Pith. sign in

REVIEW 4 major objections 5 minor 63 references

Bringing SAM to new heights: Leveraging elevation data for tree crown segmentation from drone imagery

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper argues that learning task-specific prompts for SAM, with elevation data added, gives the best tree crown instance segmentation from drone imagery.

desk verdict Useful benchmark and a clean but modest method, undermined by an overclaim that its own Table 3 contradicts; worth reviewing after corrections. read the letter →

arxiv 2506.04970 v1 pith:OK53HR7E submitted 2025-06-05 cs.CV

Mélisande Teng , Arthur Ouaknine , Etienne Laliberté , Yoshua Bengio , David Rolnick , Hugo Larochelle This is my paper
classification cs.CV
keywords treecrownsegmentationinstanceSegmentAnythingModelDigitalSurfacedroneimagerypromptlearningforestmonitoringremotesensing
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 asks whether the Segment Anything Model (SAM), a foundation model for image segmentation, can be turned into a reliable tool for mapping individual tree crowns from high-resolution drone imagery. The authors compare SAM used out of the box, SAM given hand-crafted or predicted prompts, and custom convolutional networks such as Mask R-CNN, across plantations, a temperate forest, and a tropical forest. They find that SAM alone underperforms a custom Mask R-CNN, but that a module which learns task-specific prompts for SAM outperforms the CNN baselines. They also show that adding a Digital Surface Model (DSM) computed from the same drone images improves results, most strongly where trees are well separated. Their proposed model, BalSAM, fuses elevation information into SAM's prompt learning and achieves the best results in the plantation setting.

What carries the argument

The load-bearing mechanism is the learnable prompt generator coupled to SAM's frozen image encoder and mask decoder. BalSAM extends the RSPrompter architecture by adding a small three-layer CNN that encodes the Digital Surface Model and fuses its output with SAM's image embedding through an element-wise sum; this fused embedding is fed to the mask decoder alongside the learned prompts. The DSM encoder lets the model exploit canopy height and crown shape as dense information without retraining SAM's core.

What would settle it

An independent evaluation with fresh, unmodified annotations and prompt parameters chosen only by validation (not by dataset-specific hand tuning) in which a custom Mask R-CNN with DSM matches or beats the learned-prompt SAM methods would show the central claim does not hold.

Watch

Extended reading notes

Core claim

The central claim is that the best recipe for tree crown instance segmentation from drone imagery is not SAM used directly, nor a CNN alone, but a learned prompting module integrated with SAM, optionally fed with elevation information. Out-of-the-box SAM, even when prompted with treetop locations derived from the DSM, produces many false positives and merges overlapping crowns, and scores far below a custom Mask R-CNN. However, methods that train a small module to generate prompts for SAM (RSPrompter, and BalSAM with DSM) outperform Mask R-CNN on all three datasets in multi-class mean average precision. Adding the DSM as an extra channel or as a learned embedding reliably improves the CNN and SAM-based models, with the largest gains in the plantation dataset where trees are distinct and the ground is visible.

Load-bearing premise

The comparison assumes that manually correcting missing tree annotations and hand-picking per-dataset prompt parameters (like the minimum distance between treetop prompts) does not unfairly advantage the proposed methods.

Editorial extensions

If this is right

  • Forest monitoring pipelines can skip manual tree crown labeling at scale, using a trained prompt module around SAM instead of a fully custom CNN.
  • Publishing drone orthomosaics together with photogrammetric DSMs is a low-cost addition that improves segmentation accuracy, especially in plantations and other open canopies.
  • SAM out-of-the-box is not a drop-in solution for tree crown segmentation; it needs task-specific adaptation to compete.
  • Learned prompting helps most on rare classes, which are the ones ecologists often care about.

Reading between the lines

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

  • The DSM encoder design suggests that other cheap structural modalities (e.g., canopy height models from LiDAR, or multi-season imagery) could be fused into SAM the same way, potentially improving species classification in closed forests.
  • Because the learned prompter is task- but not site-specific, the same approach may transfer to other fine-grained instance segmentation problems with scarce labels, such as mapping shrubs, crops, or individual buildings.
  • An implicit implication of the work is that benchmark conclusions in this area are sensitive to annotation completeness; future studies should report how missing crowns are handled.
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 benchmarks methods for individual tree crown instance segmentation from high-resolution drone RGB imagery on three datasets (Quebec Plantations, SBL, BCI), comparing SAM in automatic mode, SAM prompted by DSM local maxima, Mask R-CNN and Faster R-CNN variants with and without DSM input, RSPrompter, and a new BalSAM method that adds a trainable DSM encoder to RSPrompter. All trained comparisons use three seeds and report single-class and multi-class mAP/wmAP. The abstract and Section 5.1 claim that learned SAM prompting methods outperform Mask R-CNN-based models on all datasets and that DSM information generally improves performance, but the reported tables only support these claims for some datasets and metrics.

Significance. If the central claims held, the paper would provide a useful benchmark for tree crown instance segmentation and a strong argument for parameter-efficient SAM prompting with auxiliary elevation data. The experiments are carefully run with three seeds, standard errors, and detailed hyperparameters, and the paper is among the first to benchmark SAM-based methods on these three drone datasets. However, the central generalization claim is directly contradicted by the authors' own numbers on the BCI dataset, and the DSM benefit is inconsistent across datasets. The empirical work is valuable, but the current framing overstates the conclusions.

major comments (4)
  1. [Section 5.1, Table 3] The sentence 'RSPrompter and BalSAM models outperform Mask R-CNN-based models (integrating or not the DSM) in terms of multi-class mAP and wmAP on all three datasets' is contradicted by the BCI results in Table 3. For wmAP, Mask R-CNN+DSM encoder achieves 11.86±0.27, RSPrompter achieves 11.53±0.34, and BalSAM achieves 10.42±0.27; for multi-class mAP the three methods are essentially tied (8.30, 8.44, 8.48 with overlapping standard errors). The claim should be revised to a per-dataset statement, and the BCI result should be discussed as a counterexample rather than glossed over.
  2. [Section 5.1, Tables 1-3] The claim that integrating DSM 'generally improves' model performance is not consistently supported. On SBL (Table 2), Mask R-CNN+DSM has wmAP 26.82±0.15 versus 27.27±0.18 for Mask R-CNN without DSM, and RSPrompter wmAP 29.44±0.83 versus BalSAM wmAP 29.12±0.81, so DSM integration does not help the strongest methods on this dataset. The discussion should report the effect per dataset and metric rather than as a general trend.
  3. [Section 3 and Appendix B.6] The BCI test-set construction involves manual decisions: the text says 'We manually correct for missing annotations by masking out parts of the imagery that contain unannotated trees,' and the SAM+DSM prompt method uses a hand-set local-maxima minimum distance (50 for Quebec Plantations, 20 for SBL and BCI). Because BCI is the dataset where the central comparative claim fails, and because these choices affect both labels and prompts, the paper should provide a sensitivity analysis or a clear argument that these decisions do not bias the comparison in favor of the proposed methods.
  4. [Section 5.1, Table 1] The statement that BalSAM 'shows potential over other methods' is supported on Quebec Plantations, but the advantage over RSPrompter there is moderate (wmAP 64.84±0.86 versus 62.37±1.41) and on SBL BalSAM is slightly below RSPrompter. The conclusion should be phrased as a per-dataset observation, not as a general superiority claim.
minor comments (5)
  1. [Table 1] There is a typo in the Mask R-CNN row: '81.82(±0.21' is missing a closing parenthesis.
  2. [Appendix B.4] The word 'derading' in 'leading to gridded segmentation patterns, derading the segmentations overall' should be 'degrading'.
  3. [Section 3] The phrase 'UA V Canadian (Quebec) Plantations dataset' contains an unusual spacing in 'UA V'; this should be 'UAV'.
  4. [Section 4.3] The description of the first-layer initialization for Mask R-CNN+DSM and Faster R-CNN+DSM is clear, but the sentence 'Then, we copy back the ImageNet pre-trained backbone's weights of the first layer onto the RGB channels' is slightly confusing and could be reworded to say the weights are copied from the pre-trained first layer to the RGB channels of the new four-channel first layer.
  5. [Section 4.3 and Appendix B.6] The paper states that only a 'representative sample' of code is provided in the supplementary material. For reproducibility of a benchmark paper, the full code and configuration files should be released.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the study is an empirical benchmark against independently trained baselines, with no derivation chain that reduces to its inputs.

full rationale

This paper makes no first-principles derivation and contains no fitted parameter that is later renamed as a prediction. Its conclusions are comparative claims drawn from held-out test-set metrics (mAP, wmAP, mIoU) for independently trained models: SAM out-of-the-box variants, Mask R-CNN baselines, Faster/Mask R-CNN+SAM prompting variants, RSPrompter, and the proposed BalSAM. Each method is trained on the same train/validation/test splits and evaluated on test data, so the central comparison is empirically self-contained rather than circular. The only reuse of prior work by overlapping authors is the adoption of AOI splits from Ramesh et al. [63] for the SBL dataset and a modification of that work's hierarchical loss; both are standard evaluation-protocol and loss-design choices, and neither forces the reported outcome. The paper's claim that RSPrompter and BalSAM outperform Mask R-CNN-based models 'on all three datasets' is in fact weakened by the BCI wmAP numbers in Table 3, where Mask R-CNN+DSM encoder achieves 11.86 versus 11.53 for RSPrompter and 10.42 for BalSAM, and by SBL wmAP where RSPrompter (29.44) slightly exceeds BalSAM (29.12). That is an internal-consistency/correctness concern about the strength of the stated conclusion, not a circularity concern under the definitions used here. Similarly, the manual correction of BCI annotations and per-dataset prompt parameters (e.g., local-maxima minimum distances of 50 vs 20 in Appendix B.6) could affect fairness of the comparison, but they do not constitute a self-referential derivation. No step of the paper's argument is defined in terms of its conclusion, and no load-bearing claim is justified solely by a self-citation. The appropriate finding is therefore no significant circularity, score 0.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

No derivation is attempted; the paper is an empirical benchmark. The free parameters are experimental choices, not fitted constants in a mathematical model, but they influence the reported comparisons.

free parameters (4)
  • local_maxima_min_distance = 50 (Plantations), 20 (SBL, BCI)
    Hand-set in Appendix B.6 for SAM+DSM prompts; controls which treetop candidates become prompts, directly affects the SAM+DSM baseline.
  • points_per_side = 100 and 10
    SAM automatic mode density; tested as two variants in Tables 1-3.
  • NMS thresholds = score threshold 0.5, IoU threshold 0.5
    Applied to all SAM out-of-the-box outputs as described in Appendix B.6.
  • class grouping thresholds = >20 trees per species (Plantations); taxonomic family grouping (BCI)
    Chosen to ensure classes appear in train and test splits; changes the label space and hence all reported metrics.
assumptions (3)
  • domain assumption SAM is a capable zero-shot segmentation model when appropriately prompted.
    The comparison's starting point; if SAM were not competent at generic segmentation, the negative results would be trivial. Invoked throughout Section 4.
  • domain assumption Photogrammetric DSM is an accurate proxy for canopy height structure.
    Relied on in the introduction and related work, citing Santoro et al. [57], to justify using DSM instead of LiDAR as an input modality.
  • domain assumption mAP and wmAP computed with test-class weights are valid measures of instance segmentation performance for this ecology use case.
    Used in Section 4.2; the choice of metrics affects all conclusions about which method is best.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bringing SAM to new heights: Leveraging elevation data for tree crown segmentation from drone imagery." pith.science (2026). https://pith.science/paper/OK53HR7E

@misc{pith2026250604970,
  author       = {Pith},
  title        = {Pith review of: Bringing SAM to new heights: Leveraging elevation data for tree crown segmentation from drone imagery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OK53HR7E}},
  note         = {Machine review of arXiv:2506.04970}
}
read the original abstract

Information on trees at the individual level is crucial for monitoring forest ecosystems and planning forest management. Current monitoring methods involve ground measurements, requiring extensive cost, time and labor. Advances in drone remote sensing and computer vision offer great potential for mapping individual trees from aerial imagery at broad-scale. Large pre-trained vision models, such as the Segment Anything Model (SAM), represent a particularly compelling choice given limited labeled data. In this work, we compare methods leveraging SAM for the task of automatic tree crown instance segmentation in high resolution drone imagery in three use cases: 1) boreal plantations, 2) temperate forests and 3) tropical forests. We also study the integration of elevation data into models, in the form of Digital Surface Model (DSM) information, which can readily be obtained at no additional cost from RGB drone imagery. We present BalSAM, a model leveraging SAM and DSM information, which shows potential over other methods, particularly in the context of plantations. We find that methods using SAM out-of-the-box do not outperform a custom Mask R-CNN, even with well-designed prompts. However, efficiently tuning SAM end-to-end and integrating DSM information are both promising avenues for tree crown instance segmentation models.

Figures

Figures reproduced from arXiv: 2506.04970 by the authors.

Figure 1
Figure 1. Examples of the raw image, annotations and DSM (normalized for the purpose of visualiza￾tion) on each of the datasets under consideration. We compare methods on three datasets repre￾senting different realistic application contexts: boreal plantations, temperate forests and tropical forests. As we discuss further in Section 5, each case presents different data characteristics. Planta￾tions (created for timber product… view at source ↗
Figure 2
Figure 2. Overview of our BalSAM method. We also considered using the DSM image as a dense prompt, but obtained very poor segmentation masks, as dense prompts are intended to be binary masks (see Appendix B.3). Mask R-CNN and variations We consider Mask R-CNN as a comparison, since this architecture has previously been successfully used for tree crown instance segmentation on aerial imagery [35, 36, 11, 37]. We compare Mask R… view at source ↗
Figure 3
Figure 3. Qualitative results comparing methods presented in Sec. 4 on the Quebec Plantations and BCI test sets. Samples were chosen at random in the test set. For the SAM and SAM+DSM prompts columns, colours do not correspond to particular classes since SAM does not classify instances. Colours in other columns correspond to classes and are consistent across columns. BalSAM is able to produce higher quality segmentations foll… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Distribution of trees of each of the considered families in the BCI dataset, ordered by [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Examples of SAM automatic predictions: (a) A success case. (b) SAM segments everything, [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Overview of our SAM+DSM prompts method. B.6 Implementation details We first evaluate SAM in its automatic mode on the test set tiles with a points per side (pps) value of 100 (default parameter) and 10. For SAM+DSM prompts, the local maxima in the DSM are obtained with…
Figure 7
Figure 7. Figure 7: Examples of images with overlayed local maxima prompts for the Quebec Plantations (left [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Examples of image and prediction when the DSM is fed as a mask prompt to SAM. [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Examples of ground truth, Mask R-CNN and Mask R-CNN [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Variations on BalSAM • First layer: Kernel size (2, 2), with 192 output channels, and a stride of (2, 2). • Second layer: Kernel size (8, 8), with 768 output channels, and a stride of (8, 8). • Third layer: Kernel size (1, 1), with 256 output channels, and a stride of…
Figure 11
Figure 11. Figure 11: shows the per-class mAP performance of different models on the Quebec Plantations test set [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Confusion matrix for the Mask R-CNN model predictions on the Quebec Plantations test [PITH_FULL_IMAGE:figures/full_fig_p021_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 51 canonical work pages

  1. [1]

    Managing forests for climate change mitigation

    Josep G Canadell and Michael R Raupach. Managing forests for climate change mitigation. science, 320(5882):1456–1457, 2008. 1

    work page 2008

  2. [2]

    Ten golden rules for reforestation to optimize carbon sequestration, biodiversity recovery and livelihood benefits.Global Change Biology, 27(7):1328–1348, 2021

    Alice Di Sacco, Kate A Hardwick, David Blakesley, Pedro HS Brancalion, Elinor Breman, Loic Cecilio Rebola, Susan Chomba, Kingsley Dixon, Stephen Elliott, Godfrey Ruyonga, et al. Ten golden rules for reforestation to optimize carbon sequestration, biodiversity recovery and livelihood benefits.Global Change Biology, 27(7):1328–1348, 2021. 1

    work page 2021

  3. [3]

    Formulating allometric equations for estimating biomass and carbon stock in small diameter trees.Forest Ecology and Management, 261(11):1945–1949, 2011

    Vishal Singh, Ashish Tewari, Satya PS Kushwaha, and Vinay K Dadhwal. Formulating allometric equations for estimating biomass and carbon stock in small diameter trees.Forest Ecology and Management, 261(11):1945–1949, 2011. 1

    work page 1945

  4. [4]

    Damena Edae Daba and Teshome Soromessa. The accuracy of species-specific allometric equations for estimating aboveground biomass in tropical moist montane forests: case study of Albizia grandibracteata and Trichilia dregeana.Carbon balance and management, 14:1–13,

    work page

  5. [5]

    Abu Mulatu, Mesele Negash, and Zerihun Asrat. Species-specific allometric models for reducing uncertainty in estimating above ground biomass at Moist Evergreen Afromontane forest of Ethiopia.Scientific Reports, 14(1):1147, 2024. 1

    work page 2024

  6. [6]

    Tallo: A global tree allometry and crown architecture database.Global change biology, 28(17):5254–5268, 2022

    Tommaso Jucker, Fabian Jörg Fischer, Jérôme Chave, David A Coomes, John Caspersen, Arshad Ali, Grace Jopaul Loubota Panzou, Ted R Feldpausch, Daniel Falster, Vladimir A Usoltsev, et al. Tallo: A global tree allometry and crown architecture database.Global change biology, 28(17):5254–5268, 2022. 1

    work page 2022

  7. [7]

    Afforestation, Reforestation, and Revegetation v1.0, 2023

    Verra. Afforestation, Reforestation, and Revegetation v1.0, 2023. URL https://verra.org/methodologies/ vm0047-afforestation-reforestation-and-revegetation-v1-0/ . Accessed: 2025-02-08. 1

    work page 2023

  8. [8]

    Mask R-CNN

    Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask R-CNN. InProceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. 1, 2

    work page 2017

Show all 63 references
  1. [9]

    Focal loss for dense object detection.arXiv preprint arXiv:1708.02002, 2017

    T Lin. Focal loss for dense object detection.arXiv preprint arXiv:1708.02002, 2017. 1

    2017 arXiv

  2. [10]

    Individual tree-crown detection in RGB imagery using semi-supervised deep learning neural networks

    Ben G Weinstein, Sergio Marconi, Stephanie Bohlman, Alina Zare, and Ethan White. Individual tree-crown detection in RGB imagery using semi-supervised deep learning neural networks. Remote Sensing, 11(11):1309, 2019. 1, 2

    2019

  3. [11]

    Accurate delineation of individual tree crowns in tropical forests from aerial RGB imagery using Mask R-CNN.Remote Sensing in Ecology and Conservation, 9(5):641–655, 2023

    James GC Ball, Sebastian HM Hickman, Tobias D Jackson, Xian Jing Koay, James Hirst, William Jay, Matthew Archer, Mélaine Aubry-Kientz, Grégoire Vincent, and David A Coomes. Accurate delineation of individual tree crowns in tropical forests from aerial RGB imagery using Mask R-...

    2023

  4. [12]

    ReforesTree: A dataset for estimating tropical forest carbon stock with deep learning and aerial imagery

    Gyri Reiersen, David Dao, Björn Lütjens, Konstantin Klemmer, Kenza Amara, Attila Steinegger, Ce Zhang, and Xiaoxiang Zhu. ReforesTree: A dataset for estimating tropical forest carbon stock with deep learning and aerial imagery. InProceedings of the AAAI Conference on Artificia...

    2022

  5. [13]

    Sub-continental- scale carbon stocks of individual trees in African drylands.Nature, 615(7950):80–86, 2023

    Compton Tucker, Martin Brandt, Pierre Hiernaux, Ankit Kariryaa, Kjeld Rasmussen, Jennifer Small, Christian Igel, Florian Reiner, Katherine Melocik, Jesse Meyer, et al. Sub-continental- scale carbon stocks of individual trees in African drylands.Nature, 615(7950):80–86, 2023. 2

    2023

  6. [14]

    Segment Anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment Anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4015–4026,

  7. [15]

    SAM on medical images: A comprehensive study on three prompt modes.arXiv preprint arXiv:2305.00035, 2023

    Dongjie Cheng, Ziyuan Qin, Zekun Jiang, Shaoting Zhang, Qicheng Lao, and Kang Li. SAM on medical images: A comprehensive study on three prompt modes.arXiv preprint arXiv:2305.00035, 2023. 2

    2023 arXiv

  8. [16]

    Armin Moghimi, Mario Welzel, Turgay Celik, and Torsten Schlurmann. A comparative performance analysis of popular deep learning models and Segment Anything Model (SAM) for river water segmentation in close-range remote sensing imagery.IEEE Access, 2024. 2

    2024

  9. [17]

    SAM fails to segment anything?–SAM-adapter: Adapting SAM in underperformed scenes: Camouflage, shadow, medical image segmentation, and more.arXiv preprint arXiv:2304.09148, 2023

    Tianrun Chen, Lanyun Zhu, Chaotao Ding, Runlong Cao, Yan Wang, Zejian Li, Lingyun Sun, Papa Mao, and Ying Zang. SAM fails to segment anything?–SAM-adapter: Adapting SAM in underperformed scenes: Camouflage, shadow, medical image segmentation, and more.arXiv preprint arXiv:2304...

    2023 arXiv

  10. [18]

    The Segment Anything Model (SAM) for remote sensing applications: From zero to one shot.International Journal of Applied Earth Observation and Geoinformation, 124:103540, 2023

    Lucas Prado Osco, Qiusheng Wu, Eduardo Lopes de Lemos, Wesley Nunes Gonçalves, Ana Paula Marques Ramos, Jonathan Li, and José Marcato Junior. The Segment Anything Model (SAM) for remote sensing applications: From zero to one shot.International Journal of Applied Earth Observat...

    2023

  11. [19]

    Segmate Python segmentation toolkit, 2023

    Vahid Reza Khazaie and Marshall Wang. Segmate Python segmentation toolkit, 2023. URL https://github.com/VectorInstitute/SegMate. Python package. 2, 3

    2023

  12. [20]

    RSPrompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model.IEEE Transactions on Geoscience and Remote Sensing, 2024

    Keyan Chen, Chenyang Liu, Hao Chen, Haotian Zhang, Wenyuan Li, Zhengxia Zou, and Zhenwei Shi. RSPrompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model.IEEE Transactions on Geoscience and Remote Sensing, 2024. 2, 3, 5, 19, 20

    2024

  13. [21]

    Lefebvre and E

    I. Lefebvre and E. Laliberté. UA V LiDAR, UA V Imagery, tree segmentations and ground measurements for estimating tree biomass in Canadian (Quebec) plantations, 2024. URL https://doi.org/10.20383/103.0979. 2, 3

    2024 doi

  14. [22]

    Influence of temperate forest autumn leaf phenology on segmentation of tree species from UA V imagery using deep learning.Remote Sensing of Environment, 311:114283, 2024

    Myriam Cloutier, Mickaël Germain, and Etienne Laliberté. Influence of temperate forest autumn leaf phenology on segmentation of tree species from UA V imagery using deep learning.Remote Sensing of Environment, 311:114283, 2024. 2, 4, 15

    2024

  15. [23]

    Barro Colorado Island 50-ha plot crown maps: manually segmented and instance segmented

    Vincente Vasquez, Katherine Cushman, Pablo Ramos, Cecilia Williamson, Paulino Villareal, Luisa Fernanda Gomez Correa, and Helen Muller-Landau. Barro Colorado Island 50-ha plot crown maps: manually segmented and instance segmented. (version 2). smithsonian tropical research ins...

    2023

  16. [24]

    Deep semantic segmentation of trees using multispectral images.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15:7589–7604, 2022

    Irem Ulku, Erdem Akagündüz, and Pedram Ghamisi. Deep semantic segmentation of trees using multispectral images.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15:7589–7604, 2022. 2

    2022

  17. [25]

    An unexpectedly large count of trees in the West African Sahara and Sahel.Nature, 587(7832): 78–82, 2020

    Martin Brandt, Compton J Tucker, Ankit Kariryaa, Kjeld Rasmussen, Christin Abel, Jennifer Small, Jerome Chave, Laura Vang Rasmussen, Pierre Hiernaux, Abdoul Aziz Diouf, et al. An unexpectedly large count of trees in the West African Sahara and Sahel.Nature, 587(7832): 78–82, 2020. 2

    2020

  18. [26]

    Deep Learning Based Oil Palm Tree Detection and Counting for High-Resolution Remote Sensing Images.Remote Sensing, 9(1):22, December 2016

    Weijia Li, Haohuan Fu, Le Yu, and Arthur Cracknell. Deep Learning Based Oil Palm Tree Detection and Counting for High-Resolution Remote Sensing Images.Remote Sensing, 9(1):22, December 2016. ISSN 2072-4292. doi: 10.3390/rs9010022. URL https://www.mdpi.com/ 2072-4292/9/1/22. 2 11

    2016 doi

  19. [27]

    Review on Convolutional Neural Networks (CNN) in vegetation remote sensing.ISPRS Journal of Photogrammetry and Remote Sensing, 173:24–49, March 2021

    Teja Kattenborn, Jens Leitloff, Felix Schiefer, and Stefan Hinz. Review on Convolutional Neural Networks (CNN) in vegetation remote sensing.ISPRS Journal of Photogrammetry and Remote Sensing, 173:24–49, March 2021. ISSN 09242716. doi: 10.1016/j.isprsjprs.2020.12.010. URL https...

    2021 doi

  20. [28]

    Explainable identification and mapping of trees using UA V RGB image and deep learning.Scientific Reports, 11(1):903, January 2021

    Masanori Onishi and Takeshi Ise. Explainable identification and mapping of trees using UA V RGB image and deep learning.Scientific Reports, 11(1):903, January 2021. ISSN 2045-

    2021

  21. [29]

    Comparison of classical methods and Mask R-CNN for automatic tree detection and mapping using uav imagery.Remote Sensing, 14(2):295, 2022

    Kunyong Yu, Zhenbang Hao, Christopher J Post, Elena A Mikhailova, Lili Lin, Gejin Zhao, Shangfeng Tian, and Jian Liu. Comparison of classical methods and Mask R-CNN for automatic tree detection and mapping using uav imagery.Remote Sensing, 14(2):295, 2022. 2

    2022

  22. [30]

    Sebastian Dersch, Alfred Schoettl, Peter Krzystek, and Marco Heurich. Towards complete tree crown delineation by instance segmentation with Mask R-CNN and DETR using UA V-based multispectral imagery and lidar data.ISPRS Open Journal of Photogrammetry and Remote Sensing, 8:1000...

    2023

  23. [31]

    Transferability of a Mask R-CNN model for the delineation and classification of two species of regenerating tree crowns to untrained sites.Science of Remote Sensing, 9:100109, 2024

    Andrew J Chadwick, Nicholas C Coops, Christopher W Bater, Lee A Martens, and Barry White. Transferability of a Mask R-CNN model for the delineation and classification of two species of regenerating tree crowns to untrained sites.Science of Remote Sensing, 9:100109, 2024. 2

    2024

  24. [32]

    Tree in- stance segmentation with temporal contour graph

    Adnan Firoze, Cameron Wingren, Raymond A Yeh, Bedrich Benes, and Daniel Aliaga. Tree in- stance segmentation with temporal contour graph. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2193–2202, 2023. 2

    2023

  25. [33]

    Individual tree crown delineation in high-resolution remote sensing images based on U-Net.Neural Computing and Applications, 34(24):22197–22207, 2022

    Maximilian Freudenberg, Paul Magdon, and Nils Nölke. Individual tree crown delineation in high-resolution remote sensing images based on U-Net.Neural Computing and Applications, 34(24):22197–22207, 2022. 2

    2022

  26. [34]

    U-Net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-Net: Convolutional networks for biomedical image segmentation. InMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part I...

    2015

  27. [35]

    Multi-species individual tree segmentation and identification based on improved Mask R-CNN and UA V imagery in mixed forests.Remote Sensing, 14(4):874, 2022

    Chong Zhang, Jiawei Zhou, Huiwen Wang, Tianyi Tan, Mengchen Cui, Zilu Huang, Pei Wang, and Li Zhang. Multi-species individual tree segmentation and identification based on improved Mask R-CNN and UA V imagery in mixed forests.Remote Sensing, 14(4):874, 2022. 2, 5

    2022

  28. [36]

    Yingbo Li, Guoqi Chai, Yueting Wang, Lingting Lei, and Xiaoli Zhang. ACE R-CNN: An Attention Complementary and Edge Detection-Based Instance Segmentation Algorithm for Individual Tree Species Identification Using UA V RGB Images and LiDAR Data.Remote Sensing, 14(13):3035, June...

    2022 doi

  29. [37]

    Individual Tree AGB Estimation of Malania oleifera Based on UA V-RGB Imagery and Mask R-CNN.Forests, 14(7):1493, July 2023

    Maojia Gong, Weili Kou, Ning Lu, Yue Chen, Yongke Sun, Hongyan Lai, Bangqian Chen, Juan Wang, and Chao Li. Individual Tree AGB Estimation of Malania oleifera Based on UA V-RGB Imagery and Mask R-CNN.Forests, 14(7):1493, July 2023. ISSN 1999-4907. doi: 10.3390/f14071493. URLhtt...

    2023 doi

  30. [38]

    François A. Gougeon. A Crown-Following Approach to the Automatic Delineation of Individual Tree Crowns in High Spatial Resolution Aerial Images.Canadian Journal of Remote Sensing, 21(3):274–284, August 1995. ISSN 0703-8992, 1712-7971. doi: 10.1080/07038992.1995.10874622. URL h...

    1995

  31. [39]

    Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis.Machine Vision and Applications, 11 (2):64–73, October 1998

    Tomas Brandtberg and Fredrik Walter. Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis.Machine Vision and Applications, 11 (2):64–73, October 1998. ISSN 0932-8092, 1432-1769. doi: 10.1007/s001380050091. URL http...

    1998 doi

  32. [40]

    TIDA: an algorithm for the delineation of tree crowns in high spatial resolu- tion remotely sensed imagery.Computers & Geosciences, 28(1):33–44, February 2002

    Darius S Culvenor. TIDA: an algorithm for the delineation of tree crowns in high spatial resolu- tion remotely sensed imagery.Computers & Geosciences, 28(1):33–44, February 2002. ISSN 00983004. doi: 10.1016/S0098-3004(00)00110-2. URL https://linkinghub.elsevier. com/retrieve/p...

    2002 doi

  33. [41]

    Mats Erikson. Species classification of individually segmented tree crowns in high-resolution aerial images using radiometric and morphologic image measures.Remote Sensing of Environ- ment, 91(3-4):469–477, June 2004. ISSN 00344257. doi: 10.1016/j.rse.2004.04.006. URL https://...

    2004 doi

  34. [42]

    Quackenbush

    Yinghai Ke and Lindi J. Quackenbush. A review of methods for automatic individual tree- crown detection and delineation from passive remote sensing.International Journal of Remote Sensing, 32(17):4725–4747, September 2011. ISSN 0143-1161, 1366-5901. doi: 10.1080/01431161.2010....

    2011

  35. [43]

    V o, John Brandt, Justine Spore, Sayantan Majumdar, Daniel Haziza, Janaki Vamaraju, Theo Moutakanni, Piotr Bojanowski, Tracy Johns, Brian White, Tobias Tiecke, and Camille Couprie

    Jamie Tolan, Hung-I Yang, Benjamin Nosarzewski, Guillaume Couairon, Huy V . V o, John Brandt, Justine Spore, Sayantan Majumdar, Daniel Haziza, Janaki Vamaraju, Theo Moutakanni, Piotr Bojanowski, Tracy Johns, Brian White, Tobias Tiecke, and Camille Couprie. Very high resolution...

    2024

  36. [44]

    Kelly, Martin Schwartz, Sassan Saatchi, Philippe Ciais, Sebas- tian Pokutta, Martin Brandt, and Fabian Gieseke

    Jan Pauls, Max Zimmer, Una M. Kelly, Martin Schwartz, Sassan Saatchi, Philippe Ciais, Sebas- tian Pokutta, Martin Brandt, and Fabian Gieseke. Estimating Canopy Height at Scale. In Ruslan Salakhutdinov, Zico Kolter, Katherine Heller, Adrian Weller, Nuria Oliver, Jonathan Scarle...

    2024

  37. [45]

    Wagner, Sophia Roberts, Alison L

    Fabien H. Wagner, Sophia Roberts, Alison L. Ritz, Griffin Carter, Ricardo Dalagnol, Samuel Favrichon, Mayumi C.M. Hirye, Martin Brandt, Philippe Ciais, and Sassan Saatchi. Sub- meter tree height mapping of California using aerial images and LiDAR-informed U-Net model.Remote Se...

    2024

  38. [46]

    Fabien H. Wagner, Ricardo Dalagnol, Griffin Carter, Mayumi CM Hirye, Shivraj Gill, Le Bi- enfaiteur Sagang Takougoum, Samuel Favrichon, Michael Keller, Jean PHB Ometto, Lorena Alves, Cynthia Creze, Stephanie P. George-Chacon, Shuang Li, Zhihua Liu, Adugna Mullissa, Yan Yang, E...

    2025 arXiv

  39. [47]

    Manuel Weber, Carly Beneke, and Clyde Wheeler. Unified Deep Learning Model for Global Prediction of Aboveground Biomass, Canopy Height, and Cover from High-Resolution, Multi- Sensor Satellite Imagery.Remote Sensing, 17(9):1594, April 2025. ISSN 2072-4292. doi: 10.3390/rs170915...

    2025 doi

  40. [48]

    Landau, Luke J

    Tony Chang, Kiarie Ndegwa, Andreas Gros, Vincent A. Landau, Luke J. Zachmann, Bogdan State, Mitchell A. Gritts, Colton W. Miller, Nathan E. Rutenbeck, Scott Conway, and Guy Bayes. VibrantVS: A High-Resolution Vision Transformer for Forest Canopy Height Estimation. Remote Sensi...

    2025 doi

  41. [49]

    LiDAR-based Norwegian tree species detection using deep learning, November 2023

    Martijn Vermeer, Jacob Alexander Hay, David Völgyes, Zsófia Koma, Johannes Breidenbach, and Daniele Stefano Maria Fantin. LiDAR-based Norwegian tree species detection using deep learning, November 2023. URL http://arxiv.org/abs/2311.06066. arXiv:2311.06066 [cs]. 2

    2023 arXiv

  42. [50]

    Automated forest inventory: Analysis of high-density airborne LiDAR point clouds with 3D deep learning.Remote Sensing of Environment, 305:114078, May 2024

    Binbin Xiang, Maciej Wielgosz, Theodora Kontogianni, Torben Peters, Stefano Puliti, Rasmus Astrup, and Konrad Schindler. Automated forest inventory: Analysis of high-density airborne LiDAR point clouds with 3D deep learning.Remote Sensing of Environment, 305:114078, May 2024. ...

    2024

  43. [51]

    Individual tree-crown delineation and treetop detection in high-spatial-resolution aerial imagery.Photogrammetric Engineering & Remote Sensing, 70(3):351–357, 2004

    Le Wang, Peng Gong, and Gregory S Biging. Individual tree-crown delineation and treetop detection in high-spatial-resolution aerial imagery.Photogrammetric Engineering & Remote Sensing, 70(3):351–357, 2004. 3 13

    2004

  44. [52]

    Hongyu Huang, Xu Li, and Chongcheng Chen. Individual tree crown detection and delineation from very-high-resolution UA V images based on bias field and marker-controlled watershed segmentation algorithms.IEEE Journal of selected topics in applied earth observations and remote ...

    2018

  45. [53]

    Individual tree crown segmentation of a larch plantation using airborne laser scanning data based on region growing and canopy morphology features.Remote Sensing, 12 (7):1078, 2020

    Zhenyu Ma, Yong Pang, Di Wang, Xiaojun Liang, Bowei Chen, Hao Lu, Holger Weinacker, and Barbara Koch. Individual tree crown segmentation of a larch plantation using airborne laser scanning data based on region growing and canopy morphology features.Remote Sensing, 12 (7):1078, 2020. 3

    2020

  46. [54]

    Individual tree crown delineation from high-resolution UA V images in broadleaf forest.Ecological Informatics, 61:101207, 2021

    Mojdeh Miraki, Hormoz Sohrabi, Parviz Fatehi, and Mathias Kneubuehler. Individual tree crown delineation from high-resolution UA V images in broadleaf forest.Ecological Informatics, 61:101207, 2021. 3

    2021

  47. [55]

    Zhenbang Hao, Lili Lin, Christopher J Post, Elena A Mikhailova, Minghui Li, Yan Chen, Kunyong Yu, and Jian Liu. Automated tree-crown and height detection in a young forest plantation using mask region-based convolutional neural network (Mask R-CNN).ISPRS Journal of Photogramme...

    2021

  48. [56]

    Deep learning enables image-based tree counting, crown segmentation, and height prediction at national scale

    Sizhuo Li, Martin Brandt, Rasmus Fensholt, Ankit Kariryaa, Christian Igel, Fabian Gieseke, Thomas Nord-Larsen, Stefan Oehmcke, Ask Holm Carlsen, Samuli Junttila, et al. Deep learning enables image-based tree counting, crown segmentation, and height prediction at national scale...

    2023

  49. [57]

    Santoro, Paulo G

    Giulio B. Santoro, Paulo G. Molin, José M. S. M. Viveiros, Giovanna de Andrade Ferreira, Vinicius M. Costa, Leo E. Haneda, Melodie K. S. D. Sinegalia, Laury Cullen Jr, Pedro H. S. Brancalion, Carlos A. Silva, and Danilo R. A. de Almeida. Monitoring the structure of restored fo...

    2025 doi

  50. [58]

    Felix Schiefer, Teja Kattenborn, Annett Frick, Julian Frey, Peter Schall, Barbara Koch, and Sebastian Schmidtlein. Mapping forest tree species in high resolution UA V-based RGB-imagery by means of convolutional neural networks.ISPRS Journal of Photogrammetry and Remote Sensing...

    2020 doi

  51. [60]

    Adapting Segment Anything Model to aerial land cover classification with low-rank adaptation.IEEE Geoscience and Remote Sensing Letters, 21:1–5, 2024

    Bowei Xue, Han Cheng, Qingqing Yang, Yi Wang, and Xiaoning He. Adapting Segment Anything Model to aerial land cover classification with low-rank adaptation.IEEE Geoscience and Remote Sensing Letters, 21:1–5, 2024. 3

    2024

  52. [61]

    UV-SAM: Adapting Segment Anything Model for urban village identification

    Xin Zhang, Yu Liu, Yuming Lin, Qingmin Liao, and Yong Li. UV-SAM: Adapting Segment Anything Model for urban village identification. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38(20), pages 22520–22528, 2024. 3

    2024

  53. [62]

    Leveraging Prompt-Based Segmentation Models and Large Dataset to Improve Detection of Trees.Proceedings of the Conference on Robots and Vision, may 28 2024

    Vincent Grondin, Philippe Massicotte, Mohamed Gaha, François Pomerleau, and Philippe Giguère. Leveraging Prompt-Based Segmentation Models and Large Dataset to Improve Detection of Trees.Proceedings of the Conference on Robots and Vision, may 28 2024. https://crv.pubpub.org/pub...

    2024

  54. [63]

    Acer”. We choose to keep genus level classes as separate classes instead of grouping them all into an “Other

    Venkatesh Ramesh, Arthur Ouaknine, and David Rolnick. Tree semantic segmentation from aerial image time series.arXiv preprint arXiv:2407.13102, 2024. 4, 15, 23 A Dataset In this section, we provide more details about the composition of the datasets and the splits that we used ...

    2024 arXiv

  55. [2322]

    URL https://www.nature.com/articles/ s41598-020-79653-9

    doi: 10.1038/s41598-020-79653-9. URL https://www.nature.com/articles/ s41598-020-79653-9. 2

    doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.