pith. sign in

arxiv: 2403.12072 · v4 · submitted 2024-02-13 · 💻 cs.CV · cs.LG

Floralens: a Deep Learning Model for the Portuguese Native Flora

Ant\'onio Filgueiras , Eduardo R. B. Marques , Lu\'is M. B. Lopes , Miguel Marques , Hugo Silva This is my paper

Pith reviewed 2026-05-24 03:32 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords deep learningplant species identificationconvolutional neural networksflora image datasetcitizen scienceimage classificationbiodiversity monitoringmachine learning application
0
0 comments X

The pith

Curated public data and standard deep learning tools produce a model for native flora identification that matches leading platforms.

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

The paper assembles a dataset of images for the native flora of a specific region from high-quality public research-grade sources and supplements it with additional observations. It then applies off-the-shelf deep convolutional neural networks via cloud services to train a model that reaches accuracy levels comparable to established citizen science systems. The resulting model supports image-based species identification and is shared through a public project along with the training dataset. A sympathetic reader would care because the work shows how accessible machine-learning methods can create functional tools for regional biodiversity when dataset construction receives careful attention.

Core claim

By anchoring a dataset in high-quality data from botanical sources and adding further sampled research-grade observations, then training via standard deep convolutional neural networks on off-the-shelf cloud services, the authors produce a model that performs accurate image-based identification of native plant species at levels comparable to state-of-the-art platforms. The model is integrated into a public website for ongoing use, and the full training dataset is released openly for others to build upon.

What carries the argument

The Floralens model, a deep convolutional neural network trained on a carefully assembled image dataset of native flora species.

If this is right

  • The model enables public access to automated identification for citizen science projects focused on plants.
  • The openly shared dataset allows direct comparison or extension by other researchers working on similar identification tasks.
  • The same combination of curated public data and standard training services can be repeated for other geographic regions or groups of species.
  • Integration into websites makes the identification capability available to non-specialists without requiring them to build models from scratch.

Where Pith is reading between the lines

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

  • Such models could support field conservation work by letting volunteers quickly flag unusual or protected species during surveys.
  • Pairing the image model with location or seasonal data might further reduce errors in real-world use.
  • The method could be tested on images taken under varying lighting or growth stages to measure how robust the accuracy remains outside the original dataset conditions.

Load-bearing premise

Public research-grade data sources supply accurately labeled images that represent the full range of the native flora without substantial label errors or collection biases.

What would settle it

A test showing that the model achieves markedly lower accuracy on an independent collection of images from the same region would indicate the central claim does not hold.

Figures

Figures reproduced from arXiv: 2403.12072 by Ant\'onio Filgueiras, Eduardo R. B. Marques, Hugo Silva, Lu\'is M. B. Lopes, Miguel Marques.

Figure 1
Figure 1. Figure 1: Detail of the FloraOn web application. 1 Introduction The improvements in processing speed, storage capacity, and imaging sensors for mobile devices paved the way for Citizen Science [1] applications and Web services that allow amateur enthusiasts to participate in science projects. One successful case study is nature observation, specifically the photographic recording of animals, plants, and fungi in the… view at source ↗
Figure 2
Figure 2. Figure 2: Dataset histograms (x-axis: number of images; y-axis: number of species). This source-based prioritization intends to define a dataset where images are less prone to identification errors. It takes into account the curation processes associated with each data source. FloraOn is curated by botanists and features high-quality images. These often feature subtle details that help secure the identification of a… view at source ↗
Figure 3
Figure 3. Figure 3: Overall, it comprises three stages: (1) preparing the data set for training; (2) [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: GAMLV interface for model training and deployment. Once the dataset is imported onto AutoML, training may proceed, requiring only the user to make high-level choices for the type of model to be generated and the maximum training time, as illustrated in Figure 4a. Since we wish to use the model as part of web or mobile applications (cf. Section 5) rather than deploying it on a Google Cloud server, we select… view at source ↗
Figure 5
Figure 5. Figure 5: Layers of the CNN model (fragment). family of functions called convolutions, which are especially suited for detecting image features (e.g., edges). In the simplest type of convolution, operating over 2D matrices, each position in the output vector called a feature map, is the dot product of a sliding window over the input matrix with a filter defined by the internal weights of the neuron. For details, see… view at source ↗
Figure 6
Figure 6. Figure 6: shows the results for precision and recall for the Floralens model applied to the test set given in [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FLTS results: Top-1, Top-5 and MRR categories of species that are endangered (part of the IUCN red list), endemic, protected, or rare. In both cases, the FloraOn site [10] is the reference for the presented subsets of species. The number of species in each subset is given between parenthesis in the figures (e.g., 42 species of ferns). The y-axis reference ticks correspond to the overall FLTS values for the… view at source ↗
Figure 8
Figure 8. Figure 8: Results according to species growth form and special categories. 4.2 PlantCLEF and Wikipedia Test Sets We considered two additional test sets: a random sample of 10,000 labeled images from the PlantCLEF’22-23 [46, 59] competition, and a sample of close to 1,500 images from Wikipedia. The PlantCLEF data we used was only a tiny sample of the entire “trusted” training set of PlantCLEF [60] that comprises appr… view at source ↗
Figure 9
Figure 9. Figure 9: Species, genus, and family identification results for all test datasets (a,b and c, respectively). results for the (comparatively few) 95 observations (9% of the cases) with 3 or more images showing even more pronounced improvements. 4.5 Classification using Geographical Location Some automated classification platforms use the geographical location associated with observations to improve the accuracy of th… view at source ↗
Figure 10
Figure 10. Figure 10: Results for classification using multiple images. System (MGRS) grid zone with a resolution of 10×10 km. For any given species, it keeps a set of MGRS grid elements for which observations have been reported. We collected this data for almost all species covered by the Floralens dataset – 1654 out of 1678 (98%). Next, we identified all images related to observations in Continental Portugal in the FLTS thro… view at source ↗
Figure 11
Figure 11. Figure 11: Results for geographic data filter. significant. The operation of the filter is also illustrated in Figure 11c in terms of the fraction of images whose classification results have been affected by the filter. The plot distinguishes the following two cases: (1) the ground truth (species) rank is improved, and; (2) the ground truth is filtered out from the results. For D = 20km we obtain the maximum fractio… view at source ↗
Figure 12
Figure 12. Figure 12: Pl@ntNet API: comparative MRR values. In [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Pl@ntNet API: comparative MRR values grouped by species growth form and special categories. April 10, 2025 21/29 [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Biolens – web application screenshots. 5.2 Biolens App We also recently developed a prototype version of a mobile application that can run on Android and iOS devices. The Android version is available for download at the Biolens website. A few screenshots of the application are shown in [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Biolens – mobile application screenshots. 5.3 Dataset and Results The dataset and the Python notebooks with all the code used for the results of Section 4 are available publicly from Zenodo [16]. The dataset contains the mapping between the image labels (ground truth), the image URLs from which they were retrieved, URLs for a site we maintain where all images are also stored, and GBIF identifiers when app… view at source ↗
read the original abstract

Machine-learning techniques, especially deep convolutional neural networks, are pivotal for image-based identification of biological species in many Citizen Science platforms. In this paper, we describe the construction of a dataset for the Portuguese native flora based on publicly available research-grade datasets, and the derivation of a high-accuracy model from it using off-the-shelf deep convolutional neural networks. We anchored the dataset in high-quality data provided by Sociedade Portuguesa de Bot\^anica and added further sampled data from research-grade datasets available from GBIF. We find that with a careful dataset design, off-the-shelf machine-learning cloud services such as Google's AutoML Vision produce accurate models, with results comparable to those of Pl@ntNet, a state-of-the-art citizen science platform. The best model we derived, dubbed Floralens, has been integrated into the public website of Project Biolens, where we gather models for other taxa as well. The dataset used to train the model is also publicly available on Zenodo.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript describes the assembly of a dataset for Portuguese native flora by combining high-quality records from the Sociedade Portuguesa de Botânica with research-grade observations from GBIF. It then trains a deep convolutional model using Google's AutoML Vision service, claims that the resulting Floralens model attains accuracy comparable to the Pl@ntNet platform, and reports that the model has been deployed on the Biolens project website with the training data released on Zenodo.

Significance. If the performance claims are substantiated, the work would supply a geographically focused identification tool that lowers the barrier for citizen-science applications in Portugal. The reliance on public datasets and an off-the-shelf cloud service is a practical strength that could be replicated for other regional floras.

major comments (3)
  1. [Abstract] Abstract: the central claim that Floralens produces 'results comparable to those of Pl@ntNet' is unsupported by any reported accuracy figures, dataset cardinality, train/validation/test split sizes, or error analysis. Without these quantities the comparability assertion cannot be assessed.
  2. [Dataset construction] Dataset construction (implied in Abstract and methods description): the premise that the SPB+GBIF research-grade subset supplies accurately labeled, representative samples of Portuguese native flora is unverified; no expert re-labeling audit, coverage statistics relative to a complete Portuguese flora checklist, or estimate of residual misidentification rates is supplied.
  3. [Model derivation] Model derivation: the statement that 'off-the-shelf machine-learning cloud services such as Google's AutoML Vision produce accurate models' is presented without any quantitative validation metrics or ablation against alternative architectures or training regimes, rendering the 'careful dataset design' claim unevaluable.
minor comments (2)
  1. The manuscript would benefit from a dedicated Results section containing standard computer-vision metrics (top-1/top-5 accuracy, per-class F1, confusion matrix) and a direct numerical comparison table with Pl@ntNet if such data exist.
  2. Clarify the precise number of taxa and images retained after any filtering steps; these cardinalities are prerequisites for interpreting any future accuracy numbers.

Simulated Author's Rebuttal

3 responses · 2 unresolved

We thank the referee for the constructive comments. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that Floralens produces 'results comparable to those of Pl@ntNet' is unsupported by any reported accuracy figures, dataset cardinality, train/validation/test split sizes, or error analysis. Without these quantities the comparability assertion cannot be assessed.

    Authors: We agree that the abstract should contain the supporting quantitative details. In the revised manuscript we will expand the abstract to report the model's top-1 accuracy, total dataset cardinality, train/validation/test split sizes, and a concise error analysis that underpins the comparability statement with Pl@ntNet. revision: yes

  2. Referee: [Dataset construction] Dataset construction (implied in Abstract and methods description): the premise that the SPB+GBIF research-grade subset supplies accurately labeled, representative samples of Portuguese native flora is unverified; no expert re-labeling audit, coverage statistics relative to a complete Portuguese flora checklist, or estimate of residual misidentification rates is supplied.

    Authors: The dataset is built exclusively from research-grade GBIF records and verified SPB observations. We will add coverage statistics against the Portuguese flora checklist. An expert re-labeling audit and residual misidentification estimate were not performed; these would require resources outside the present study scope. revision: partial

  3. Referee: [Model derivation] Model derivation: the statement that 'off-the-shelf machine-learning cloud services such as Google's AutoML Vision produce accurate models' is presented without any quantitative validation metrics or ablation against alternative architectures or training regimes, rendering the 'careful dataset design' claim unevaluable.

    Authors: The manuscript already contains the AutoML Vision validation metrics; we will present them more explicitly in the methods and results sections. Ablation experiments against other architectures lie outside the scope of demonstrating feasibility with an off-the-shelf service and are noted as future work. revision: partial

standing simulated objections not resolved
  • Expert re-labeling audit of the dataset
  • Quantitative estimate of residual misidentification rates

Circularity Check

0 steps flagged

No circularity; standard ML pipeline on external public datasets with no self-referential definitions or load-bearing self-citations.

full rationale

The paper constructs a dataset from external public sources (Sociedade Portuguesa de Botânica and GBIF research-grade records) and trains off-the-shelf models via Google's AutoML Vision. The central claim of comparability to Pl@ntNet is an empirical performance report on held-out data, not a derivation that reduces to fitted inputs or prior self-citations by construction. No equations, ansatzes, or uniqueness theorems are invoked; the derivation chain consists of standard data collection followed by supervised training and evaluation. This is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The claim rests on the representativeness and label accuracy of the assembled dataset plus the effectiveness of the off-the-shelf AutoML service; no new entities are postulated.

free parameters (1)
  • AutoML Vision training configuration
    Specific settings and hyperparameters chosen within the cloud service are not detailed and implicitly affect the final model.
axioms (1)
  • domain assumption Research-grade observations from GBIF and Sociedade Portuguesa de Botânica are accurately labeled and representative of Portuguese native flora.
    Dataset construction in the abstract relies on this premise without stated verification steps.

pith-pipeline@v0.9.0 · 5715 in / 1239 out tokens · 55131 ms · 2026-05-24T03:32:14.873438+00:00 · methodology

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

65 extracted references · 65 canonical work pages

  1. [1]

    Citizen Science: A Developing Tool for Expanding Science Knowledge and Scientific Literacy

    Bonney R, Cooper CB, Dickinson J, Kelling S, Phillips T, Rosenberg KV, et al. Citizen Science: A Developing Tool for Expanding Science Knowledge and Scientific Literacy. BioScience. 2009;59(11):977–984

    work page 2009

  2. [2]

    Connecting to nature through tech? The case of the iNaturalist app

    Altrudi S. Connecting to nature through tech? The case of the iNaturalist app. Convergence. 2021;27(1):124–141

    work page 2021

  3. [3]

    Supporting citizen scientists with automatic species identification using deep learning image recognition models

    Schermer M, Hogeweg L. Supporting citizen scientists with automatic species identification using deep learning image recognition models. Biodiversity Information Science and Standards. 2018

    work page 2018

  4. [4]

    Machine learning for image based species identification

    W¨ aldchen J, M¨ ader P. Machine learning for image based species identification. Methods in Ecology and Evolution. 2018;9(11):2216–2225

    work page 2018

  5. [5]

    The Flora Incognita app–interactive plant species identification

    M¨ ader P, Boho D, Rzanny M, Seeland M, Wittich HC, Deggelmann A, et al. The Flora Incognita app–interactive plant species identification. Methods in Ecology and Evolution. 2021;12(7):1335–1342

    work page 2021

  6. [6]

    Pl@ntNet app in the era of deep learning

    Affouard A, Go¨ eau H, Bonnet P, Lombardo JC, Joly A. Pl@ntNet app in the era of deep learning. In: International Conference on Learning Representations. Toulon, France; 2017

    work page 2017

  7. [7]

    Perspectives in machine learning for wildlife conservation

    Tuia D, Kellenberger B, Beery S, Costelloe BR, Zuffi S, Risse B, et al. Perspectives in machine learning for wildlife conservation. Nature Communications. 2022;13(1):792

    work page 2022

  8. [8]

    Applications for deep learning in ecology

    Christin S, Hervet E, Lecomte N. Applications for deep learning in ecology. Methods in Ecology and Evolution. 2019;10(10):1632–1644

    work page 2019

  9. [9]

    https://rubisco.dcc.fc.up.pt/biolens

    Biolens; 2020. https://rubisco.dcc.fc.up.pt/biolens

    work page 2020

  10. [10]

    https://www.flora-on.pt

    Flora-On: Flora de Portugal Interactiva; 2014. https://www.flora-on.pt

    work page 2014

  11. [11]

    The GBIF integrated publishing toolkit: facilitating the efficient publishing of biodiversity data on the Internet

    Robertson T, D¨ oring M, Guralnick R, Bloom D, Wieczorek J, Braak K, et al. The GBIF integrated publishing toolkit: facilitating the efficient publishing of biodiversity data on the Internet. PLOS One. 2014;9(8)

    work page 2014

  12. [12]

    Identifica¸ c˜ ao Taxon´ omica em Biologia Usando Inteligˆ encia Artificial; 2022

    Lopes LMB, Marques ERB, Mamede T, Filgueiras A, Marques M, Coutinho M. Identifica¸ c˜ ao Taxon´ omica em Biologia Usando Inteligˆ encia Artificial; 2022. Available from: https://rce.casadasciencias.org/rceapp/art/2022/050/

    work page 2022

  13. [13]

    A Portuguese Flora Identification Tool Using Deep Learning

    Marques M. A Portuguese Flora Identification Tool Using Deep Learning. Masters thesis, Faculty of Sciences, University of Porto; 2021. https://hdl.handle.net/10216/130189

    work page 2021

  14. [14]

    Floralens: a deep learning model for portuguese flora

    Filgueiras A. Floralens: a deep learning model for portuguese flora. Masters thesis, Faculty of Sciences, University of Porto; 2022. https://hdl.handle.net/10216/145701. April 10, 2025 26/29

    work page 2022

  15. [15]

    On using Deep Learning for Automatic Taxonomic Identification of Butterflies

    Mamede T. On using Deep Learning for Automatic Taxonomic Identification of Butterflies. BSC project report, Faculty of Sciences, University of Porto; 2020. https: //www.dcc.fc.up.pt/~edrdo/supervision/tmamede_lepidoptera.pdf

    work page 2020

  16. [16]

    The Floralens Dataset for Portuguese Flora; 2024

    Filgueiras A, Marques ERB, Lopes LMB, Marques M. The Floralens Dataset for Portuguese Flora; 2024. https://doi.org/10.5281/zenodo.10639701

    work page doi:10.5281/zenodo.10639701 2024

  17. [17]

    Deep-plant: Plant identification with convolutional neural networks

    Lee SH, Chan CS, Wilkin P, Remagnino P. Deep-plant: Plant identification with convolutional neural networks. In: IEEE International Conference on Image Processing; 2015. p. 452–456

    work page 2015

  18. [18]

    Large-scale plant classification with deep neural networks

    Heredia I. Large-scale plant classification with deep neural networks. In: Computing Frontiers Conference; 2017. p. 259–262

    work page 2017

  19. [19]

    Deep learning for plant identification in natural environment

    Sun Y, Liu Y, Wang G, Zhang H. Deep learning for plant identification in natural environment. Computational Intelligence and Neuroscience. 2017

    work page 2017

  20. [20]

    Plant identification: Experts vs

    Bonnet P, Go¨ eau H, Hang ST, Lasseck M,ˇSulc M, Mal´ ecot V, et al. Plant identification: Experts vs. machines in the era of deep learning: deep learning techniques challenge flora experts. Multimedia Tools and Applications for Environmental & Biodiversity Informatics. 2018; p. 131–149

    work page 2018

  21. [21]

    https://observation.org/pages/nia-explain/

    Observation.org - Explanation NIA; 2022. https://observation.org/pages/nia-explain/

    work page 2022

  22. [22]

    Visual Feature Extraction by a Multilayered Network of Analog Threshold Elements

    Fukushima K. Visual Feature Extraction by a Multilayered Network of Analog Threshold Elements. IEEE Transactions on Systems Science and Cybernetics. 1969;5(4):322–333

    work page 1969

  23. [23]

    Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position

    Fukushima K. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological Cybernetics. 1980;36(4):193–202

    work page 1980

  24. [24]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Kolesnikov A, Dosovitskiy A, Weissenborn D, Heigold G, Uszkoreit J, Beyer L, et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. In: International Conference on Learning Representations. Virtual Conference; 2021

    work page 2021

  25. [25]

    Are Transformers more robust than CNNs? In: Advances in Neural Information Processing Systems; 2021

    Bai Y, Mei J, Yuille AL, Xie C. Are Transformers more robust than CNNs? In: Advances in Neural Information Processing Systems; 2021. p. 26831–26843

    work page 2021

  26. [26]

    Pl@ntNet news – Covering all countries floras and new identification AI; 2023

    Pl@ntNet. Pl@ntNet news – Covering all countries floras and new identification AI; 2023. https://plantnet.org/en/2023/07/05/ covering-all-countries-floras-new-identification-ai/

    work page 2023

  27. [27]

    https://www.inaturalist.org/pages/computer_vision_demo

    iNaturalist - Computer Vision Explorations; 2024. https://www.inaturalist.org/pages/computer_vision_demo

    work page 2024

  28. [28]

    Pl@ntnet-my business

    Joly A, Bonnet P, Affouard A, Lombardo JC, Go¨ eau H. Pl@ntnet-my business. In: International Conference on Multimedia. ACM; 2017. p. 551–555

    work page 2017

  29. [29]

    https://observation.org/apps/obsidentify/

    Observation.org - ObsIdentify; 2022. https://observation.org/apps/obsidentify/

    work page 2022

  30. [30]

    https://multi-source.docs.biodiversityanalysis.eu/index.html

    Naturalis Biodiversity Center - Nature Identification API v2; 2022. https://multi-source.docs.biodiversityanalysis.eu/index.html. April 10, 2025 27/29

    work page 2022

  31. [31]

    https://aws.amazon.com/rekognition/

    Amazon Rekognition; 2024. https://aws.amazon.com/rekognition/

    work page 2024

  32. [32]

    https://developer.apple.com/documentation/createml

    Apple Create ML; 2024. https://developer.apple.com/documentation/createml

    work page 2024

  33. [33]

    https: //azure.microsoft.com/en-us/solutions/automated-machine-learning/

    Automated Machine Learning (AutoML) - Microsoft Azure; 2024. https: //azure.microsoft.com/en-us/solutions/automated-machine-learning/

    work page 2024

  34. [34]

    Non-coding deep learning models for tomato biotic and abiotic stress classification using microscopic images

    Choudhary M, Sentil S, Jones JB, Paret ML. Non-coding deep learning models for tomato biotic and abiotic stress classification using microscopic images. Frontiers in Plant Science. 2023;14:1292643

    work page 2023

  35. [35]

    Code-free deep learning for multi-modality medical image classification

    Korot E, Guan Z, Ferraz D, Wagner SK, Zhang G, Liu X, et al. Code-free deep learning for multi-modality medical image classification. Nature Machine Intelligence. 2021;3(4):288–298

    work page 2021

  36. [36]

    Google Auto ML versus Apple Create ML for histopathologic cancer diagnosis; which algorithms are better? arXiv preprint arXiv:190308057

    Borkowski AA, Wilson CP, Borkowski SA, Thomas LB, Deland LA, Grewe SJ, et al. Google Auto ML versus Apple Create ML for histopathologic cancer diagnosis; which algorithms are better? arXiv preprint arXiv:190308057. 2019

    work page 2019

  37. [37]

    Testing the suitability of automated machine learning, hyperspectral imaging and CIELAB color space for proximal in situ fertilization level classification

    Malounas I, Lentzou D, Xanthopoulos G, Fountas S. Testing the suitability of automated machine learning, hyperspectral imaging and CIELAB color space for proximal in situ fertilization level classification. Smart Agricultural Technology. 2024;8:100437

    work page 2024

  38. [38]

    Liang D, Xue F. Integrating automated machine learning and interpretability analysis in architecture, engineering and construction industry: A case of identifying failure modes of reinforced concrete shear walls. Computers in Industry. 2023;147:103883

    work page 2023

  39. [39]

    AI-GeoSpecies: integrate artificial intelligence into your citizen science app; 2023

    Justamante A, Joly A, Lombardo JC, Robert F, Chouet M, Li˜ n´ an S, et al.. AI-GeoSpecies: integrate artificial intelligence into your citizen science app; 2023. https://doi.org/10.5281/zenodo.7657594

    work page doi:10.5281/zenodo.7657594 2023

  40. [40]

    http://sftp.kew.org/pub/data-repositories/WCVP/

    WCVP: World Checklist of Vascular Plants; 2023. http://sftp.kew.org/pub/data-repositories/WCVP/

    work page 2023

  41. [41]

    Flowers, leaves or both? How to obtain suitable images for automated plant identification

    Rzanny M, M¨ ader P, Deggelmann A, Chen M, W¨ aldchen J. Flowers, leaves or both? How to obtain suitable images for automated plant identification. Plant Methods. 2019;15(1):1–11

    work page 2019

  42. [42]

    Florid – a Nationwide Identification Service for Plants from Photos and Habitat Information; 2024

    Brun P, de Witte L, Popp MR, Zurell D, Karger DN, Descombes P, et al.. Florid – a Nationwide Identification Service for Plants from Photos and Habitat Information; 2024. Available from: https://ssrn.com/abstract=4830448

    work page 2024

  43. [43]

    Mind your app: Could plant ID applications lead to an increase in extinction risk? Phytotaxa

    Berjano, R and Lopez-Tirado, J and Martin-Escobar, I and Martinez-Sagarra, G and Nieto-Lugilde, D and Sanchez-Romero, J and De La Estrella, M . Mind your app: Could plant ID applications lead to an increase in extinction risk? Phytotaxa. 2023;609(1):65–68

    work page 2023

  44. [44]

    The iNaturalist species classification and detection dataset

    Van Horn G, Mac Aodha O, Song Y, Cui Y, Sun C, Shepard A, et al. The iNaturalist species classification and detection dataset. In: IEEE Conference on Computer Vision and Pattern Recognition; 2018. p. 8769–8778

    work page 2018

  45. [45]

    Plant identification based on noisy web data: the amazing performance of deep learning (LifeCLEF 2017)

    Go¨ eau H, Bonnet P, Joly A. Plant identification based on noisy web data: the amazing performance of deep learning (LifeCLEF 2017). In: Conference and Labs of the Evaluation Forum; 2017. April 10, 2025 28/29

    work page 2017

  46. [46]

    Overview of PlantCLEF 2022: Image-based plant identification at global scale

    Go¨ eau H, Bonnet P, Joly A. Overview of PlantCLEF 2022: Image-based plant identification at global scale. In: Conference and Labs of the Evaluation Forum. vol. 3180; 2022. p. 1916–1928

    work page 2022

  47. [47]

    iNaturalist Research-grade Observations

    iNaturalist contributors, iNaturalist (2022). iNaturalist Research-grade Observations. iNaturalist.org.; 2023. https://doi.org/10.15468/ab3s5x

    work page doi:10.15468/ab3s5x 2022

  48. [48]

    Observation.org - Nature data from around the World

    de Vries H, Lemmens M. Observation.org - Nature data from around the World

    work page

  49. [49]

    https://doi.org/10.15468/5nilie

    work page doi:10.15468/5nilie

  50. [50]

    Pl@ntNet observations

    Affouard A, Joly A, Lombardo JC, Champ J, Goeau H, Bonnet P. Pl@ntNet observations. Version 1.2. Pl@ntNet; 2023. https://doi.org/10.15468/gtebaa

    work page doi:10.15468/gtebaa 2023

  51. [51]

    The GBIF integrated publishing toolkit: facilitating the efficient publishing of biodiversity data on the internet

    Robertson T, D¨ oring M, Guralnick R, Bloom D, Wieczorek J, Braak K, et al. The GBIF integrated publishing toolkit: facilitating the efficient publishing of biodiversity data on the internet. PLOS One. 2014;9(8)

    work page 2014

  52. [52]

    https://observation.org/pages/validation/

    Observation.org - Validation; 2023. https://observation.org/pages/validation/

    work page 2023

  53. [53]

    Research Grade

    What is the data quality assessment and how do observations qualify to become “Research Grade”?; 2023. https://www.inaturalist.org/pages/help#quality

    work page 2023

  54. [54]

    In: Google AutoML: Cloud Vision

    Bisong E. In: Google AutoML: Cloud Vision. Apress; 2019. p. 581–598

    work page 2019

  55. [55]

    https://cloud.google.com/vision/automl/docs/

    AutoML Vision Documentation; 2023. https://cloud.google.com/vision/automl/docs/

    work page 2023

  56. [56]

    https://www.tensorflow.org/lite/

    TensorFlow Lite, ML for Mobile and Edge Devices; 2023. https://www.tensorflow.org/lite/

    work page 2023

  57. [57]

    https://www.tensorflow.org/js/

    TensorFlow.js, Machine Learning for Javascript developers; 2023. https://www.tensorflow.org/js/

    work page 2023

  58. [58]

    Deep Learning

    Goodfellow I, Bengio Y, Courville A. Deep Learning. MIT Press; 2016

    work page 2016

  59. [59]

    MnasNet: Platform-Aware Neural Architecture Search for Mobile

    Tan M, Chen B, Pang R, Vasudevan V, Sandler M, Howard A, et al. MnasNet: Platform-Aware Neural Architecture Search for Mobile. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition; 2019. p. 2815–2823

    work page 2019

  60. [60]

    https://www.imageclef.org/PlantCLEF2022

    PlantCLEF2022, Image-based plant identification at global scale; 2023. https://www.imageclef.org/PlantCLEF2022

    work page 2023

  61. [61]

    https://lab.plantnet.org/LifeCLEF/PlantCLEF2022/train

    PlantCLEF’22 trusted training set; 2022. https://lab.plantnet.org/LifeCLEF/PlantCLEF2022/train

    work page 2022

  62. [62]

    https://www.mediawiki.org/wiki/Wikimedia_REST_API

    Wikimedia REST API; 2023. https://www.mediawiki.org/wiki/Wikimedia_REST_API

    work page 2023

  63. [63]

    Assessing the accuracy of free automated plant identification applications

    Hart, Adam G and Bosley, Hayley and Hooper, Chloe and Perry, Jessica and Sellors-Moore, Joel and Moore, Oliver and Goodenough, Anne E . Assessing the accuracy of free automated plant identification applications. People and Nature. 2023;5(3):929–937

    work page 2023

  64. [64]

    Pl@ntNet API for developers; 2023

    Pl@ntNet. Pl@ntNet API for developers; 2023. https://my.plantnet.org

    work page 2023

  65. [65]

    https://powo.science.kew.org/taxon/urn:lsid:ipni.org: names:192416-1

    Royal Botanical Gardens, Kew: Plants of the World Online: Chamaeleon gummifer; 2023. https://powo.science.kew.org/taxon/urn:lsid:ipni.org: names:192416-1. April 10, 2025 29/29

    work page 2023