calibration method, Klein et al. 2024 for procedural tomato data) supplies methodology and context, and the botanical basis is attributed to independent sources such as Went (1944) and Torres Quezada (2023). None of these citations is used to forbid alternatives or to define the evaluation outcome. The paper also openly acknowledges a limitation: 'a rigorous evaluation under a matched protocol across all fine-tuning variants, including Full FT, remains ongoing' (Section 6), which is an honest statement of incompleteness rather than a disguised circular step. Overall, the claimed prediction is not equivalent to any fitted input; it is a real transfer experiment with known evaluation-pipeline caveats, so no significant circularity is present.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

No fundamentally new physical entities are introduced. The main free choices are model-building and training hyperparameters that shape the synthetic distribution and the final checkpoints; the core external-evaluation claim does not fit parameters to the real test sets.

free parameters (4)
  • α (WiSE-FT interpolation coefficient) = 0.75
    Chosen by hand to balance zero-shot and fine-tuned weights (Section 3.4); directly determines all WiSE-FT variant results.
  • LoRA rank and learning rate
    Not reported in the paper; chosen by hand and directly control LoRA transfer performance.
  • Stochastic morphology noise σ = σ ≤ 0.1 (µ = 1.0)
    Chosen by hand to diversify synthetic plants; controls the synthetic data distribution (Appendix B.1).
  • L-system growth and geometry parameters
    Plastochron, sigmoid growth functions, tropism elasticities, and leaf geometry were manually calibrated via interactive fitting (Appendices A.3-A.6); the synthetic domain depends on these choices.
axioms (4)
  • domain assumption SAM 3 provides a strong promptable visual prior that can be specialized without losing its generalization ability.
    The entire fine-tuning and weight-interpolation strategy assumes the pre-trained prior is preserved during synthetic fine-tuning (Section 3.4).
  • domain assumption The L-system + UE5 simulation faithfully represents commercial tomato greenhouse morphology, lighting, and management practices.
    Sim-to-real transfer depends on this; the model was manually calibrated and not quantitatively validated against measured real geometry (Section 3.3, Appendix A.3).
  • domain assumption Real benchmark annotations are sufficient for fruit IoU despite non-exhaustive labels.
    The paper acknowledges that LaboroTomato and Rob2Pheno non-exhaustive labels can penalize comprehensive models (Section 4.2), so the reported rankings partly depend on this assumption.
  • domain assumption Label-free confidence metrics on the unlabeled greenhouse are informative about segmentation quality.
    No ground truth exists for the commercial greenhouse dataset; TTA agreement and confidence are assumed to correlate with correctness, and the authors call these metrics 'purely indicative' (Section 4.3).

pith-pipeline@v1.3.0-alltime-deepseek · 19870 in / 11205 out tokens · 134724 ms · 2026-08-01T14:57:45.095719+00:00 · methodology

0 comments
read the original abstract

Vision-based automation is an excellent candidate for reducing manual labor in greenhouse crop production and phenotyping. However, progress is constrained by the lack of annotated training data. Recent advances in vision-based foundational models have shown promising results in zero-shot generalization to novel domains, but their performance drops in complex agricultural environments. In this work, we present a sim-to-real framework for tomato plant segmentation that combines synthetic data generation with fine-tuning of a foundation model. We model a commercial cherry tomato greenhouse and use it to generate a large-scale synthetic dataset under diverse viewpoints, lighting conditions, and plant morphology. Subsequently, we fine-tune the Segment Anything Model 3 (SAM 3) on the synthetic dataset, specializing its text-conditioned segmentation behavior for greenhouse crop organs while retaining the general visual prior that makes zero-shot transfer possible. By evaluating our framework on multiple real-world greenhouse datasets, we demonstrate that combining synthetic data with SAM 3 fine-tuning significantly improves segmentation performance and model confidence. To support community benchmarking, we publicly release the procedural model, the generated synthetic dataset, and our fine-tuned SAM 3 weights.

Figures

Figures reproduced from arXiv: 2607.18576 by Ahmed M. Eltawil, Dominik L. Michels, Gianluca Setti, Hakim Ghazzai, Jonathan Klein, Katja Froehlich, Mikolaj Cieslak, Najmeddine Dhieb, Samy Mounir, Soeren Pirk, Wojciech Palubicki.

Figure 1
Figure 1. Figure 1: Example of a synthetic image and its corresponding labels. Semantic masks are colored red for fruit, green [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison of the supervised model DeepLabV3+, zero-shot SAM3, and our fully fine-tuned [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Prompt simplicity analysis. A single image is segmented using concise organ-level prompts: “stem”, “leaf” [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Simulation of bifurcated infructescence: photograph (left) vs. model visualization (right). [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Simulation of the leaning and lowering management practice. [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Variation in plant morphology over time. [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

148 extracted references · 26 canonical work pages

  1. [1]

    Tomato (solanum lycopersicum): A model fruit-bearing crop

    Seisuke Kimura and Neelima Sinha. Tomato (solanum lycopersicum): A model fruit-bearing crop. Cold Spring Harbor Protocols, 2008 0 (11): 0 pdb.emo105, 2008. doi:10.1101/pdb.emo105

    work page doi:10.1101/pdb.emo105 2008

  2. [2]

    Cobb, Genevieve DeClerck, Anthony Greenberg, Randy Clark, and Susan McCouch

    Joshua N. Cobb, Genevieve DeClerck, Anthony Greenberg, Randy Clark, and Susan McCouch. Next-generation phenotyping: requirements and strategies for enhancing our understanding of genotype--phenotype relationships and its relevance to crop improvement. Theoretical and Applied Genetics, 126 0 (4): 0 867--887, Apr 2013. ISSN 1432-2242. doi:10.1007/s00122-013-2066-0

    work page doi:10.1007/s00122-013-2066-0 2013

  3. [4]

    Quantitative extraction and evaluation of tomato fruit phenotypes based on image recognition

    Yihang Zhu, Qing Gu, Yiying Zhao, Hongjian Wan, Rongqing Wang, Xiaobin Zhang, and Yuan Cheng. Quantitative extraction and evaluation of tomato fruit phenotypes based on image recognition. Frontiers in Plant Science, 13, 2022. ISSN 1664-462X. doi:10.3389/fpls.2022.859290

    arXiv 2022

  4. [5]

    Tomato multi-angle multi-pose dataset for fine-grained phenotyping

    Yujie Zhang, Sabine Struckmeyer, Andreas Kolb, and Sven Reichardt. Tomato multi-angle multi-pose dataset for fine-grained phenotyping. Sci Data, 13: 0 309, 2 2026. ISSN 2052-4463. doi:https://doi.org/10.1038/s41597-026-06926-9

    work page doi:10.1038/s41597-026-06926-9 2026

  5. [6]

    Kasper Johansen, Mitchell J. L. Morton, Yoann Malbeteau, Bruno Aragon, Samer Al-Mashharawi, Matteo G. Ziliani, Yoseline Angel, Gabriele Fiene, Sónia Negrão, Magdi A. A. Mousa, Mark A. Tester, and Matthew F. McCabe. Predicting biomass and yield in a tomato phenotyping experiment using uav imagery and random forest. Frontiers in Artificial Intelligence, Vol...

    arXiv 2020

  6. [7]

    3dphenomvs: A low-cost 3d tomato phenotyping pipeline using 3d reconstruction point cloud based on multiview images

    Yinghua Wang, Songtao Hu, He Ren, Wanneng Yang, and Ruifang Zhai. 3dphenomvs: A low-cost 3d tomato phenotyping pipeline using 3d reconstruction point cloud based on multiview images. Agronomy, 12 0 (8), 2022. ISSN 2073-4395. doi:10.3390/agronomy12081865

    work page doi:10.3390/agronomy12081865 2022

  7. [9]

    White, J

    Emma L. White, J. Alex Thomasson, Brent Auvermann, Newell R. Kitchen, Leland Sandy Pierson, Dana Porter, Craig Baillie, Hendrik Hamann, Gerrit Hoogenboom, Todd Janzen, Rajiv Khosla, James Lowenberg-DeBoer, Matt McIntosh, Seth Murray, Dave Osborn, Ashoo Shetty, Craig Stevenson, Joe Tevis, and Fletcher Werner. Report from the conference, `identifying obstac...

    work page doi:10.1007/s11119-020-09738-y 2021

  8. [10]

    Michels, Soren Pirk, Chia-Chun Fu, and Wojciech Palubicki

    Mikolaj Cieslak, Umabharathi Govindarajan, Alejandro Garcia, Anuradha Chandrashekar, Torsten Hadrich, Aleksander Mendoza-Drosik, Dominik L. Michels, Soren Pirk, Chia-Chun Fu, and Wojciech Palubicki. Generating diverse agricultural data for vision-based farming applications. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2024

  9. [11]

    Jonathan Klein, Rebekah Waller, Sören Pirk, Wojtek Pałubicki, Mark Tester, and Dominik L. Michels. Synthetic data at scale: a development model to efficiently leverage machine learning in agriculture. Frontiers in Plant Science, Volume 15 - 2024, 2024. ISSN 1664-462X. doi:10.3389/fpls.2024.1360113

    arXiv 2024

  10. [12]

    Effectiveness of training with procedurally generated synthetic images of crop plants

    Nazifa Khan, Mikolaj Cieslak, Mark Eramian, and Ian McQuillan. Effectiveness of training with procedurally generated synthetic images of crop plants. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 5490--5500, June 2025

    2025

  11. [13]

    Winsyn: A high resolution testbed for synthetic data

    Tom Kelly, John Femiani, and Peter Wonka. Winsyn: A high resolution testbed for synthetic data. In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 22456--22465, 2024. doi:10.1109/CVPR52733.2024.02119

    arXiv 2024

  12. [14]

    Bhattacharyya

    Hyungtae Lee, Yan Zhang, Heesung Kwon, and Shuvra S. Bhattacharyya. Exploring the potential of synthetic data to replace real data. In 2024 IEEE International Conference on Image Processing (ICIP), pages 1005--1011, 2024. doi:10.1109/ICIP51287.2024.10647508

    arXiv 2024

  13. [15]

    Sam 3: Segment anything with concepts, 2025

    Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoubhik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman Rädle, Triantafyllos Afouras, Effrosyni Mavroudi, Katherine Xu, Tsung-Han Wu, Yu Zhou, Liliane ...

    2025

  14. [16]

    Segment anything is not always perfect: An investigation of sam on different real-world applications

    Wei Ji, Jingjing Li, Qi Bi, Tingwei Liu, Wenbo Li, and Li Cheng. Segment anything is not always perfect: An investigation of sam on different real-world applications. Machine Intelligence Research, 21 0 (4): 0 617--630, Aug 2024. ISSN 2731-5398. doi:10.1007/s11633-023-1385-0

    work page doi:10.1007/s11633-023-1385-0 2024

  15. [18]

    Tsaftaris

    Feng Chen, Mario Valerio Giuffrida, and Sotirios A. Tsaftaris. Adapting vision foundation models for plant phenotyping. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, pages 604--613, October 2023

    2023

  16. [19]

    Evaluating the efficacy of segment anything model for delineating agriculture and urban green spaces in multiresolution aerial and spaceborne remote sensing images

    Baoling Gui, Anshuman Bhardwaj, and Lydia Sam. Evaluating the efficacy of segment anything model for delineating agriculture and urban green spaces in multiresolution aerial and spaceborne remote sensing images. Remote Sensing, 16 0 (2), 2024. ISSN 2072-4292. doi:10.3390/rs16020414

    work page doi:10.3390/rs16020414 2024

  17. [20]

    Agri-fm+: A self-supervised foundation model for agricultural vision

    Md Jaber Al Nahian, Tapotosh Ghosh, Farnaz Sheikhi, and Farhad Maleki. Agri-fm+: A self-supervised foundation model for agricultural vision. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 5550--5562, June 2025

    2025

  18. [21]

    Few-shot adaptation of grounding dino for agricultural domain

    Rajhans Singh, Rafael Bidese Puhl, Kshitiz Dhakal, and Sudhir Sornapudi. Few-shot adaptation of grounding dino for agricultural domain. 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 5332--5342, 2025

    2025

  19. [22]

    Bailey, and J

    Lars Lundqvist, Earl Ranario, Hamid Kamangir, Heesup Yun, Christine Diepenbrock, Brian N. Bailey, and J. Mason Earles. Does your vfm speak plant? the botanical grammar of vision foundation models for object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 9768--9776, June 2026

    2026

  20. [23]

    Kai Shen, Xue Zhou, Santhi Daggubati, Xu Wang, Liyike Ji, Daeun Choi, and Uchechukwu D. Ilodibe. A field-based plant video segmentation pipeline (fpvsp) for off-line plant tracking through zero-shot segmentation and motion features. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 9824--9833, June 2026

    2026

  21. [25]

    The use of plant models in deep learning: an application to leaf counting in rosette plants

    Jordan Ubbens, Mikolaj Cieslak, Przemyslaw Prusinkiewicz, and Ian Stavness. The use of plant models in deep learning: an application to leaf counting in rosette plants. Plant Methods, 14 0 (1): 0 6, Jan 2018. ISSN 1746-4811. doi:10.1186/s13007-018-0273-z

    work page doi:10.1186/s13007-018-0273-z 2018

  22. [26]

    D. Ward, P. Moghadam, and N. Hudson. Deep leaf segmentation using synthetic data. In Proceedings of the British Machine Vision Conference (BMVC) Workshop on Computer Vision Problems in Plant Phenotyping (CVPPP), 2018

    2018

  23. [27]

    Schwing, Robert Brunner, Hrant Khachatrian, Hovnatan Karapetyan, Ivan Dozier, Greg Rose, David Wilson, Adrian Tudor, Naira Hovakimyan, Thomas S

    Mang Tik Chiu, Xingqian Xu, Yunchao Wei, Zilong Huang, Alexander G. Schwing, Robert Brunner, Hrant Khachatrian, Hovnatan Karapetyan, Ivan Dozier, Greg Rose, David Wilson, Adrian Tudor, Naira Hovakimyan, Thomas S. Huang, and Honghui Shi. Agriculture-vision: A large aerial image database for agricultural pattern analysis. In 2020 IEEE/CVF Conference on Comp...

    arXiv 2020

  24. [28]

    Konovalov, Bronson Philippa, Peter Ridd, Jake C

    Alex Olsen, Dmitry A. Konovalov, Bronson Philippa, Peter Ridd, Jake C. Wood, Jamie Johns, Wesley Banks, Benjamin Girgenti, Owen Kenny, James Whinney, Brendan Calvert, Mostafa Rahimi Azghadi , and Ronald D. White. DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning . Scientific Reports, 9 0 (2058), 2 2019. doi:10.1038/s41598-018-38343-3

    work page doi:10.1038/s41598-018-38343-3 2058

  25. [29]

    A realistic synthetic mushroom scenes dataset

    Dafni Anagnostopoulou, George Retsinas, Niki Efthymiou, Panagiotis Filntisis, and Petros Maragos. A realistic synthetic mushroom scenes dataset. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 6282--6289, June 2023

    2023

  26. [30]

    Cropdeep: The crop vision dataset for deep-learning-based classification and detection in precision agriculture

    Yang-Yang Zheng, Jian-Lei Kong, Xue-Bo Jin, Xiao-Yi Wang, Ting-Li Su, and Min Zuo. Cropdeep: The crop vision dataset for deep-learning-based classification and detection in precision agriculture. Sensors, 19 0 (5), 2019. doi:10.3390/s19051058

    work page doi:10.3390/s19051058 2019

  27. [31]

    Deep convolutional neural networks for image-based convolvulus sepium detection in sugar beet fields

    Junfeng Gao, Andrew P French, Michael P Pound, Yong He, Tony P Pridmore, and Jan G Pieters. Deep convolutional neural networks for image-based convolvulus sepium detection in sugar beet fields. Plant Methods, 16 0 (1): 0 29, March 2020

    2020

  28. [32]

    Domain generalization for crop segmentation with standardized ensemble knowledge distillation

    Simone Angarano, Mauro Martini, Alessandro Navone, and Marcello Chiaberge. Domain generalization for crop segmentation with standardized ensemble knowledge distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 5450--5459, June 2024

    2024

  29. [34]

    Plantdreamer: Achieving realistic 3d plant models with diffusion-guided gaussian splatting

    Zane K J Hartley, Lewis A G Stuart, Andrew P French, and Michael P Pound. Plantdreamer: Achieving realistic 3d plant models with diffusion-guided gaussian splatting. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, pages 7069--7079, October 2025

    2025

  30. [36]

    Sapkota, Sorin Popescu, Nithya Rajan, Ramon G

    Bishwa B. Sapkota, Sorin Popescu, Nithya Rajan, Ramon G. Leon, Chris Reberg-Horton, Steven Mirsky, and Muthukumar V. Bagavathiannan. Use of synthetic images for training a deep learning model for weed detection and biomass estimation in cotton. Scientific Reports, 12 0 (1): 0 19580, Nov 2022. ISSN 2045-2322. doi:10.1038/s41598-022-23399-z

    work page doi:10.1038/s41598-022-23399-z 2022

  31. [37]

    Synthset: Generative diffusion model for semantic segmentation in precision agriculture

    Andrew Heschl, Mauricio Murillo, Keyhan Najafian, and Farhad Maleki. Synthset: Generative diffusion model for semantic segmentation in precision agriculture. In Alessio Del Bue, Cristian Canton, Jordi Pont-Tuset, and Tatiana Tommasi, editors, Computer Vision -- ECCV 2024 Workshops, pages 267--283, Cham, 2025. Springer Nature Switzerland. ISBN 978-3-031-91835-3

    2024

  32. [38]

    Beyond annotations: Efficient wheat head segmentation using l-systems, game engines, and student-teacher models

    Hosein Beheshtifard, Elijah Mickelson, Keyhan Najafian, and Farhad Maleki. Beyond annotations: Efficient wheat head segmentation using l-systems, game engines, and student-teacher models. In Alessio Del Bue, Cristian Canton, Jordi Pont-Tuset, and Tatiana Tommasi, editors, Computer Vision -- ECCV 2024 Workshops, pages 319--334, Cham, 2025. Springer Nature ...

    2024

  33. [40]

    Efficient wheat head segmentation with minimal annotation: A generative approach

    Jaden Myers, Keyhan Najafian, Farhad Maleki, and Katie Ovens. Efficient wheat head segmentation with minimal annotation: A generative approach. Journal of Imaging, 10 0 (7), 2024. ISSN 2313-433X. doi:10.3390/jimaging10070152

    work page doi:10.3390/jimaging10070152 2024

  34. [43]

    A dataset for semantic and instance segmentation of modern fruit orchards

    Tieqiao Wang, Abhinav Jain, Liqiang He, Cindy Grimm, and Sinisa Todorovic. A dataset for semantic and instance segmentation of modern fruit orchards. In 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 5381--5391, 2025. doi:10.1109/CVPRW67362.2025.00535

    arXiv 2025

  35. [46]

    De Visser, Gerie van der Heijden, and Gerhard Buck-Sorlin

    Pieter H. De Visser, Gerie van der Heijden, and Gerhard Buck-Sorlin. Optimizing illumination in the greenhouse using a 3d model of tomato and a ray tracer. Frontiers in Plant Science, Volume 5, 2014. ISSN 1664-462X. doi:10.3389/fpls.2014.00048

    arXiv 2014

  36. [48]

    A functional–structural plant model for dwarf tomato ideotype identification in vertical farming

    Michele Butturini, Katarína Smoleňová, Jacopo Restina, Jacob Stolz, Jorad de Vries, and Leo F M Marcelis. A functional–structural plant model for dwarf tomato ideotype identification in vertical farming. in silico Plants, 8 0 (1): 0 diaf024, 12 2025. ISSN 2517-5025. doi:10.1093/insilicoplants/diaf024

    work page doi:10.1093/insilicoplants/diaf024 2025

  37. [49]

    Junchen Deng, Samhita Marri, Jonathan Klein, Wojtek Pałubicki, Sören Pirk, Girish Chowdhary, and Dominik L. Michels. Gazebo plants: Simulating plant-robot interaction with cosserat rods, 2024

    2024

  38. [51]

    LaboroTomato : Instance segmentation dataset

    Roman Trigubenko, Dewen Xu, and Hiromasa Fujihara. LaboroTomato : Instance segmentation dataset. https://github.com/laboroai/LaboroTomato, 2020. Accessed: 2024-05-22

    2020

  39. [52]

    Tomato fruit detection and counting in greenhouses using deep learning

    Manya Afonso, Hubert Fonteijn, Felipe Schadeck Fiorentin, Dick Lensink, Marcel Mooij, Nanne Faber, Gerrit Polder, and Ron Wehrens. Tomato fruit detection and counting in greenhouses using deep learning. Frontiers in Plant Science, Volume 11 - 2020, 2020. ISSN 1664-462X. doi:10.3389/fpls.2020.571299

    arXiv 2020

  40. [53]

    Tsironis, S

    V. Tsironis, S. Bourou, and C. Stentoumis. tomatOD : Evaluation of object detection algorithms on a new real-world tomato dataset. In ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2020

    2020

  41. [54]

    Benchmark of deep learning and a proposed hsv colour space models for the detection and classification of greenhouse tomato

    Germano Moreira, Sandro Augusto Magalhães, Tatiana Pinho, Filipe Neves dos Santos, and Mário Cunha. Benchmark of deep learning and a proposed hsv colour space models for the detection and classification of greenhouse tomato. Agronomy, 12 0 (2), 2022. ISSN 2073-4395. doi:10.3390/agronomy12020356

    work page doi:10.3390/agronomy12020356 2022

  42. [56]

    The Algorithmic Beauty of Plants

    P Prusinkiewicz and A Lindenmayer. The Algorithmic Beauty of Plants. Springer, New York, 1990. With JS Hanan, FD Fracchia, DR Fowler, MJM de Boer, and L Mercer

    1990

  43. [57]

    M e ch and P

    R. M e ch and P. Prusinkiewicz. Visual models of plants interacting with their environment. In Proceedings of the 23rd Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH '96, pages 397--410, New York, NY, USA, 1996. ACM

    1996

  44. [59]

    F. W. Went. Morphological observations on the tomato plant. Bulletin of the Torrey Botanical Club, 71 0 (1): 0 77--92, 1944. ISSN 00409618, 23258055

    1944

  45. [60]

    Torres Quezada

    E. Torres Quezada. Basic tomato ( Lycopersicon esculentum) physiology and morphology. Extension Publication SPES-508, Virginia Cooperative Extension, 2023

    2023

  46. [61]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations (ICLR), 2022 b

    2022

  47. [62]

    Robust fine-tuning of zero-shot models

    Mitchell Wortsman, Gabriel Ilharco, Jong Wook Kim, Mike Li, Simon Kornblith, Rebecca Roelofs, Raphael Gontijo Lopes, Hannaneh Hajishirzi, Ali Farhadi, Hongseok Namkoong, and Ludwig Schmidt. Robust fine-tuning of zero-shot models. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7949--7961, 2022 a . doi:10.1109/CVPR52688...

    arXiv 2022

  48. [63]

    Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time

    Mitchell Wortsman, Gabriel Ilharco, Samir Ya Gadre, Rebecca Roelofs, Raphael Gontijo-Lopes, Ari S Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, and Ludwig Schmidt. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time. In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csaba...

    2022

  49. [64]

    Encoder-decoder with atrous separable convolution for semantic image segmentation

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Vittorio Ferrari, Martial Hebert, Cristian Sminchisescu, and Yair Weiss, editors, Computer Vision -- ECCV 2018, pages 833--851, Cham, 2018. Springer International Publishing. ISBN 978-3-03...

    2018

  50. [65]

    Segformer: Simple and efficient design for semantic segmentation with transformers

    Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M Alvarez, and Ping Luo. Segformer: Simple and efficient design for semantic segmentation with transformers. In Neural Information Processing Systems (NeurIPS), 2021

    2021

  51. [66]

    Your ViT is Secretly an Image Segmentation Model

    Tommie Kerssies, Niccol\` o Cavagnero, Alexander Hermans, Narges Norouzi, Giuseppe Averta, Bastian Leibe, Gijs Dubbelman, and Daan de Geus . Your ViT is Secretly an Image Segmentation Model . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    2025

  52. [68]

    Zero-shot hierarchical plant segmentation via foundation segmentation models and text-to-image attention

    Junhao Xing, Ryohei Miyakawa, Yang Yang, Xinpeng Liu, Risa Shinoda, Hiroaki Santo, Yosuke Toda, and Fumio Okura. Zero-shot hierarchical plant segmentation via foundation segmentation models and text-to-image attention. In 2026 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 2510--2519, 2026. doi:10.1109/WACV61042.2026.00247

    arXiv 2026

  53. [69]

    An opinion on imaging challenges in phenotyping field crops

    Derek Kelly, Avimanyou Vatsa, Wade Mayham, Linh Ng \^o , Addie Thompson, and Toni Kazic. An opinion on imaging challenges in phenotyping field crops. Machine Vision and Applications, 27 0 (5): 0 681--694, Jul 2016. ISSN 1432-1769. doi:10.1007/s00138-015-0728-4

    work page doi:10.1007/s00138-015-0728-4 2016

  54. [70]

    Recognition and localization methods for vision-based fruit picking robots: A review

    Yunchao Tang, Mingyou Chen, Chenglin Wang, Lufeng Luo, Jinhui Li, Guoping Lian, and Xiangjun Zou. Recognition and localization methods for vision-based fruit picking robots: A review. Frontiers in Plant Science, Volume 11 - 2020, 2020. ISSN 1664-462X. doi:10.3389/fpls.2020.00510

    arXiv 2020

  55. [71]

    Quantifying the importance of a realistic tomato (solanum lycopersicum) leaflet shape for 3-d light modelling

    Jonathan Vermeiren, Selwyn L Y Villers, Lieve Wittemans, Wendy Vanlommel, Jeroen van Roy, Herman Marien, Jonas R Coussement, and Kathy Steppe. Quantifying the importance of a realistic tomato (solanum lycopersicum) leaflet shape for 3-d light modelling. Annals of Botany, 126 0 (4): 0 661--670, 12 2019. ISSN 0305-7364. doi:10.1093/aob/mcz205

    work page doi:10.1093/aob/mcz205 2019

  56. [72]

    Development of a tomato functional–structural plant model for digital twin applications

    Katarína Smoleňová, Nastassia Vilfan, Daniela Bustos-Korts, Tim van Daalen, Pieter de Visser, Elias Kaiser, Rick van de Zedde, and Jochem Evers. Development of a tomato functional–structural plant model for digital twin applications. in silico Plants, 7 0 (2): 0 diaf022, 11 2025. ISSN 2517-5025. doi:10.1093/insilicoplants/diaf022

    work page doi:10.1093/insilicoplants/diaf022 2025

  57. [73]

    Xfrog - procedural organic 3d modeler [3d models]

    Bernd Lintermann and Oliver Deussen. Xfrog - procedural organic 3d modeler [3d models]. Software/Tool, 1999. URL https://www.xfrog.com/. Procedural organic modeling tool for 3D plants

    1999

  58. [74]

    Prusinkiewicz, R

    P. Prusinkiewicz, R. Karwowski, and B. Lane. The L+C plant-modeling language. In J. Vos, editor, Functional-structural modeling in crop production, pages 27--42. Springer, Dordrecht, 2007

    2007

  59. [75]

    The use of positional information in the modeling of plants

    Przemyslaw Prusinkiewicz, Lars M\" u ndermann, Radoslaw Karwowski, and Brendan Lane. The use of positional information in the modeling of plants. In Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH '01, pages 289--300, New York, NY, USA, 2001. ACM

    2001

  60. [76]

    Mohanty, David P

    Sharada P. Mohanty, David P. Hughes, and Marcel Salathé. Using deep learning for image-based plant disease detection. Frontiers in Plant Science, 7, Sep 2016

    2016

  61. [77]

    COSYS-AIRSIM : A real-time simulation framework expanded for complex industrial applications

    Wouter Jansen, Edward Verreycken, Dimitrios Zermas, and Lode Claesen. COSYS-AIRSIM : A real-time simulation framework expanded for complex industrial applications. In Annual Modeling and Simulation Conference (ANNSIM), pages 37--48. IEEE, 2023. doi:10.23919/ANNSIM57446.2023.10194672

    arXiv 2023

  62. [78]

    in silico Plants , volume =

    Cieslak, Mikolaj and Khan, Nazifa and Ferraro, Pascal and Soolanayakanahally, Raju and Robinson, Stephen J and Parkin, Isobel and McQuillan, Ian and Prusinkiewicz, Przemyslaw , title =. in silico Plants , volume =. 2021 , month =

    2021

  63. [79]

    and Pirk, Soren and Fu, Chia-Chun and Palubicki, Wojciech , title =

    Cieslak, Mikolaj and Govindarajan, Umabharathi and Garcia, Alejandro and Chandrashekar, Anuradha and Hadrich, Torsten and Mendoza-Drosik, Aleksander and Michels, Dominik L. and Pirk, Soren and Fu, Chia-Chun and Palubicki, Wojciech , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops , month =. 2024 ...

    2024

  64. [80]

    Graph-Grammars and Their Application to Computer Science , pages =

    Applications of L-systems to computer imagery , author =. Graph-Grammars and Their Application to Computer Science , pages =. 1987 , organization =

    1987

  65. [81]

    and Lindenmayer, A

    Prusinkiewicz, P. and Lindenmayer, A. and Hanan, J. , title =. Computer Graphics , year =

  66. [82]

    The Algorithmic Beauty of Plants , publisher =

    Prusinkiewicz, P and Lindenmayer, A , year =. The Algorithmic Beauty of Plants , publisher =

  67. [83]

    and Mjolsness, Eric , title =

    Prusinkiewicz, Przemyslaw and Hammel, Mark S. and Mjolsness, Eric , title =. Proceedings of the 20th Annual Conference on Computer Graphics and Interactive Techniques , series =. 1993 , pages =

    1993

  68. [84]

    The Use of Positional Information in the Modeling of Plants , booktitle =

    Prusinkiewicz, Przemyslaw and M\". The Use of Positional Information in the Modeling of Plants , booktitle =. 2001 , pages =

    2001

  69. [85]

    R. M. Visual models of plants interacting with their environment , booktitle =. 1996 , pages =

    1996

  70. [86]

    , title =

    Prusinkiewicz, P. , title =. Acta Horticulturae , year = 2004, volume =

    2004

  71. [87]

    Prusinkiewicz and R

    P. Prusinkiewicz and R. Karwowski and B. Lane , year = 2007, title =. Functional-structural modeling in crop production , publisher =

    2007

  72. [88]

    Mercer, P

    L. Mercer, P. Prusinkiewicz, J. Hanan , title =. Proceedings of Graphics Interface’90 , year =

  73. [89]

    Frontiers in Plant Science , author =

    Using deep learning for image-based plant disease detection , volume =. Frontiers in Plant Science , author =. 2016 , month =

    2016

  74. [90]

    Annual Modeling and Simulation Conference (ANNSIM) , pages =

    Jansen, Wouter and Verreycken, Edward and Zermas, Dimitrios and Claesen, Lode , title =. Annual Modeling and Simulation Conference (ANNSIM) , pages =. 2023 , organization =

    2023

  75. [91]

    and de Visser, P

    Sarlikioti, V. and de Visser, P. H. B. and Buck-Sorlin, G. H. and Marcelis, L. F. M. , title =. Annals of Botany , volume =. 2011 , month =. doi:10.1093/aob/mcr221 , eprint =

    work page doi:10.1093/aob/mcr221 2011

  76. [92]

    Annals of Botany , volume =

    Vermeiren, Jonathan and Villers, Selwyn L Y and Wittemans, Lieve and Vanlommel, Wendy and van Roy, Jeroen and Marien, Herman and Coussement, Jonas R and Steppe, Kathy , title =. Annals of Botany , volume =. 2019 , month =

    2019

  77. [93]

    2020 , issn =

    High resolution 3D simulation of light climate and thermal performance of a solar greenhouse model under tomato canopy structure , journal =. 2020 , issn =. doi:https://doi.org/10.1016/j.renene.2020.06.144 , author =

    work page doi:10.1016/j.renene.2020.06.144 2020

  78. [94]

    in silico Plants , volume =

    Smoleňová, Katarína and Vilfan, Nastassia and Bustos-Korts, Daniela and van Daalen, Tim and de Visser, Pieter and Kaiser, Elias and van de Zedde, Rick and Evers, Jochem , title =. in silico Plants , volume =. 2025 , month =

    2025

  79. [95]

    in silico Plants , volume =

    Butturini, Michele and Smoleňová, Katarína and Restina, Jacopo and Stolz, Jacob and de Vries, Jorad and Marcelis, Leo F M , title =. in silico Plants , volume =. 2025 , month =

    2025

  80. [96]

    and van der Heijden, Gerie and Buck-Sorlin, Gerhard , title =

    De Visser, Pieter H. and van der Heijden, Gerie and Buck-Sorlin, Gerhard , title =. Frontiers in Plant Science , volume =. 2014 , doi =

    2014

Showing first 80 references.