Pith. sign in

REVIEW 4 major objections 5 minor 80 references

Segmentation of arbitrary features in very high resolution remote sensing imagery

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read EcoMapper claims a fully automated pipeline can segment arbitrary features in very high resolution remote sensing imagery without dataset-specific tuning, matching context-specific models and adding a simple rule for choosing ground…

desk verdict A genuinely useful open-source segmentation pipeline (EcoMapper) with a real but oversold heuristic (Cording Index) that needs external validation before the scientific claims can stand. read the letter →

arxiv 2412.16046 v1 pith:PZQXAN42 submitted 2024-12-20 cs.CV

Henry Cording , Yves Plancherel , Pablo Brito-Parada This is my paper · ORCID
classification cs.CV
keywords remotesensingsemanticsegmentationveryhighresolutionimagerygroundsamplingdistanceCordingIndexEcoMapperUAVdeeplearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces EcoMapper, an automated pipeline that turns raw geospatial data into trained segmentation models with no per-dataset hand-tuning. The central claim is that this generic pipeline matches the performance of context-specific models: on a real-world drone orthomosaic it segmented chayote plantations and roads with mean intersection-over-union scores of 0.668 and 0.798, respectively. The paper further claims a simple resolution rule, the Cording Index: if a feature's smallest visible attribute measures between $f_{s1}$ and $f_{s2}$, the critical ground sampling distance at which segmentation performance begins to fall lies in $(f_{s1}/3, f_{s2}/3)$. A planning workflow built on this rule lets surveyors choose image resolution and survey extent before flying. If the claims hold, very high resolution mapping becomes reusable across features and regions rather than rebuilt for each study.

What carries the argument

The central mechanisms are the EcoMapper pipeline and the Cording Index. EcoMapper automates tiling, class-balanced sampling, model training, and post-processing via crop-based merging of overlapping predictions, so the same code applies to any labeled feature. The Cording Index, Eq. (3), is the empirical identity $GSD_F \in (f_{s1}/3, f_{s2}/3)$, where $f_{s1}$ and $f_{s2}$ are lower and upper size bounds of the feature's smallest visible attribute in overhead imagery; it converts a field measurement of a feature into a recommended ground sampling distance interval.

What would settle it

Train the same EcoMapper pipeline on an independent dataset of a feature with known SVA sizes, at GSDs just below and above the interval $(f_{s1}/3, f_{s2}/3)$. If mIoU remains flat or still high above $f_{s2}/3$, or collapses below $f_{s1}/3$, the Cording Index's predictive claim is falsified.

Watch

Extended reading notes

Core claim

The paper's core discovery is that a single, untuned training pipeline can produce segmentation models competitive with bespoke per-dataset solutions, and that the resolution at which such models fail can be predicted from feature geometry. Using the EcoMapper pipeline, the authors trained a vision-transformer segmentation model on a real-world UAV orthomosaic, segmenting chayote and roads with mean IoUs of 0.668 and 0.798 respectively. By training 99 models across twelve ground sampling distances, three degradation methods, and decreasing training-set sizes, they found that performance degrades at GSDs dependent on the size of the feature's smallest visible attribute, leading to the proposed Cording Index interval in Eq. (3). They also report that at 0.15 m/px, segmentation performance remained within 85.91% and 91.48% of the best UAV-resolution scores for chayote and roads, respectively.

Load-bearing premise

The paper's central claim rests on the assumption that the critical ground sampling distance for a feature is one third of its smallest visible attribute, with that attribute chosen reliably by a human; this factor and selection rule are empirical and were calibrated on the same performance curves used to validate them.

Editorial extensions

If this is right

  • A single EcoMapper run can be pointed at a new labeled feature without model architecture choices or hyperparameter tuning, so feature-specific studies reduce to labeling and running the pipeline.
  • For features whose smallest visible attribute is measurable, the Cording Index gives a survey-ready GSD interval, letting planners trade resolution against flight time and cost.
  • Models trained on about 30% of the original training tiles retained performance; below that, segmentation degraded sharply, indicating a floor on useful survey extent.
  • At 0.15 m/px, performance remained at 85.91% and 91.48% of the best UAV-resolution scores for chayote and roads, suggesting that very high resolution satellite products can substitute for UAV imagery for some features.

Reading between the lines

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

  • Editorial inference: if the Cording Index's factor-3 rule holds across independent features, it could be adapted to choose not only ground sampling distance but also tile size or model input resolution, since both change the effective meters-per-pixel.
  • Editorial inference: the paper's evidence comes from one geographic site and two feature classes; a decisive test would apply EcoMapper to features with SVAs spanning orders of magnitude across multiple regions and sensors.
  • Editorial inference: because the index only needs a size range of a visible sub-attribute, it may transfer to ecological surveys and disaster mapping where the target is a small object embedded in a large scene, not just continuous fields like chayote.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces EcoMapper, an open-source, automated pipeline for semantic segmentation of arbitrary features in very high resolution remote sensing imagery, with no dataset-specific model tuning. The authors evaluate EcoMapper by training Mask2Former models on a single UAV orthomosaic of a mining site in the Philippines, segmenting two features (chayote plantations and roads). Across 99 training runs they simulate varying ground sampling distance, feature size, and dataset size, and from these results they propose the Cording Index (CI, Eq. 3), which states that the critical GSD for a feature lies between one third of the lower and upper bounds of the feature's 'smallest visible attribute' size. They also propose a field-survey workflow based on these findings. The central claims are that EcoMapper achieves competitive performance without per-dataset tuning and that CI enables derivation of optimal GSD from feature size.

Significance. If fully substantiated, EcoMapper would be a useful engineering contribution: it integrates geospatial preprocessing, training, and inference, is released as open-source code, addresses spatial leakage in train/test splitting, and provides concrete lessons about GSD and dataset size. The attention to reproducibility (pinned dependencies, fixed seed, journaling, tutorials) is commendable. However, the scientific novelty of the paper rests largely on the Cording Index, and this relationship is not independently validated; it is calibrated on the same empirical curves that are later used as confirmation. With the current single-dataset, single-seed evidence, the general 'arbitrary features' claim is also broader than what the experiments demonstrate. The strengths of the software contribution do not by themselves establish the paper's central scientific claims.

major comments (4)
  1. [Section 4.2, Eq. (3)] The Cording Index is introduced as an empirical relationship, and the text explicitly states that it was 'established from empirical study of the results in section 3, and from figures reported in prior works [30, 61]'. The validation in Section 4.2.1 and Appendix M then uses the same Fig. 8 degradation curves and the same references [30, 61] that were used to set the divisor 3 and the SVA bounds; this is circular. In addition, 'critical GSD' is never given a quantitative definition (e.g., a threshold relative drop in IoU), so any monotone degradation curve can be made consistent with the interval by adjusting the SVA bounds. A predictive test on held-out features or datasets, with pre-registered SVA measurements and a defined critical-GSD criterion, is required before Eq. (3) can be regarded as a general relationship.
  2. [Section 2.5.1 and Table 1] All quantitative results are based on a single seed and single train/validation/test split per condition, with no error bars or repeated runs reported. Claims about the precise GSD at which performance 'sharply drops' (e.g., Chayote at 0.12 m/px or roads around 1 m/px in Fig. 8) are therefore fragile; a single run could shift these thresholds. Since these thresholds are used to support the Cording Index interval, reporting mean and standard deviation over at least three seeds, or an equivalent statistical analysis, is necessary for the central quantitative claims.
  3. [Section 4.1] The claim that EcoMapper performance is 'competitive with prior studies' is based on comparisons across different datasets, features, sensors, and evaluation protocols; no matched baselines are run. For example, [53–56] are cited for features comparable to Chayote and [57–60] for roads, but no scores are tabulated alongside the EcoMapper results in Table 1, and the text does not state whether those studies report class IoU, mIoU, Dice, or F1 on their own test splits. This comparison should either be removed or substantiated with a controlled benchmark on a common dataset using identical metrics and data splits.
  4. [Section 3.2 and Section 1.2] The paper's stated goal is segmentation of 'arbitrary features' without dataset-specific tuning, but the evaluation covers only two features in a single orthomosaic from one geographic region. While the two features differ in size and spectral properties, they do not demonstrate generality across sensors, landscapes, or feature types. The 'arbitrary' claim would require at least a second dataset from a different region or sensor, or a clear bounding of the claim in the abstract and introduction.
minor comments (5)
  1. [Appendix K, Table K.2] The Border IoU value for method C at GSD 0.10 m/px is listed as 212.56, which appears to be a typo for 21.56; this should be corrected.
  2. [Section 4.2, Eq. (3)] The notation 'GSDF' and 'fs1, fs2' is not introduced before Eq. (3); please define the subscripts and state the units explicitly.
  3. [Figure 9 caption] The caption contains the typo 'mechnasims'; it should read 'mechanisms'.
  4. [Appendix M] The SVA for cows is taken from external body-width tables even though the cow species in [30] is unspecified; the post-hoc matching of observed degradation with these external measurements should be acknowledged as a limitation, not presented as confirmation.
  5. [Section 2.5.1] The text says 'the same seed for all libraries was used throughout the research project,' but it does not specify which seed was used; for reproducibility, the seed value should be stated.

Circularity Check

2 steps flagged · score 4.0 of 10

EcoMapper is independently benchmarked, but the Cording Index's factor-3 interval is calibrated on the same curves and prior figures that are later presented as confirmation, so the derived GSD relationship is only partially self-contained.

  1. fitted input called prediction [Section 4.2, Eq. 3 and Section 4.2.1]
    "The index was established from empirical study of the results in section 3, and from figures reported in prior works [30, 61]. ... This interval is appropriate, as Chayote segmentation performance declined steadily beyond 0.10–0.12 m/px and improved marginally towards 0.08 m/px (c.f. Fig. 8)."

    Eq. 3 has one free parameter, the divisor 3, and the paper says it was 'established from empirical study of the results in section 3'. Section 4.2.1 then confirms the index for chayote and roads by citing the same Fig. 8 curves. For those features the 'prediction' is thus a post-hoc redescription of the observed decline: SVA bounds were chosen after inspecting the imagery or its degradation behaviour, and the divisor was set so the interval tracks the fall-off. No quantitative definition of 'critical GSD' is given, so any monotone decline can be fit by adjusting SVA bounds and divisor. The independent content rests only on Appendix M, not on chayote/road validation.

  2. fitted input called prediction [Section 4.2 and Appendix M]
    "The index was established from empirical study of the results in section 3, and from figures reported in prior works [30, 61]. ... In appendix M, additional applications of CI to other features from previous studies are made. CI provides sensible GSD intervals for these cases, even though different models and datasets were used."

    The 'additional applications' are not independent: Appendix M applies CI to vineyard LAI from [61] and cows/sheep from [30], the exact two prior works used to establish the index. Thus the supposed cross-study validation reuses the calibration sources. The features, models, and datasets differ, so this is not wholly circular, but the claim that CI was confirmed on prior work is weakened because the same figures informed the choice of the factor 3.

full rationale

The EcoMapper pipeline claim is not circular: two features were segmented by models trained without dataset-specific tuning, scored on a spatially separated held-out test set, and compared with published context-specific models; that part is an external benchmark. The circularity is confined to the Cording Index. Eq. 3's divisor 3 is not derived from theory but is stated to have been 'established from empirical study of the results in section 3, and from figures reported in prior works [30, 61]'. The in-study validation for chayote and roads points back to the same Fig. 8 curves used to set the constant, and the Appendix M 'additional applications' reuse [30] and [61], the same prior works used in calibration. This makes the central derived GSD relationship partially circular: for the in-study features the interval is a post-hoc fit, while the external cases share their calibration source. There is no load-bearing self-citation problem. Because EcoMapper's main contribution is independently validated and the CI has some external content, the overall circularity is moderate, not severe.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The main load-bearing quantities are the factor 3 in the Cording Index and the subjective SVA measure. The paper does not invent new physical entities; it relies on standard ML assumptions and the transferability of a single-site model.

free parameters (1)
  • CI divisor = 3
    In Eq. 3 the ground sampling distance is defined as feature size divided by 3. The divisor is described as derived from the observed performance curves in Fig. 8 and prior works, but no independent fitting procedure is given; it is a hand-selected constant used to match the data.
assumptions (4)
  • domain assumption Features in overhead imagery have a measurable 'smallest visible attribute' that drives detectability.
    Section 4.2 requires users to identify and measure an SVA; for irregular features this is ambiguous and no objective procedure is given.
  • domain assumption Models trained on one UAV scene of a mining site transfer to arbitrary features and regions.
    The abstract and Section 1 claim EcoMapper segments arbitrary features, but evaluation only covers chayote and roads in a single orthomosaic.
  • domain assumption Synthetic downscaling accurately emulates real changes in ground sampling distance.
    Section 2.4 uses three degradation methods; method B, for example, preserves label resolution while degrading images, which does not match real capture, and changes in tile count under method C affect training set size as a confound.
  • domain assumption Mask2Former with a Cityscapes-pretrained Swin-S backbone is a strong enough base model for arbitrary feature segmentation without tuning.
    Section 2.5.1 justifies the model choice by benchmark results, but the transferability to arbitrary RS features is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Segmentation of arbitrary features in very high resolution remote sensing imagery." pith.science (2026). https://pith.science/paper/PZQXAN42

@misc{pith2026241216046,
  author       = {Pith},
  title        = {Pith review of: Segmentation of arbitrary features in very high resolution remote sensing imagery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PZQXAN42}},
  note         = {Machine review of arXiv:2412.16046}
}
read the original abstract

Very high resolution (VHR) mapping through remote sensing (RS) imagery presents a new opportunity to inform decision-making and sustainable practices in countless domains. Efficient processing of big VHR data requires automated tools applicable to numerous geographic regions and features. Contemporary RS studies address this challenge by employing deep learning (DL) models for specific datasets or features, which limits their applicability across contexts. The present research aims to overcome this limitation by introducing EcoMapper, a scalable solution to segment arbitrary features in VHR RS imagery. EcoMapper fully automates processing of geospatial data, DL model training, and inference. Models trained with EcoMapper successfully segmented two distinct features in a real-world UAV dataset, achieving scores competitive with prior studies which employed context-specific models. To evaluate EcoMapper, many additional models were trained on permutations of principal field survey characteristics (FSCs). A relationship was discovered allowing derivation of optimal ground sampling distance from feature size, termed Cording Index (CI). A comprehensive methodology for field surveys was developed to ensure DL methods can be applied effectively to collected data. The EcoMapper code accompanying this work is available at https://github.com/hcording/ecomapper .

Figures

Figures reproduced from arXiv: 2412.16046 by the authors.

Figure 1
Figure 1. A high level overview of EcoMapper’s architecture. Data pre- and post-processing, as well as model training, evaluation, and inference are fully automated. With the availability of very high resolution (VHR) RS imagery and increasingly powerful DL models, a novel opportunity emerges to map a multitude of features, such as insects and fauna, at centimeter scale across entire landscapes. This would facilitate biodiver… view at source ↗
Figure 2
Figure 2. Overview of the labeling process. Top row: QGIS labeling. (a) Partial view of a Chayote plantation in the Sto. Niño region; (b) an overlay of the manually drawn label map for Chayote, labels were palettized for visualization. Blue indicates “Chayote”, red indicates “Border” (uncertainty). Bottom row: CVAT labeling. (c) Input image; (d) points are placed indicating the feature to label; (e) the label (cyan) is genera… view at source ↗
Figure 3
Figure 3. Methods of resolution degradation. The original image (top left) can be downsized (A), downsized and upscaled to the original tile dimensions (B), or the orignal orthomosaic can be downsized and split into tiles anew, yielding fewer tiles that cover more spatial distance and appear “zoomed out” (C). As argued in section 1, a meaningful evaluation of EcoMapper had to consider principal survey characteristics: feature… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Training, validation, and test sets after manually labeling and splitting the Sto. Niño datset. Ground truth labels are indicated, showing a favorable distribution of classes in all three sets for both features. Spacing was introduced between the train and evaluation s…
Figure 5
Figure 5. Figure 5: EcoMapper performance in different tasks. (a) Image splitting using a stride of 0.5; (b) prediction merging, input size indicates total size of all tiles; (c) GPU training with PyTorch via MMSegmentation. Code profiling was conducted to measure the efficiency of data p…
Figure 6
Figure 6. Figure 6: Comparison of ground truth with model predictions in the test set. For (a) and (b), cyan indicates Chayote, red indicates border. For (c) and (d), cyan, green and yellow visualize asphalt, dirt, and sand roads, respectively (road types for illustration, not distiguishe…
Figure 7
Figure 7. Figure 7: Closeup view of the difference between Chayote ground truth labels and predictions. Blue indicates agreement, red areas are false positives. White outline describes ground truth. (A) Largest disagreements occur in regions not containing Chayote, but contextually consid…
Figure 8
Figure 8. Figure 8: Chayote (a, c) and road (b, d) IoU model scores relative to GSD and dataset size. Slight differences in image counts in (c) and (d) are due to the difference in splits shown in [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Field survey methodology with mechnasims to accomodate for DL model requirements. Survey feasibility is informed through study extent, feature size, GSD, and the number of training samples resulting from these characteristics. Actions are proposed for cases of data sho…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 56 canonical work pages

  1. [1]

    Megan M. Coffer. ‘Balancing Privacy Rights and the Production of High-Quality Satellite Imagery’. In:Environmental Science and Technology54.11 (May 2020), pp. 6453–6455.doi: https://doi.org/10.1021/acs.est.0c02365

    work page doi:10.1021/acs.est.0c02365 2020

  2. [2]

    G. A. Fotso Kamga et al. ‘Advancements in satellite image classification: methodologies, techniques, approaches and applications’. In:International Journal of Remote Sensing42.20 (Sept. 2021), pp. 7662–7722.doi: https://doi.org/10.1080/01431161.2021.1954261

    arXiv 2021

  3. [3]

    ‘Optical satellite images services for precision agricultural use: a review’

    Asmae Dakir, Barramou Fatima Zahra and Alami Bachir Omar. ‘Optical satellite images services for precision agricultural use: a review’. In:remote sensing 4 (2021), p. 18. doi: https://doi.org/http://dx.doi.org/10.25046/aj060337

    work page doi:10.25046/aj060337 2021

  4. [4]

    McKellar, Nicholas G

    Ann E. McKellar, Nicholas G. Shephard and Dominique Chabot. ‘Dual visible-thermal camera approach facilitates drone surveys of Colonial Marshbirds’. In:Remote Sensing in Ecology and Conservation 7.2 (2020), pp. 214–226.doi: https://doi.org/10.1002/rse2.183

    work page doi:10.1002/rse2.183 2020

  5. [5]

    ‘Drones as a tool to monitor human impacts and vegetation changes in parks and protected areas’

    Francisco Javier Ancin-Murguzur et al. ‘Drones as a tool to monitor human impacts and vegetation changes in parks and protected areas’. In:Remote Sensing in Ecology and Con- servation 6.1 (Sept. 2019). Ed. by Ned Horning and Anna Carter, pp. 105–113.doi: https: //doi.org/10.1002/rse2.127

    work page doi:10.1002/rse2.127 2019

  6. [6]

    ‘Mapping coral reefs using consumer-grade drones and structure from motion photogrammetry techniques’

    Elisa Casella et al. ‘Mapping coral reefs using consumer-grade drones and structure from motion photogrammetry techniques’. In: Coral Reefs 36.1 (Nov. 2016), pp. 269–275. doi: https://doi.org/10.1007/s00338-016-1522-0

    work page doi:10.1007/s00338-016-1522-0 2016

  7. [7]

    ‘Individual tree detection and species classification of Amazo- nian palms using UAV images and deep learning’

    Matheus Pinheiro Ferreira et al. ‘Individual tree detection and species classification of Amazo- nian palms using UAV images and deep learning’. In:Forest Ecology and Management475 (Nov. 2020), p. 118397.doi: https://doi.org/10.1016/j.foreco.2020.118397

    work page arXiv 2020

  8. [8]

    ‘Convolutional Neural Networks accurately predict cover fractions of plant species and communities in Unmanned Aerial Vehicle imagery’

    Teja Kattenborn et al. ‘Convolutional Neural Networks accurately predict cover fractions of plant species and communities in Unmanned Aerial Vehicle imagery’. In:Remote Sensing in Ecology and Conservation6.4 (Feb. 2020). Ed. by Ned Horning and Nicola Clerici, pp. 472–486. doi: https://doi.org/10.1002/rse2.146

    work page doi:10.1002/rse2.146 2020

Show all 80 references
  1. [9]

    ‘Deep learning based banana plant detection and counting using high-resolution red-green-blue (RGB) images collected from unmanned aerial vehicle (UAV)’

    Bipul Neupane, Teerayut Horanont and Nguyen Duy Hung. ‘Deep learning based banana plant detection and counting using high-resolution red-green-blue (RGB) images collected from unmanned aerial vehicle (UAV)’. In:PLOS ONE 14.10 (Oct. 2019). Ed. by Krishna Prasad Vadrevu, e022390...

    2019 doi

  2. [10]

    ‘Automatic Recognition of Soybean Leaf Diseases Using UAV Images and Deep Convolutional Neural Networks’

    Everton Castelão Tetila et al. ‘Automatic Recognition of Soybean Leaf Diseases Using UAV Images and Deep Convolutional Neural Networks’. In:IEEE Geoscience and Remote Sensing Letters 17.5 (2020), pp. 903–907.doi: https://doi.org/10.1109/LGRS.2019.2932385

    2020

  3. [11]

    ‘Exploiting the centimeter resolution of UAV multispectral imagery to improve remote-sensing estimates of canopy structure and biochemistry in sugar beet crops’

    Sylvain Jay et al. ‘Exploiting the centimeter resolution of UAV multispectral imagery to improve remote-sensing estimates of canopy structure and biochemistry in sugar beet crops’. In: Remote Sensing of Environment 231 (2019), p. 110898. issn: 0034-4257. doi: https : //doi.org...

    2019 doi

  4. [12]

    ‘Deep Learning for Remote Sensing Data: A Technical Tutorial on the State of the Art’

    Liangpei Zhang, Lefei Zhang and Bo Du. ‘Deep Learning for Remote Sensing Data: A Technical Tutorial on the State of the Art’. In:IEEE Geoscience and Remote Sensing Magazine4.2 (2016), pp. 22–40.doi: https://doi.org/10.1109/MGRS.2016.2540798

    2016

  5. [13]

    Ball, Derek T

    John E. Ball, Derek T. Anderson and Chee Seng Chan Sr. ‘Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community’. In:Journal of Applied Remote Sensing11.4 (2017), p. 042609.doi: https://doi.org/10.1117/1.JRS.11. 042609

    2017 doi

  6. [14]

    ‘Deep learning in remote sensing applications: A meta-analysis and review’

    Lei Ma et al. ‘Deep learning in remote sensing applications: A meta-analysis and review’. In: ISPRS Journal of Photogrammetry and Remote Sensing152 (2019), pp. 166–177.issn: 0924-2716. doi: https://doi.org/10.1016/j.isprsjprs.2019.04.015

    2019 doi

  7. [15]

    ‘Machine Learning and Deep Learning in Remote Sensing and Urban Application: A Systematic Review and Meta-Analysis’

    Rissouni Youssef, Moumen Aniss and Chao Jamal. ‘Machine Learning and Deep Learning in Remote Sensing and Urban Application: A Systematic Review and Meta-Analysis’. In: Proceedings of the 4th Edition of International Conference on Geo-IT and Water Resources 2020, Geo-IT and Wat...

    2020

  8. [16]

    ‘Land’s complex role in climate change’

    Sr Pielke Roger A., Rezaul Mahmood and Clive McAlpine. ‘Land’s complex role in climate change’. In:Physics Today69.11 (Nov. 2016), pp. 40–46.issn: 0031-9228. doi: https://doi. org/10.1063/PT.3.3364

    2016 doi

  9. [17]

    Impacts, Risks, and Adaptation in the United States: The Fourth National Climate Assessment, Volume II

    Benjamin Sleeter et al.Chapter 5 : Land Cover and Land Use Change. Impacts, Risks, and Adaptation in the United States: The Fourth National Climate Assessment, Volume II. Tech. rep

  10. [18]

    ‘Climate Change Impacts on Plant Pathogens and Plant Diseases’

    Yigal Elad and Ilaria Pertot. ‘Climate Change Impacts on Plant Pathogens and Plant Diseases’. In: Journal of Crop Improvement28.1 (2014), pp. 99–139.doi: https://doi.org/10.1080/ 15427528.2014.865412

    2014

  11. [19]

    R. K. Mall et al. ‘Water resources and climate change: An Indian perspective’. In:Current Science 90.12 (2006), pp. 1610–1626.issn: 00113891. url: http://www.jstor.org/stable/ 24091910 (visited on 10/06/2023)

    2006

  12. [20]

    Tsouros, Stamatia Bibi and Panagiotis G

    Dimosthenis C. Tsouros, Stamatia Bibi and Panagiotis G. Sarigiannidis. ‘A Review on UAV- Based Applications for Precision Agriculture’. In:Information 10.11 (Nov. 2019), p. 349.doi: https://doi.org/10.3390/info10110349

    2019 doi

  13. [21]

    ‘A Study on the Detection of Cattle in UAV Images Using Deep Learning’

    Jayme Garcia Arnal Barbedo et al. ‘A Study on the Detection of Cattle in UAV Images Using Deep Learning’. In:Sensors 19.24 (2019). issn: 1424-8220. doi: https://doi.org/10.3390/ s19245436

    2019

  14. [22]

    Heil et al

    Benjamin J. Heil et al. ‘Reproducibility standards for machine learning in the life sciences’. In: Nature Methods18.10 (Oct. 2021), pp. 1132–1135.issn: 1548-7105. doi: https://doi.org/10. 1038/s41592-021-01256-7. 20

    2021

  15. [23]

    Machicao et al

    J. Machicao et al. ‘Mitigation Strategies to Improve Reproducibility of Poverty Estimations From Remote Sensing Images Using Deep Learning’. In:Earth and Space Science9.8 (Aug. 2022). doi: https://doi.org/10.1029/2022ea002379

    2022 doi

  16. [24]

    Saif Hassan Onim et al

    Md. Saif Hassan Onim et al. ‘LULC classification by semantic segmentation of satellite images using FastFCN’. In:CoRR abs/2011.06825 (Oct. 2020).doi: https://doi.org/10.48550/ arXiv.2011.06825

    2011 doi arXiv

  17. [25]

    Azavea/Element 84, Robert Cheetham.Raster Vision: An open source library and framework for deep learning on satellite and aerial imagery. Jan. 2023.doi: https://doi.org/10.5281/ zenodo.8018177

    2023

  18. [26]

    ‘PyTorch: An Imperative Style, High-Performance Deep Learning Library’

    Adam Paszke et al. ‘PyTorch: An Imperative Style, High-Performance Deep Learning Library’. In: CoRR abs/1912.01703 (2019). doi: https://doi.org/10.48550/arXiv.1912.01703

    2019 doi arXiv

  19. [27]

    MMSegmentation: OpenMMLab Semantic Segmentation Toolbox and Benchmark

    OpenMMLab. MMSegmentation: OpenMMLab Semantic Segmentation Toolbox and Benchmark. https://github.com/open-mmlab/mmsegmentation. 2023. (Visited on 24/07/2023)

    2023

  20. [28]

    ‘Segment Anything’

    Alexander Kirillov et al. ‘Segment Anything’. In:arXiv:2304.02643 (2023)

    2023 arXiv

  21. [29]

    ‘Estimates of maize plant density from UAV RGB images using Faster-RCNN detection model: impact of the spatial resolution’

    Kaaviya Velumani et al. ‘Estimates of maize plant density from UAV RGB images using Faster-RCNN detection model: impact of the spatial resolution’. In:Plant Phenomics 2021 (2021). doi: https://doi.org/10.34133/2021/9824843

    2021 doi

  22. [30]

    ‘Automated aerial animal detection when spatial resolution conditions are varied’

    Jasper Brown et al. ‘Automated aerial animal detection when spatial resolution conditions are varied’. In:Computers and Electronics in Agriculture193 (2022), p. 106689.issn: 0168-1699. doi: https://doi.org/10.1016/j.compag.2022.106689

    2022

  23. [31]

    ‘Analysing the Interactions Between Training Dataset Size, Label Noise and Model Performance in Remote Sensing Data’

    Jonas Gütter, Julia Niebling and Xiao Xiang Zhu. ‘Analysing the Interactions Between Training Dataset Size, Label Noise and Model Performance in Remote Sensing Data’. In:IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium. 2022, pp. 303–306.doi: https...

    2022

  24. [32]

    Calhoun et al

    Zachary D. Calhoun et al. ‘Self-Supervised Encoders Are Better Transfer Learners in Remote Sensing Applications’. In:Remote Sensing14.21 (2022). issn: 2072-4292. doi: https://doi. org/10.3390/rs14215500

    2022 doi

  25. [33]

    QGIS Association

    QGIS Development Team.QGIS Geographic Information System. QGIS Association. 2023. url: https://www.qgis.org

    2023

  26. [34]

    Version 2.2.0

    CVAT.ai Corporation.Computer Vision Annotation Tool (CVAT). Version 2.2.0. Sept. 2022. url: https://github.com/opencv/cvat

    2022

  27. [35]

    The Sphinx documentation generator

    Sphinx. The Sphinx documentation generator. May 2023.url: https://github.com/sphinx- doc/sphinx (visited on 22/07/2023)

    2023

  28. [36]

    Jazzband. 2023. url: https://github.com/jazzband/pip-tools (visited on 24/07/2023)

    2023

  29. [37]

    2023.doi: https://doi.org/https://doi.org/10.5281/ZENODO

    Even Rouault et al.GDAL. 2023.doi: https://doi.org/https://doi.org/10.5281/ZENODO. 5884351

    2023 doi

  30. [38]

    Anaconda Software Distribution

    Anaconda. Anaconda Software Distribution. https://www.anaconda.com/. July 2023. (Visited on 22/07/2023). 21

    2023

  31. [39]

    OpenMMLab. 2023. url: https://openmmlab.com/ (visited on 27/08/2023)

    2023

  32. [40]

    Seunghan Park et al.MultiEarth 2023 Deforestation Challenge – Team FOREVER. 2023. arXiv: 2306.11762 [cs.CV]

    2023 arXiv

  33. [41]

    Wenwei Zhang et al.K-Net: Towards Unified Image Segmentation. 2021. arXiv:2106.14855 [cs.CV]

    2021 arXiv

  34. [42]

    Chen Liang et al.GMMSeg: Gaussian Mixture based Generative Semantic Segmentation Models

  35. [43]

    ‘Albumentations: Fast and Flexible Image Augmentations’

    Alexander Buslaev et al. ‘Albumentations: Fast and Flexible Image Augmentations’. In:In- formation 11.2 (2020). issn: 2078-2489. doi: https://doi.org/10.3390/info11020125

    2020 doi

  36. [44]

    Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton. ‘ImageNet Classification with Deep Convolutional Neural Networks’. In:Commun. ACM 60.6 (May 2017), pp. 84–90.issn: 0001-

    2017

  37. [45]

    ‘Data augmentation for improving deep learning in image classification problem’

    Agnieszka Mikołajczyk and Michał Grochowski. ‘Data augmentation for improving deep learning in image classification problem’. In:2018 International Interdisciplinary PhD Workshop (IIPhDW). 2018, pp. 117–122.doi: https://doi.org/10.1109/IIPHDW.2018.8388338

    2018

  38. [46]

    ‘A survey on image data augmentation for deep learning’

    Connor Shorten and Taghi M Khoshgoftaar. ‘A survey on image data augmentation for deep learning’. In:Journal of big data6.1 (2019), pp. 1–48

    2019

  39. [47]

    Karasiak et al

    N. Karasiak et al. ‘Spatial dependence between training and test sets: another pitfall of classification accuracy assessment in remote sensing’. In:Machine Learning111.7 (Apr. 2021), pp. 2715–2740. doi: https://doi.org/10.1007/s10994-021-05972-1

    2021 doi

  40. [48]

    ‘Masked-attention Mask Transformer for Universal Image Segmentation’

    Bowen Cheng et al. ‘Masked-attention Mask Transformer for Universal Image Segmentation’. In: 2022. doi: https://doi.org/10.48550/arXiv.2112.01527

    2022 doi arXiv

  41. [49]

    ‘The Cityscapes Dataset for Semantic Urban Scene Understanding’

    Marius Cordts et al. ‘The Cityscapes Dataset for Semantic Urban Scene Understanding’. In: CoRR abs/1604.01685 (2016). arXiv:1604.01685. url: http://arxiv.org/abs/1604.01685

    2016 arXiv

  42. [50]

    ‘Scene Parsing through ADE20K Dataset’

    Bolei Zhou et al. ‘Scene Parsing through ADE20K Dataset’. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017, pp. 5122–5130. doi: https : //doi.org/10.1109/CVPR.2017.544

    2017 doi

  43. [51]

    Ze Liu et al.Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. 2021. arXiv: 2103.14030 [cs.CV]

    2021 arXiv

  44. [52]

    Ilya Loshchilov and Frank Hutter.Decoupled Weight Decay Regularization. 2019. arXiv:1711. 05101 [cs.LG]

    2019

  45. [53]

    ‘An automated, high-performance approach for detecting and charac- terizing broccoli based on UAV remote-sensing and transformers: A case study from Haining, China’

    Chengquan Zhou et al. ‘An automated, high-performance approach for detecting and charac- terizing broccoli based on UAV remote-sensing and transformers: A case study from Haining, China’. In:International Journal of Applied Earth Observation and Geoinformation114 (2022), p. 10...

    2022

  46. [54]

    ‘Urban Trees Mapping Using Multi-Scale Rgb Image and Deep Learning Vision Transformer-Based’

    Felipe David Georges Gomes et al. ‘Urban Trees Mapping Using Multi-Scale Rgb Image and Deep Learning Vision Transformer-Based’. In:SSRN Electronic Journal (2022). doi: https://doi.org/10.2139/ssrn.4167085. 22

    2022 doi

  47. [55]

    Gibril et al

    Mohamed Barakat A. Gibril et al. ‘Large-Scale Date Palm Tree Segmentation from Multiscale UAV-Based and Aerial Images Using Deep Vision Transformers’. In:Drones 7.2 (2023). issn: 2504-446X. doi: https://doi.org/10.3390/drones7020093

    2023 doi

  48. [56]

    ‘AMDNet: A Modern UAV RGB Remote-Sensing Tree Species Image Segmentation Model Based on Dual-Attention Residual and Structure Re-Parameterization’

    Haozhe Huang et al. ‘AMDNet: A Modern UAV RGB Remote-Sensing Tree Species Image Segmentation Model Based on Dual-Attention Residual and Structure Re-Parameterization’. In: Forests 14.3 (2023). issn: 1999-4907. doi: https://doi.org/10.3390/f14030549

    2023 doi

  49. [57]

    ‘Building Extraction With Vision Transformer’

    Libo Wang et al. ‘Building Extraction With Vision Transformer’. In:IEEE Transactions on Geoscience and Remote Sensing60 (2022), pp. 1–11.doi: https://doi.org/10.1109/TGRS. 2022.3186634

    2022

  50. [58]

    ‘DCS-TransUperNet: Road Segmentation Network Based on CSwin Trans- former with Dual Resolution’

    Zheng Zhang et al. ‘DCS-TransUperNet: Road Segmentation Network Based on CSwin Trans- former with Dual Resolution’. In:Applied Sciences12.7 (2022). issn: 2076-3417. doi: https: //doi.org/10.3390/app12073511

    2022 doi

  51. [59]

    In:Mathematics 10.24 (2022)

    Satyawant Kumar, Abhishek Kumar and Dong-Gyu Lee. In:Mathematics 10.24 (2022). issn: 2227-7390. doi: https://doi.org/10.3390/math10244735

    2022 doi

  52. [60]

    ‘Adaptive enhanced swin transformer with U-net for remote sensing image segmentation’

    Xingjian Gu et al. ‘Adaptive enhanced swin transformer with U-net for remote sensing image segmentation’. In:Computers and Electrical Engineering102 (2022), p. 108223.issn: 0045-7906. doi: https://doi.org/10.1016/j.compeleceng.2022.108223

    2022

  53. [61]

    ‘Leaf Area Index Estimation of Pergola- Trained Vineyards in Arid Regions Based on UAV RGB and Multispectral Data Using Machine Learning Methods’

    Osman Ilniyaz, Alishir Kurban and Qingyun Du. ‘Leaf Area Index Estimation of Pergola- Trained Vineyards in Arid Regions Based on UAV RGB and Multispectral Data Using Machine Learning Methods’. In:Remote Sensing14.2 (2022). issn: 2072-4292. doi: https://doi.org/ 10.3390/rs14020415

    2022 doi

  54. [62]

    ‘Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classification’

    Jayme Garcia Arnal Barbedo. ‘Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classification’. In:Computers and Electronics in Agriculture153 (2018), pp. 46–53.issn: 0168-1699. doi: https://doi.org/10.1016/j. comp...

    2018 doi

  55. [63]

    ‘On the Impact of Data Set Size in Transfer Learning Using Deep Neural Networks’

    Deepak Soekhoe, Peter van der Putten and Aske Plaat. ‘On the Impact of Data Set Size in Transfer Learning Using Deep Neural Networks’. In:Advances in Intelligent Data Analysis XV. Ed. by Henrik Boström et al. Cham: Springer International Publishing, 2016, pp. 50–60

    2016

  56. [64]

    G. Bradski. ‘The OpenCV Library’. In:Dr. Dobb’s Journal of Software Tools(2000)

    2000

  57. [65]

    Ashish Vaswani et al.Attention Is All You Need. 2023. arXiv:1706.03762 [cs.CL]

    2023 arXiv

  58. [66]

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

    Alexey Dosovitskiy et al. ‘An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale’. In:CoRR abs/2010.11929 (2020). arXiv:2010.11929. url: https://arxiv.org/ abs/2010.11929

    2020 arXiv

  59. [67]

    Robin Strudel et al.Segmenter: Transformer for Semantic Segmentation. 2021. arXiv:2105. 05633 [cs.CV]

    2021

  60. [68]

    Sixiao Zheng et al.Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers. 2021. arXiv:2012.15840 [cs.CV]. 23

    2021 arXiv

  61. [69]

    Schwing and Alexander Kirillov.Per-Pixel Classification is Not All You Need for Semantic Segmentation

    Bowen Cheng, Alexander G. Schwing and Alexander Kirillov.Per-Pixel Classification is Not All You Need for Semantic Segmentation. 2021. arXiv:2107.06278 [cs.CV]

    2021 arXiv

  62. [70]

    Vitis vinifera — European grape

    NativePlantTrust. Vitis vinifera — European grape. url: https://gobotany.nativeplanttrust. org/species/vitis/vinifera/ (visited on 24/08/2023)

    2023

  63. [71]

    url: https://plants.ces

    North Carolina Extension Gardener Plant Toolbox.Vitis vinifera. url: https://plants.ces. ncsu.edu/plants/vitis-vinifera/ (visited on 24/08/2023)

    2023

  64. [72]

    J. O. L Cerqueira et al. ‘Relationship between Zoometric Measurements in Holstein-Friesian Cow and Cubicle Size in Dairy Farms’. In:International Journal of Morphology31.1 (Mar. 2013), pp. 55–63.doi: https://doi.org/10.4067/s0717-95022013000100008

    2013 doi

  65. [73]

    Alexis Santana.Black Angus Cattle Dimensions and Drawings. Nov. 2022.url: https://www. dimensions.com/element/black-angus-cattle

    2022

  66. [74]

    Alexis Santana.Holstein Friesian Cattle Dimensions and Drawings. Nov. 2022.url: https: //www.dimensions.com/element/holstein-friesian-cattle

    2022

  67. [75]

    Dairy cow comfort: tie-stall dimensions

    Ontario. Dairy cow comfort: tie-stall dimensions. Aug. 2022.url: https://www.ontario.ca/ page/dairy-cow-comfort-tie-stall-dimensions

    2022

  68. [76]

    confidence scores

    Derick Omondi. Domestic Sheep (Ovis aries). Nov. 2022. url: https://www.dimensions. com/element/domestic-sheep-ovis-aries. 24 A Application of the Segment Anything Model (SAM) to a section of the Sto. Niño dataset (a) (b) (c) (d) Figure A.1: Application of SAM to a Chayote pla...

    2022

  69. [80]

    1280 Circular

    studied the ability of multiple machine learning models such as support vector regression and k-nearest neighbor to estimate the leaf area index (LAI), which measures the total leaf area for a given ground area, indicating the density of leaves. The study was conducted over a ...

  70. [782]

    doi: https://doi.org/10.1145/3065386

    doi

  71. [2018]

    doi: https://doi.org/10.7930/nca4.2018.ch5

    2018 doi

  72. [2022]

    arXiv: 2210.02025 [cs.CV]

    arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.