Pith. sign in

REVIEW 3 major objections 6 minor 33 references

Whole leaf vein networks expose three gene–geography links

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 06:26 UTC pith:6RTGG6T4

load-bearing objection Genuinely new pipeline for whole-network leaf vein association, but the real-data 'significant' interactions have no inferential support—fixable, but currently overclaimed. the 3 major comments →

arxiv 2607.22763 v1 pith:6RTGG6T4 submitted 2026-07-23 cs.LG stat.ML

An Integrated Deep Learning and Statistical Framework for Whole-Network Gene--Environment Association with Leaf Vascular Architecture

Geran Zhao , Yangsheng Wang , Xiaotian Dai , Guifang Fu This is my paper
classification cs.LG stat.ML MSC 62H2062P1068T45
keywords leaf vascular architecturegene-environment associationcanonical correlation analysisedge detectiontransformerwhole-network phenotypezero-inflated dataPopulus
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 argues that a leaf's complete vascular network—every vein, not just summary statistics like vein density—can serve as the phenotype in gene–environment association studies. The authors build a pipeline that fine-tunes a transformer-based edge detector to turn RGB leaf images into whole-network edge maps, then applies a semiparametric sparse canonical correlation analysis to relate these bivariate image responses to 118 genetic and geographic predictors. Simulations show the statistical step recovers known causal predictors, and on a real Populus dataset the pipeline selects three gene–geography interaction terms with nonzero coefficients. If the claim holds, researchers no longer need to throw away most of the structural information in leaf images when hunting for genotype–phenotype associations.

Core claim

The central claim is that leaf vascular architecture can be represented as a whole-network image phenotype and that this representation, together with a statistical model built for sparse zero-inflated images, can detect gene–environment associations that summary traits would miss. To the authors' knowledge this is the first study to do so. The concrete results are: the fine-tuned EDTER model extracts complete vein networks, the new annotated leaf database (built from pseudo-labels) enables that fine-tuning, SSCCA handles the bivariate front/back edge maps, and the real-data analysis selects GCPM_1053-1 × longitude, GCPM_1036-1 × elevation, and GCPM_1131 × latitude as the only nonzero canoni

What carries the argument

EDTER (Edge Detection with Transformers), fine-tuned on a leaf image database with pseudo-labels from a diffusion-based edge detector, converts an RGB leaf into a pixel-level edge map of the whole vascular network; SSCCA (Semiparametric Sparse Canonical Correlation Analysis) then estimates the correlation between bivariate image responses and high-dimensional predictors through a truncated latent Gaussian copula, using an inverse bridge function to turn Kendall's tau into latent correlations and an l1 penalty to select a sparse set of predictors. The copula structure is what lets the method handle edge maps that are mostly zero-valued background pixels.

Load-bearing premise

The load-bearing premise is that the pseudo-labeled edge maps used to fine-tune EDTER are accurate enough to represent the true vascular network; no manual validation of those labels is reported, and if they systematically miss or invent veins, the whole phenotype and the three found interactions are biased.

What would settle it

Manually trace veins on a random subset of the 100 Populus leaves (or the 324 training leaves) and compare with the fine-tuned EDTER output; if the automated edge maps have a high false-positive or false-negative rate on tertiary and quaternary veins, re-run SSCCA on hand-corrected edge maps and check whether the three gene–geography interactions disappear.

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

If this is right

  • Leaf vascular architecture can enter gene–environment studies as a whole image rather than a few summary traits, preserving local details and global topology.
  • SSCCA provides a working statistical model for bivariate, repeatedly-measured image responses that are sparse and zero-inflated, with variable selection built in.
  • Pseudo-labels from a diffusion-based edge detector are a scalable route to fine-tune edge detection for a new image domain without manual annotation.
  • The three reported gene–geography interactions (GCPM_1053-1 × longitude, GCPM_1036-1 × elevation, GCPM_1131 × latitude) are concrete hypotheses for follow-up in larger Populus populations.
  • The framework transfers to other network-structured image phenotypes, such as root architecture or vascular imaging, where ground-truth labels are scarce.

Where Pith is reading between the lines

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

  • Editorial implication: because fine-tuning used pseudo-labels as ground truth, a small manual validation study on a random subset of the 324 leaves would test whether the edge maps reliably capture fine veins; if not, the three real-data interactions may be artifacts of the edge detector rather than biology.
  • Editorial implication: the 60x40 resizing, chosen for computational balance, smooths pixel-level detail; a multi-resolution sensitivity check would show whether the three selected interactions persist at higher resolution, strengthening the claim that they are not resolution artifacts.
  • Editorial implication: SSCCA treats pixels as locations in a vector with an estimated correlation matrix, but ignores spatial adjacency; adding a spatial prior or using a tensor-variate extension could improve power for detecting interactions that affect connected vein regions.
  • Editorial implication: the strongest test of the paper's value is whether the three interactions replicate in an independent Populus sample and whether they associate with known vein traits (e.g., vein density, areole count) from the same leaves.

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 / 6 minor

Summary. The paper proposes an integrated framework for gene–environment association analysis using whole-network leaf vascular architecture as an image phenotype. The authors fine-tune the EDTER transformer-based edge detector on a new leaf dataset with pseudo-labels generated by DiffusionEdge, then apply Semiparametric Sparse Canonical Correlation Analysis (SSCCA) to bivariate (front/back) edge-map images of Populus leaves, with predictors including genetic markers, geography, and their interactions. Two simulation studies (n=200, p1=50) show that SSCCA can select true predictors with low false selection rates. In a real dataset of 100 leaves (200 images; 118 predictors; 4800 response dimensions after resizing), SSCCA selects three gene–geography interactions, which the paper calls 'significant.' The paper claims to be the first to use deep-learning whole-network leaf phenotypes in gene–environment association studies.

Significance. The paper introduces a plausible pipeline: fine-tuning a transformer-based edge detector (EDTER) with pseudo-labels from DiffusionEdge to produce whole-network leaf vein images, and then applying SSCCA to bivariate front/back image responses. If the real-data findings were supported by proper inference, this would be a valuable contribution, as it moves beyond low-dimensional vein traits and accommodates the zero-inflated structure of edge-map images. The two simulations demonstrate that SSCCA can recover planted predictors in favorable settings (n=200, p1=50, FSR ≈ 0.02), which is a useful sanity check. The bivariate response design (front and back of the same leaf) is a sensible way to exploit repeated measurements. However, the current manuscript does not establish the central biological claim, and the simulation design partially reuses the same phenotype extractor, so the evidence for the framework's end-to-end validity is incomplete.

major comments (3)
  1. [Section 4] The claim of 'three significant gene–geography interactions' is not supported by any inferential procedure. SSCCA with BIC (Eq. 3) selects variables by penalized optimization; in a problem with n=100, p1=118, p2=4800, nonzero canonical coefficients can be selected by chance. No p-values, permutation tests, confidence intervals, or multiple-testing control are reported. The simulations in Section 3 use n=200 and p1=50, with FSR≈0.02, which does not calibrate the real-data regime; a null simulation with the real-data dimensions is needed to assess the expected number of false selections. The Section 5 limitation paragraph acknowledges small sample size but does not retract the word 'significant' in the abstract and Section 4. Because the central biological claim rests on these three selections, the paper must either provide permutation-based significance evidence or explicitly downgrade th
  2. [Sections 2.2 and 3.2] The fine-tuned EDTER is trained with DiffusionEdge pseudo-labels as ground truth, and no manual validation of these labels on the 324 Mendeley leaves is reported. The same EDTER output is then used as the template in Simulations 1 and 2 and as the response in the real-data SSCCA. Simulation 2 in particular adds noise to an EDTER-extracted template and asks SSCCA to recover planted predictors; this is a positive control for SSCCA conditional on that template, not for the phenotype extraction. If pseudo-labels systematically miss tertiary/quaternary veins or hallucinate branches, the real-data associations are with edge-detector artifacts, not vascular architecture. Please provide a manual validation subset (e.g., comparison with expert annotation or an established tool such as NEFI/phenoVein) or a sensitivity analysis using alternative edge detectors to establish that the inferences are r
  3. [Section 3.2 vs Section 2.3] There is an inconsistency in pixel-intensity scales. Section 2.3 states that pixel intensities are normalized to [0,1] for SSCCA, while Simulation 2 uses threshold α=80 and truncation to [0,255] (Eq. 4). If the template edge map is in [0,1], α=80 is not meaningful; if it is in [0,255], the normalization statement is misleading. Please clarify the scale used for T_{s,t} and the threshold α.
minor comments (6)
  1. [Section 1 and 2.1.1] Typographical errors: 'inlfuence' and 'venascular' in Section 1; 'T ransformer' spacing in the Section 2.1.1 heading. A careful proofread is needed.
  2. [Section 2.2, Figure 4] The caption says 'annotated by DiffusionEdge'; since these are pseudo-labels, 'generated by DiffusionEdge' would be more accurate and avoid implying manual annotation.
  3. [Section 2.3] The package name 'mixedCCAR' is likely a typo for 'mixedCCA' (the R package for mixed-type CCA). Also, the manuscript does not state how many canonical pairs are extracted; please specify whether only the first canonical pair is used.
  4. [Section 3] The simulations set γ=10 and noise N(0,1) but do not report the signal-to-noise ratio or justify the choice of γ. A brief discussion of the signal strength relative to noise would help readers assess the simulation difficulty.
  5. [Section 4] The encoding of the 28 categorical genetic markers is not described (e.g., additive coding, number of alleles). Please specify the coding scheme used for the markers and interaction terms.
  6. [Section 5] The statement that resizing to 60×40 preserves 'complete image information' despite acknowledging smoothing is overstated. Suggest rewording to 'preserves the overall architecture' or similar.

Circularity Check

0 steps flagged

No significant circularity; the pipeline relies on external methods and the synthetic validations do not reduce to their inputs.

full rationale

The paper's claimed derivation is a pipeline: fine-tuned EDTER edge maps serve as the response phenotype, and SSCCA is applied with its published rank-based copula estimator [31]. Every load-bearing methodological input (EDTER [22], DiffusionEdge [30], BSDS500 [1], SSCCA/mixedCCAR [31]) is external and cited, not derived within this paper. The two simulations plant known coefficients beta into images generated from a template edge map (Eq. 4) and then ask SSCCA to recover them; this is an honest synthetic validation, not a prediction that reduces to its input by construction. The pseudo-label supervision from DiffusionEdge is explicitly acknowledged ('Although the edge maps generated by DiffusionEdge are not manually annotated ground-truth labels'), so fine-tuning EDTER to reproduce those labels is a stated assumption rather than a hidden circular step. The only self-citation ([9]) supports the geographic range of the Populus collection, not the statistical or biological claim, so it is not load-bearing. The paper's own limitation section concedes the small sample size and low marker resolution, and the real-data 'significant' interactions lack hypothesis-test calibration; these are statistical-evidence risks, not definitional reductions. No equation is defined in terms of the target finding, no fitted value is renamed as a prediction, and no uniqueness theorem is imported to force the variable-selection result. Therefore the derivation chain is not circular.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 0 invented entities

Central claim depends on a chain of modeling assumptions: the truncated latent Gaussian copula for zero-inflated edge maps, the bridge-function estimator, DiffusionEdge pseudo-labels as ground truth, the additive pixel-wise generative model in simulations, and the choice of 60×40 resizing. None of these is independently benchmarked against manually annotated leaf vein ground truth in this paper.

free parameters (7)
  • γ (signal scale in simulations) = 10
    Chosen by hand to make simulated genetic signal visible; not estimated from data. Appears in Eq. 4 and Simulation 1.
  • α (vein-pixel threshold in Simulation 2) = 80
    Hand-set threshold defining vein regions for predictor effects in Eq. 4; affects difficulty of simulation but not real-data estimation.
  • resized image resolution = 60×40
    Chosen after 'experimentally evaluating' 180×120, 120×80, 90×60, 60×40 in simulation studies; affects all simulation and real-data results.
  • µ (PSD projection regularization) = 0.01
    Small positive threshold added to regularized correlation matrix in Section 2.3; fixed by authors.
  • ρ (simulation predictor correlation) = 0.8
    Autoregressive correlation in simulation design; hand-set.
  • β coefficients for true predictors = 1, -2, -1, 1, 0.5
    Simulation true signals; used only to generate synthetic data, not fit to real data.
  • EDTER fine-tuning iterations = 20,000 / 40,000
    Number of additional training iterations for stages I/II; no ablation or sensitivity analysis.
axioms (6)
  • domain assumption Edge-map pixels follow a truncated latent Gaussian copula (TLNPN) model with latent nonparanormal distribution.
    Invoked in Definition in Section 2.3; if the zero-inflated, resized edge-map intensities do not satisfy this copula structure, SSCCA's estimated latent correlations are misspecified.
  • domain assumption Kendall's tau bridge function F(Σ_jk; Δ_j) correctly maps sample rank correlations to latent correlations for the mixed zero-inflated/continuous data.
    Used in Section 2.3 to estimate the latent correlation matrix; correctness depends on the copula and truncation assumptions.
  • ad hoc to paper DiffusionEdge pseudo-labels are accurate enough to serve as ground truth for fine-tuning EDTER on leaf vein images.
    Section 2.2 constructs the 'new annotated leaf image database' from DiffusionEdge outputs; no manual validation of these labels is provided.
  • ad hoc to paper A single EDTER-extracted template plus additive pixel-wise noise captures how real genetic variation affects vein architecture.
    Simulations in Section 3 generate responses by perturbing a template edge map with linear predictor effects; if real genetic effects are nonlinear or structural, simulation evidence may not transfer.
  • standard math BIC-based tuning selects the correct sparse model under the SSCCA objective.
    Uses BIC from Yoon et al. [31] Equation (3); assumes consistency of BIC for this penalized objective.
  • domain assumption EDTER architecture and pretrained weights from [22] work as specified.
    The fine-tuning strategy in Section 2.2 relies on the published EDTER architecture and pretrained weights without independent re-implementation.

pith-pipeline@v1.3.0-alltime-deepseek · 11382 in / 15223 out tokens · 149857 ms · 2026-08-01T06:26:02.029746+00:00 · methodology

0 comments
read the original abstract

Leaf veins exhibit remarkable diversity in architecture and patterning, yet existing gene--environment association studies have primarily quantified leaf venation using a small collection of low-dimensional summary traits, thereby discarding most of the structural information contained in the original images. We propose an integrated deep learning and statistical framework. The proposed framework achieves four methodological advances. First, it represents the complete leaf vascular architecture as a whole-network image phenotype. Second, it fine-tunes the deep learning-based Edge Detection with Transformers (EDTER) model to accurately extract whole-network leaf vascular architecture from RGB images by jointly learning local and global contextual features. Third, it constructs a new annotated leaf image database by integrating edge maps generated by DiffusionEdge with the Berkeley Segmentation Database (BSDS500). Fourth, it applies Semiparametric Sparse Canonical Correlation Analysis (SSCCA) to perform variable selection and model associations between repeatedly measured high-dimensional Bivariate image responses and high-dimensional predictors while simultaneously accommodating sparse, zero-inflated data represented by edge maps through a truncated latent Gaussian copula model. Two simulation studies demonstrate the performance of the proposed framework under increasing levels of complexity. Application to a real \emph{Populus} dataset identifies three significant gene--geography interactions associated with leaf vascular architecture, providing new biological insights and establishing a broadly applicable methodological framework for high-dimensional complex image phenotypes.

Figures

Figures reproduced from arXiv: 2607.22763 by Geran Zhao, Guifang Fu, Xiaotian Dai, Yangsheng Wang.

Figure 1
Figure 1. Figure 1: A schematic overview of EDTER. Elocal denotes the final edge map produced by EDTER through the fusion of global and local feature representations. 2 Methods 2.1 The Structure of EDTER Edge detection methods are commonly designed to identify object boundaries and internal edges in images. EDTER is a Transformer-based edge detector that combines global contextual information with local fine-grained cues thro… view at source ↗
Figure 2
Figure 2. Figure 2: A schematic overview of the Bi-directional Multi-Level Aggregation Decoder. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A schematic overview of the Feature Fusion Module. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Four randomly selected examples from the newly constructed leaf image dataset, together [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of the EDTER model before and after fine-tuning. The left panel illustrates [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Illustration of Simulation 1. The left panel shows the edge map of the reference leaf, [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Illustration of Simulation 2. The left panel shows the edge map of the reference leaf, [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Two randomly selected leaves from the real [PITH_FULL_IMAGE:figures/full_fig_p014_8.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

33 extracted references · 1 linked inside Pith

  1. [1]

    Contour detection and hierarchical image segmentation.IEEE transactions on pattern analysis and machine intelligence, 33(5):898–916, 2010

    Pablo Arbelaez, Michael Maire, Charless Fowlkes, and Jitendra Malik. Contour detection and hierarchical image segmentation.IEEE transactions on pattern analysis and machine intelligence, 33(5):898–916, 2010

    2010

  2. [2]

    Venation networks and the origin of the leaf economics spectrum.Ecology letters, 14(2):91–100, 2011

    Benjamin Blonder, Cyrille Violle, Lisa Patrick Bentley, and Brian J Enquist. Venation networks and the origin of the leaf economics spectrum.Ecology letters, 14(2):91–100, 2011

    2011

  3. [3]

    Viewing leaf structure and evolution from a hydraulic perspective.Functional Plant Biology, 37(6):488–498, 2010

    Tim J Brodribb, Taylor S Feild, and Lawren Sack. Viewing leaf structure and evolution from a hydraulic perspective.Functional Plant Biology, 37(6):488–498, 2010

    2010

  4. [4]

    phenovein—a tool for leaf vein segmentation and analysis.Plant physiology, 169(4):2359–2370, 2015

    Jonas Bühler, Louai Rishmawi, Daniel Pflugfelder, Gregor Huber, Hanno Scharr, Martin Hülskamp, Maarten Koornneef, Ulrich Schurr, and Siegfried Jahnke. phenovein—a tool for leaf vein segmentation and analysis.Plant physiology, 169(4):2359–2370, 2015

    2015

  5. [5]

    Quantitative analysis of venation patterns of arabidopsis leaves by supervised image analysis.The Plant Journal, 69(3): 553–563, 2012

    Stijn Dhondt, Dirk Van Haerenborgh, Caroline Van Cauwenbergh, Roeland MH Merks, Wilfried Philips, Gerrit TS Beemster, and Dirk Inzé. Quantitative analysis of venation patterns of arabidopsis leaves by supervised image analysis.The Plant Journal, 69(3): 553–563, 2012

    2012

  6. [6]

    Nefi: Network extraction from images.Scientific reports, 5(1):15669, 2015

    Michael Dirnberger, Tim Kehl, and ANEFI Neumann. Nefi: Network extraction from images.Scientific reports, 5(1):15669, 2015

    2015

  7. [7]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Pith/arXiv arXiv 2010

  8. [8]

    High dimensional semiparametric latent graphical model for mixed data.Journal of the Royal Statistical Society Series B: Statistical Methodology, 79(2):405–421, 2017

    Jianqing Fan, Han Liu, Yang Ning, and Hui Zou. High dimensional semiparametric latent graphical model for mixed data.Journal of the Royal Statistical Society Series B: Statistical Methodology, 79(2):405–421, 2017

    2017

  9. [9]

    Mapping shape quantitative trait loci using a radius-centroid-contour model.Heredity, 110(6):511–519, 2013

    G Fu, W Bo, X Pang, Z Wang, L Chen, Y Song, Z Zhang, J Li, and R Wu. Mapping shape quantitative trait loci using a radius-centroid-contour model.Heredity, 110(6):511–519, 2013

    2013

  10. [10]

    Relations between two sets of variates

    Harold Hotelling. Relations between two sets of variates. InBreakthroughs in statistics: methodology and distribution, pages 162–190. Springer, 1992

    1992

  11. [11]

    Blink: a package for the next level of genome-wide association studies with both individuals and markers in the millions.Gigascience, 8(2):giy154, 2019

    Meng Huang, Xiaolei Liu, Yao Zhou, Ryan M Summers, and Zhiwu Zhang. Blink: a package for the next level of genome-wide association studies with both individuals and markers in the millions.Gigascience, 8(2):giy154, 2019

    2019

  12. [12]

    Few-shot learning enables population-scale analysis of leaf traits in populus trichocarpa.Plant Phenomics, 5: 0072, 2023

    John Lagergren, Mirko Pavicic, Hari B Chhetri, Larry M York, Doug Hyatt, David Kainer, Erica M Rutter, Kevin Flores, Jack Bailey-Bale, Marie Klein, et al. Few-shot learning enables population-scale analysis of leaf traits in populus trichocarpa.Plant Phenomics, 5: 0072, 2023

    2023

  13. [13]

    Feature screening via distance correlation learning

    Runze Li, Wei Zhong, and Liping Zhu. Feature screening via distance correlation learning. Journal of the American Statistical Association, 107(499):1129–1139, 2012

    2012

  14. [14]

    Variation in leaf traits at different altitudes reflects the adaptive strategy of plants to environmental changes

    W Liu, L Zheng, and D Qi. Variation in leaf traits at different altitudes reflects the adaptive strategy of plants to environmental changes. ecol evol 10: 8166–8175, 2020. 15

    2020

  15. [15]

    Iterative usage of fixed and random effect models for powerful and efficient genome-wide association studies

    Xiaolei Liu, Meng Huang, Bin Fan, Edward S Buckler, and Zhiwu Zhang. Iterative usage of fixed and random effect models for powerful and efficient genome-wide association studies. PLoS genetics, 12(2):e1005767, 2016

    2016

  16. [16]

    Richer convolutional features for edge detection

    Yun Liu, Ming-Ming Cheng, Xiaowei Hu, Kai Wang, and Xiang Bai. Richer convolutional features for edge detection. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3000–3009, 2017

    2017

  17. [17]

    The fracture toughness of the leaf of the dicotyledon calophyllum inophyllum l.(guttiferae).Philosophical Transactions of the Royal Society of London

    PW Lucas, MF Choong, HTW Tan, IM Turner, and AJ Berrick. The fracture toughness of the leaf of the dicotyledon calophyllum inophyllum l.(guttiferae).Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 334(1269):95–106, 1991

    1991

  18. [18]

    Multi-model genome-wide association studies of leaf anatomical traits and vein architecture in rice.Frontiers in Plant Science, 14:1107718, 2023

    Supatthra Narawatthana, Yotwarit Phansenee, Bang-On Thammasamisorn, and Phanchita Vejchasarn. Multi-model genome-wide association studies of leaf anatomical traits and vein architecture in rice.Frontiers in Plant Science, 14:1107718, 2023

    2023

  19. [19]

    Hard and tough: the coordination between leaf mechanical resistance and drought tolerance.Flora, 288:152023, 2022

    Andrea Nardini. Hard and tough: the coordination between leaf mechanical resistance and drought tolerance.Flora, 288:152023, 2022

    2022

  20. [20]

    Global patterns of leaf mechanical properties.Ecology letters, 14(3):301–312, 2011

    Yusuke Onoda, Mark Westoby, Peter B Adler, Amy MF Choong, Fiona J Clissold, Jo- hannes HC Cornelissen, Sandra Díaz, Nathaniel J Dominy, Alison Elgart, Lucas Enrico, et al. Global patterns of leaf mechanical properties.Ecology letters, 14(3):301–312, 2011

    2011

  21. [21]

    Leaf extraction and analysis framework graphical user interface: segmenting and analyzing the structure of leaf veins and areoles.Plant Physiology, 155(1):236–245, 2011

    Charles A Price, Olga Symonova, Yuriy Mileyko, Troy Hilley, and Joshua S Weitz. Leaf extraction and analysis framework graphical user interface: segmenting and analyzing the structure of leaf veins and areoles.Plant Physiology, 155(1):236–245, 2011

    2011

  22. [22]

    Edter: Edge detection with transformer

    Mengyang Pu, Yaping Huang, Yuming Liu, Qingji Guan, and Haibin Ling. Edter: Edge detection with transformer. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1402–1412, 2022

    2022

  23. [23]

    Quantitative trait loci controlling leaf venation in arabidopsis.Plant, cell & environment, 40(8):1429–1441, 2017

    Louai Rishmawi, Jonas Bühler, Benjamin Jaegle, Martin Hülskamp, and Maarten Koornneef. Quantitative trait loci controlling leaf venation in arabidopsis.Plant, cell & environment, 40(8):1429–1441, 2017

    2017

  24. [24]

    Leaf venation: structure, function, development, evolution, ecology and applications in the past, present and future.New phytologist, 198(4): 983–1000, 2013

    Lawren Sack and Christine Scoffoni. Leaf venation: structure, function, development, evolution, ecology and applications in the past, present and future.New phytologist, 198(4): 983–1000, 2013

    2013

  25. [25]

    A database of leaf images: practice towards plant conservation with plant pathology.Mendeley Data, 2019

    Singh Chouhan Siddharth, U Singh, A Kaul, and S Jain. A database of leaf images: practice towards plant conservation with plant pathology.Mendeley Data, 2019

    2019

  26. [26]

    Plant phylogeny and growth form as drivers of the altitudinal variation in woody leaf vein traits.Frontiers in Plant Science, 10:1735, 2020

    Ruili Wang, Haoxuan Chen, Xinrui Liu, Zhibo Wang, Jingwen Wen, and Shuoxin Zhang. Plant phylogeny and growth form as drivers of the altitudinal variation in woody leaf vein traits.Frontiers in Plant Science, 10:1735, 2020

    2020

  27. [27]

    Recovering realistic texture in image super-resolution by deep spatial feature transform

    Xintao Wang, Ke Yu, Chao Dong, and Chen Change Loy. Recovering realistic texture in image super-resolution by deep spatial feature transform. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 606–615, 2018

    2018

  28. [28]

    A penalized matrix decomposi- tion, with applications to sparse principal components and canonical correlation analysis

    Daniela M Witten, Robert Tibshirani, and Trevor Hastie. A penalized matrix decomposi- tion, with applications to sparse principal components and canonical correlation analysis. Biostatistics, 10(3):515–534, 2009

    2009

  29. [29]

    Holistically-nested edge detection

    Saining Xie and Zhuowen Tu. Holistically-nested edge detection. InProceedings of the IEEE international conference on computer vision, pages 1395–1403, 2015. 16

    2015

  30. [30]

    Diffusionedge: Diffusion probabilistic model for crisp edge detection

    Yunfan Ye, Kai Xu, Yuhang Huang, Renjiao Yi, and Zhiping Cai. Diffusionedge: Diffusion probabilistic model for crisp edge detection. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 6675–6683, 2024

    2024

  31. [31]

    Sparse semiparametric canonical correlation analysis for data of mixed types.Biometrika, 107(3):609–625, 2020

    Grace Yoon, Raymond J Carroll, and Irina Gaynanova. Sparse semiparametric canonical correlation analysis for data of mixed types.Biometrika, 107(3):609–625, 2020

    2020

  32. [32]

    A unified mixed-model method for association mapping that accounts for multiple levels of relatedness.Nature genetics, 38(2):203–208, 2006

    Jianming Yu, Gael Pressoir, William H Briggs, Irie Vroh Bi, Masanori Yamasaki, John F Doebley, Michael D McMullen, Brandon S Gaut, Dahlia M Nielsen, James B Holland, et al. A unified mixed-model method for association mapping that accounts for multiple levels of relatedness.Nature genetics, 38(2):203–208, 2006

    2006

  33. [33]

    Pattern of leaf vein density and climate relationship of quercus variabilis populations remains unchanged with environmental changes.Trees, 26(2):597–607, 2012

    Yanhua Zhu, Hongzhang Kang, Qiang Xie, Zhe Wang, Shan Yin, and Chunjiang Liu. Pattern of leaf vein density and climate relationship of quercus variabilis populations remains unchanged with environmental changes.Trees, 26(2):597–607, 2012. 17

    2012