Pith. sign in

REVIEW 2 major objections 5 minor 84 references

A single fine-tuned vision foundation model with two task heads can estimate individual tree height and species from RGB drone tiles better than separate models or classical allometric equations, at roughly half the parameter cost.

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

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

T0 review · grok-4.5

2026-07-13 19:24 UTC pith:HED45IFD

load-bearing objection Solid applied CV paper: first joint height+species UAV-RGB benchmark across three biomes, plus a clean multi-task DINOv3 design that actually delivers parameter-efficient SOTA height numbers. the 2 major comments →

arxiv 2603.23669 v2 pith:HED45IFD submitted 2026-03-24 cs.CV cs.AIcs.LG

Estimating Individual Tree Height and Species from UAV Imagery

Jannik Endres , Etienne Lalibert\'e , David Rolnick , Arthur Ouaknine This is my paper
classification cs.CV cs.AIcs.LG
keywords Tree Height EstimationSpecies IdentificationRemote SensingForest MonitoringDrone ImageryVision Foundation ModelsMulti-task LearningUAV
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Forest biomass and carbon accounting need tree-level height and species, yet field measurement does not scale and LiDAR is expensive. This paper builds BIRCH-Trees, the first public benchmark of tree-centered UAV RGB images labeled for both height and species across temperate forests, tropical forests, and boreal plantations. It shows that modern vision foundation models, once fine-tuned, beat older CNNs, Mask R-CNN height binning, and even allometric equations that are given oracle species and crown radius. Their multi-task model DINOvTree shares one foundation-model backbone between two light cross-attention heads, delivering top or near-top height accuracy and competitive species accuracy while using only 54–58 percent of the parameters of the next-best pair of models. The practical claim is that cheap single-camera drone surveys can supply the two traits biomass models need, provided trees can be cropped first.

Core claim

On BIRCH-Trees, DINOvTree—a fine-tuned DINOv3 backbone with separate cross-attention heads for height regression and species classification—achieves state-of-the-art or second-best height metrics and competitive classification while using only 54–58 percent of the parameters of the second-best approach. Learned methods substantially outperform traditional allometric equations even when those equations receive ground-truth species and masks; frozen foundation models fail, so fine-tuning is required.

What carries the argument

DINOvTree: a shared fine-tuned Vision Foundation Model (DINOv3) whose patch tokens feed two task-specific heads. Each head projects tokens with an MLP, then lets a learnable query cross-attend with 2-D positional encodings; the classification head also concatenates the backbone [CLS] token. The joint loss is Dynamic Weight Average of smooth-L1 height loss and cross-entropy species loss.

Load-bearing premise

Height ground truth is defined as the 99th-percentile LiDAR canopy-height value inside a buffered crown mask, and evaluation assumes clean tree-centered crops are already available; if that proxy is systematically wrong for thin, leaning, dead or occluded trees, both training targets and reported height scores collapse.

What would settle it

Independently re-measure height and species of a held-out subset of trees in the field, recompute all metrics against those field values instead of LiDAR-derived labels, and check whether DINOvTree still ranks first or second with the same parameter advantage.

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

If this is right

  • Individual tree height can be regressed from a single centered RGB UAV tile without LiDAR at inference time.
  • Sharing one foundation-model backbone across height and species cuts parameters nearly in half with little or no accuracy loss versus two separate models.
  • Classical allometric height equations remain inferior to modern vision models even when given oracle species and crown geometry.
  • Scaling the backbone from Base to Large improves most metrics across temperate, tropical and plantation forests.
  • BIRCH-Trees becomes a concrete out-of-distribution testbed for future foundation models on high-resolution forest imagery.

Where Pith is reading between the lines

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

  • If existing crown segmenters already reach human-level accuracy, DINOvTree can be chained into an end-to-end UAV biomass pipeline without new field campaigns for every stand.
  • The observed overestimation of short trees and underestimation of tall ones points to residual scale bias that height-stratified sampling or size-aware losses might reduce.
  • The same multi-task head pattern may transfer to other drone objects—buildings, crops—where geometric size and categorical identity must be predicted together.
  • High variance on rare species suggests semi-supervised use of the abundant unlabeled canopy tiles the authors themselves flag as future work.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper introduces BIRCH-Trees, a benchmark for joint individual-tree height regression and species classification from tree-centered RGB UAV images, built from three existing sources (Quebec Trees temperate forest, BCI tropical forest, Quebec Plantations boreal). Heights are derived from concurrent LiDAR CHMs via a buffered-mask 99th-percentile (or max) procedure (Eqs. 1–2, Sec. 3.1); species labels come from the source annotations. The authors propose DINOvTree, a multi-task model that fine-tunes a shared DINOv3 ViT backbone and attaches two cross-attention heads (height query + classification query concatenated with the [CLS] token), trained with smooth-L1 and cross-entropy under Dynamic Weight Average. Extensive multi-seed comparisons against allometric equations (oracle species/masks), Mask R-CNN height bins, CNNs, hierarchical ViTs, MambaVision, AnySat, PECore and single-task DINOv3 show DINOvTree achieving top or near-top height metrics and competitive classification while using 54–58% of the parameters of the second-best approach (Tables 1–3). Ablations cover head components, loss weighting, parameter sharing and backbone scale.

Significance. If the empirical results hold, the work supplies the first multi-biome, multi-task benchmark for individual-tree height and species from single-RGB UAV crops and a parameter-efficient multi-task VFM baseline that clearly beats classical allometrics and prior Mask R-CNN pipelines. The three-dataset design, multi-seed means±SE, and systematic ablations (heads, DWA vs UW/EW/PCGrad, sharing, Base→Large scaling) make the ranking claims reproducible and useful for forest-monitoring and biomass pipelines that already assume crown detection/segmentation. The explicit comparison to oracle allometrics and the public benchmark framing are concrete contributions beyond a pure methods paper.

major comments (2)
  1. Sec. 3.1 / Eqs. (1)–(2): Height labels are the 99th percentile (or max on plantations) of LiDAR CHM values inside a buffered crown mask. The paper itself notes that dead, thin, leaning or occluded crowns produce noisier targets (App. D.5, Fig. 17). Because every method is trained and scored against the same proxy, relative rankings remain valid, but absolute MAE/RMSE/δ1.25 should be interpreted as agreement with this proxy rather than field height. A short quantitative sensitivity study (e.g., P95 vs P99 vs max, or buffer 0.05L vs 0.1L) on at least one dataset would strengthen the claim that the reported height errors are not dominated by label construction.
  2. Tables 1–3 and Sec. 5.2: Classification F1 on rare classes (e.g., Tsuga canadensis n=9 train, Betula alleghaniensis n=11 train, several BCI families <20) exhibits high seed-to-seed variance; the paper correctly flags this in the conclusion. Macro-F1 is therefore partly driven by a handful of low-count classes. Reporting per-class F1 with confidence intervals or a frequency-stratified metric (head/mid/tail) would make the “competitive classification” claim more transparent and would clarify whether DINOvTree’s multi-task design helps or hurts the tail.
minor comments (5)
  1. Abstract and Sec. 1: “first benchmark” is accurate for the joint height+species tree-centered RGB setting, but a brief footnote acknowledging prior single-task or multi-modal individual-tree datasets would avoid over-claiming absolute novelty.
  2. Fig. 3 / Eq. (1): The buffer definition uses Euclidean distance in pixel space; a one-sentence note that L is measured in pixels (not meters) would prevent unit confusion for readers coming from forestry.
  3. Sec. 5.1: Mask R-CNN is adapted by supervising only the center tree and selecting the nearest centroid at inference; this is reasonable but should be stated more prominently so that readers do not treat the numbers as a direct re-implementation of Hao/Fu.
  4. App. C: The linear height correction and negative-height exclusion for Quebec Plantations are important; a short main-text pointer would help readers who skip the appendix.
  5. Typos / consistency: “UA V” spacing in the title, occasional “in-domain” vs “in domain”, and mixed use of “δ1.25” vs “δ 1.25” should be normalized.

Circularity Check

0 steps flagged

No significant circularity: ordinary empirical multi-task VFM evaluation on a new benchmark with external LiDAR/field labels.

full rationale

The paper constructs BIRCH-Trees by extracting tree-centered RGB tiles and height labels from existing orthomosaics + concurrent LiDAR CHMs (buffered 99th-percentile, Eqs. 1-2) plus species annotations, then trains and evaluates DINOvTree (shared DINOv3 backbone + cross-attention heads, DWA loss weighting) against CNNs/ViTs/VFMs and allometric baselines on three held-out spatial splits. All reported metrics (MAE/RMSE/MSLE/δ1.25, macro-F1/Acc) and the parameter-efficiency claim are ordinary supervised-learning outcomes measured against those external labels; nothing is algebraically forced by a fitted constant, self-definition, or uniqueness theorem. Self-citations (e.g., to Teng et al. for splits/segmentations, Lefebvre et al. for plantations CHM correction) supply input data or prior engineering, not the claimed SOTA numbers. Ablations and multi-seed tables further confirm independent content. No pattern 1-6 applies.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on standard ML practice plus domain choices for labels and crops. Free parameters are ordinary training knobs and label-construction constants; axioms are remote-sensing ground-truth and deployment assumptions; invented entities are the named benchmark and model, not physical postulates.

free parameters (6)
  • CHM buffer factor (0.1L; 0.05L on plantations)
    Hand-chosen exclusion radius for boundary pixels when extracting height labels; changes which CHM pixels define h.
  • Height aggregation percentile (P99; max on plantations)
    Chosen summary statistic of CHM inside the buffer; directly defines the regression target.
  • DWA temperature T=2 and early equal weights
    Controls multi-task loss balancing; selected by empirical comparison on Quebec Trees.
  • Learning rate schedule (peak 5e-5, backbone 0.5×, warmup/cosine)
    Fitted training hyperparameters that affect final metrics.
  • Mask R-CNN height bin width 0.2 m
    Discretization chosen empirically for the competing method.
  • Smooth L1 transition threshold of 1 m
    Fixed loss shape parameter for height regression.
axioms (5)
  • domain assumption LiDAR-derived CHM values inside a buffered crown mask are accurate enough individual-tree height labels for training and evaluation.
    Sec. 3.1 and Eqs. 1–2; paper notes DTM/CHM noise and thin-tree issues but treats labels as ground truth.
  • domain assumption Tree-centered 512×512 crops from high-quality crown segmentations are a realistic evaluation setting for field deployment.
    Sec. 3.1 argues human-level segmenters exist; joint detection+traits is out of scope.
  • domain assumption Web-pretrained DINOv3 features, after full fine-tuning, transfer to high-resolution UAV forest imagery for both geometric height and fine-grained species.
    Sec. 4.1 and Tab. 1; frozen backbones fail, so the claim depends on fine-tuning working.
  • domain assumption Spatial train/val/test splits from Teng et al. sufficiently control spatial autocorrelation.
    Sec. 3.2; standard but unproven for these height labels.
  • standard math Standard deep learning optimization (AdamW, cross-entropy, smooth L1) yields valid comparative rankings of architectures.
    Implicit throughout Sec. 5; usual empirical ML practice.
invented entities (2)
  • BIRCH-Trees benchmark no independent evidence
    purpose: Provide the first shared evaluation set for joint individual tree height regression and species classification from tree-centered UAV RGB.
    Constructed by the authors from three existing datasets via a new extraction pipeline; no independent prior existence under this name.
  • DINOvTree multi-task architecture no independent evidence
    purpose: Share one VFM backbone with two cross-attention heads for simultaneous height and species prediction at lower parameter cost.
    Named model introduced in Sec. 4; composition of known components (DINOv3, cross-attention, DWA).

pith-pipeline@v1.1.0-grok45 · 34343 in / 3370 out tokens · 43511 ms · 2026-07-13T19:24:48.174782+00:00 · methodology

0 comments
read the original abstract

Accurate estimation of forest biomass, a major carbon sink, relies heavily on tree-level traits such as height and species. Unoccupied Aerial Vehicles (UAVs) capturing high-resolution imagery from a single RGB camera offer a cost-effective and scalable approach for mapping and measuring individual trees. We introduce BIRCH-Trees, the first benchmark for individual tree height and species estimation from tree-centered UAV images, spanning three datasets: temperate forests, tropical forests, and boreal plantations. We also present DINOvTree, a unified approach using a Vision Foundation Model (VFM) backbone with task-specific heads for simultaneous height and species prediction. Through extensive evaluations on BIRCH-Trees, we compare DINOvTree against commonly used vision methods, including VFMs, as well as biological allometric equations. We find that DINOvTree achieves top overall results with accurate height predictions and competitive classification accuracy while using only 54% to 58% of the parameters of the second-best approach.

Figures

Figures reproduced from arXiv: 2603.23669 by Arthur Ouaknine, David Rolnick, Etienne Lalibert\'e, Jannik Endres.

Figure 1
Figure 1. Figure 1: Overview of our benchmark BIRCH-Trees and method DINOvTree. BIRCH-Trees is a benchmark for joint individual tree height esti￾mation and species identification from UAV images. DINOvTree includes a VFM backbone with two task-specific heads. Thus, here we estimate height from RGB drone imagery, a cost-effective alternative. Recent work has successfully addressed gen￾eralized tree crown detection [5] and segm… view at source ↗
Figure 2
Figure 2. Figure 2: Examples from the BIRCH￾Trees benchmark. It consists of tree￾centered RGB UAV images from three for￾est types with height and class label. We introduce BIRCH-Trees, the first benchmark for individual tree species identification and height estimation from tree-centered RGB images. To construct BIRCH-Trees, we developed a data extraction pipeline (Sec. 3.1) and applied it to three diverse datasets: Quebec Tr… view at source ↗
Figure 3
Figure 3. Figure 3: Segmentation boundary buffering. The pixel bounded by the pink and purple contours correspond to the original segmentation S and the buffered one Sbuf. The source data comprises orthomosaics of multiple trees, each with individual segmentations and class labels. To construct our benchmark, we use manu￾ally annotated segmentations of individual tree crowns as a preprocessing step. While manual segmentations… view at source ↗
Figure 4
Figure 4. Figure 4: Class histograms of the Quebec Trees, BCI and Quebec Plantation datasets. Distributions per split are in App. A.2. 0 5 10 15 20 25 30 Height (m) 0 250 500 750 1000 1250 1500 1750 Number of Samples (a) Quebec Trees 10 20 30 40 50 Height (m) 0 20 40 60 80 100 120 Number of Samples (b) BCI 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 Height (m) 0 1000 2000 3000 4000 Number of Samples (c) Quebec Plantations [PITH_FULL… view at source ↗
Figure 5
Figure 5. Figure 5: Height histograms of the Quebec Trees, BCI and Quebec Plantation datasets with 1 m intervals. Distributions per split are in App. A.3. Quebec Plantations dataset. The Quebec Plantations dataset [39] comprises 17.7K images (train: 11.1K, val: 4.0K, test: 2.6K) from boreal plantations in Quebec, Canada, at 0.5 cm/pixel spatial resolution, making it comparable in size to Quebec Trees. Plantations contain youn… view at source ↗
Figure 6
Figure 6. Figure 6: Overview of DINOvTree. A shared VFM (blue) extracts features from an RGB image. In the height estimation head (orange), a learnable height query cross￾attends to adapted patch tokens to predict hˆ. In the classification head (yellow), a learnable classification query cross-attends to adapted patch tokens to obtain a classi￾fication token. We then concatenate it with the VFM [CLS] token to predict pˆ. The t… view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of true vs. pre￾dicted height using DINOvTree-B on the Quebec Trees test split. We select the seed with highest δ1.25 and visualize 5 samples per class. The dashed diagonal line represents ideal predictions (y = x). Tab. 6 presents the mean signed differ￾ence (MSD = hˆ − h) for DINOvTree￾B across BIRCH-Trees. Height pre￾diction bias is dataset-dependent and strongly correlates with the differ￾en… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative examples from the Quebec Trees dataset. [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative examples from the BCI dataset. [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative examples from the Quebec Plantations dataset. [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Class distributions across splits for the Quebec Trees dataset. [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Class distributions across splits for the BCI dataset. [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Class distributions across splits for the Quebec Plantations dataset. [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Height distributions across splits for the Quebec Trees dataset. [PITH_FULL_IMAGE:figures/full_fig_p029_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Height distributions across splits for the BCI dataset. [PITH_FULL_IMAGE:figures/full_fig_p029_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Height distributions across splits for the Quebec Plantations [PITH_FULL_IMAGE:figures/full_fig_p029_16.png] view at source ↗
Figure 18
Figure 18. Figure 18: Confusion matrix of DINOvTree-B on Quebec Trees. Predicted and true labels are aggregated across 5 seeds to ensure reliability. All entries are row-normalized (per true class) and displayed as percentages. Cells with value rounding to 0.0% are left blank for readability. Trees within the ‘Acer’ genus (Acer rubrum, Acer saccharum, and Acer pen￾sylvanicum) tend to be recognized more reliably than species fr… view at source ↗
Figure 19
Figure 19. Figure 19: Comparison of true vs. predicted height using DINOvTree-B on the BCI and Quebec Plantations test splits. We select the random seed yielding the highest δ1.25 and visualize up to 5 samples per class. The dashed diagonal line represents the ideal prediction (y = x). 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Classification F1-Score 0.70 0.75 0.80 0.85 0.90 0.95 1.00 H eig ht 1.25 Class Malv. Fab. Bignoni. Rubi. La… view at source ↗
Figure 21
Figure 21. Figure 21: Confusion matrix of DINOvTree-B on BCI. Predicted and true labels are aggregated across 5 seeds to ensure reliability. All entries are row-normalized (per true class) and displayed as percentages. Cells with value rounding to 0% are left blank for readability. rectly classified as ‘Fab.’ trees, which is the second most frequent class in the training set [PITH_FULL_IMAGE:figures/full_fig_p036_21.png] view at source ↗
Figure 23
Figure 23. Figure 23: Confusion matrix of DINOvTree-B on Quebec Plan￾tations. Predicted and true labels are aggregated across 5 seeds to ensure relia￾bility. All entries are row-normalized (per true class) and displayed as percentages. Cells with value rounding to 0.00% are left blank for readability. Quebec Plantations dataset. Consistent with the negligibly small average MSD for Quebec Plantations (Tab. 6), Fig. 19b displays… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

84 extracted references · 5 canonical work pages

  1. [1]

    In: CVPR

    Astruc, G., Gonthier, N., Mallet, C., Landrieu, L.: AnySat: One earth observation model for many resolutions, scales, and modalities. In: CVPR. pp. 19530–19540 (2025)

    2025

  2. [2]

    arXiv:1607.06450 [stat.ML] (2016)

    Ba, J.L., Kiros, J.R., Hinton, G.E.: Layer normalization. arXiv:1607.06450 [stat.ML] (2016)

    Pith/arXiv arXiv 2016

  3. [3]

    Bastani, F., Wolters, P., Gupta, R., Ferdinando, J., Kembhavi, A.: Satlaspretrain: Alarge-scaledatasetforremotesensingimageunderstanding.In:ICCV.pp.16772– 16782 (2023)

    2023

  4. [4]

    Baudchon, H.: GeoDataset (2024),https : / / hugobaudchon . github . io / geodataset/

    2024

  5. [5]

    In: ICLR (2026)

    Baudchon, H., Ouaknine, A., Weiss, M., Teng, M., Walla, T.R., Caron-Guay, A., Pal, C., Laliberté, E.: SelvaBox: A high-resolution dataset for tropical tree crown detection. In: ICLR (2026)

    2026

  6. [6]

    Remote Sensing15(5), 1463 (2023)

    Beloiu, M., Heinzmann, L., Rehush, N., Gessler, A., Griess, V.C.: Individual tree- crown detection and species identification in heterogeneous forests using aerial rgb imagery and deep learning. Remote Sensing15(5), 1463 (2023)

    2023

  7. [7]

    In: NeurIPS (2025)

    Bolya, D., Huang, P.Y., Sun, P., Cho, J.H., Madotto, A., Wei, C., Ma, T., Zhi, J., Rajasegaran, J., Rasheed, H., et al.: Perception encoder: The best visual embed- dings are not at the output of the network. In: NeurIPS (2025)

    2025

  8. [8]

    New Forests51(4), 573–596 (2020)

    Camarretta, N., Harrison, P.A., Bailey, T., Potts, B., Lucieer, A., Davidson, N., Hunt, M.: Monitoring forest structure to guide adaptive management of forest restoration: a review of remote sensing approaches. New Forests51(4), 573–596 (2020)

    2020

  9. [9]

    In: ECCV

    Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S.: End- to-end object detection with transformers. In: ECCV. pp. 213–229 (2020)

    2020

  10. [10]

    IEEE Geoscience and Remote Sensing Letters17(8), 1391–1395 (2019)

    Carvalho, M., Le Saux, B., Trouvé-Peloux, P., Champagnat, F., Almansa, A.: Mul- titask learning of height and semantics from aerial images. IEEE Geoscience and Remote Sensing Letters17(8), 1391–1395 (2019)

    2019

  11. [11]

    Remote Sensing of Environment311, 114283 (2024)

    Cloutier, M., Germain, M., Laliberté, E.: Influence of temperate forest autumn leaf phenology on segmentation of tree species from uav imagery using deep learning. Remote Sensing of Environment311, 114283 (2024)

    2024

  12. [12]

    In: CVPR

    Cui, Y., Jia, M., Lin, T.Y., Song, Y., Belongie, S.: Class-balanced loss based on effective number of samples. In: CVPR. pp. 9268–9277 (2019)

    2019

  13. [13]

    Bi- ological Conservation253, 108907 (2021) 16 J

    Davies, S.J., Abiem, I., Salim, K.A., Aguilar, S., Allen, D., Alonso, A., Anderson- Teixeira, K., Andrade, A., Arellano, G., Ashton, P.S., et al.: ForestGEO: Under- standing forest diversity and dynamics through a global observatory network. Bi- ological Conservation253, 108907 (2021) 16 J. Endreset al

    2021

  14. [14]

    In: CVPR

    Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: Imagenet: A large-scale hierarchical image database. In: CVPR. pp. 248–255 (2009)

    2009

  15. [15]

    In: ICLR (2020)

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale. In: ICLR (2020)

    2020

  16. [16]

    arXiv:2602.02426 [cs.CV] (2026)

    Duguay, S.O., Baudchon, H., Laliberté, E., Muller-Landau, H., Rivas-Torres, G., Ouaknine, A.: SelvaMask: Segmenting trees in tropical forests and beyond. arXiv:2602.02426 [cs.CV] (2026)

    arXiv 2026

  17. [17]

    Computers and Electronics in Agriculture219, 108785 (2024)

    Ecke, S., Stehr, F., Frey, J., Tiede, D., Dempewolf, J., Klemmt, H.J., Endres, E., Seifert, T.: Towards operational uav-based forest health monitoring: Species iden- tification and crown condition assessment by means of deep learning. Computers and Electronics in Agriculture219, 108785 (2024)

    2024

  18. [18]

    Remote Sensing12(23), 3892 (2020)

    Egli, S., Höpke, M.: Cnn-based tree species classification using high resolution rgb image data from automated uav observations. Remote Sensing12(23), 3892 (2020)

    2020

  19. [19]

    In: NeurIPS (2014)

    Eigen, D., Puhrsch, C., Fergus, R.: Depth map prediction from a single image using a multi-scale deep network. In: NeurIPS (2014)

    2014

  20. [20]

    FAO, Rome, Italy (2025).https: //doi.org/10.4060/cd6709en

    FAO: Global Forest Resources Assessment 2025. FAO, Rome, Italy (2025).https: //doi.org/10.4060/cd6709en

    doi:10.4060/cd6709en 2025

  21. [21]

    https://doi.org/10.60635/C3F593

    ForestGEO Smithsonian: 2023 high-resolution airborne LiDAR data for Barro Col- orado Island and other Smithsonian ForestGEO Sites in Central Panama (2024). https://doi.org/10.60635/C3F593

    work page doi:10.60635/c3f593 2023

  22. [22]

    Forest Ecology and Management555, 121712 (2024)

    Fu, H., Zhao, H., Jiang, J., Zhang, Y., Liu, G., Xiao, W., Du, S., Guo, W., Liu, X.: Automatic detection tree crown and height using mask r-cnn based on unmanned aerial vehicles images for biomass mapping. Forest Ecology and Management555, 121712 (2024)

    2024

  23. [23]

    IEEE Transactions on Geoscience and Remote Sensing61, 1–15 (2023)

    Gao,Z.,Sun,W.,Lu,Y.,Zhang,Y.,Song,W.,Zhang,Y.,Zhai,R.:Jointlearningof semantic segmentation and height estimation for remote sensing image leveraging contrastive learning. IEEE Transactions on Geoscience and Remote Sensing61, 1–15 (2023)

    2023

  24. [24]

    In: ICCV

    Girshick, R.: Fast r-cnn. In: ICCV. pp. 1440–1448 (2015)

    2015

  25. [25]

    In: First conference on language modeling (2024)

    Gu, A., Dao, T.: Mamba: Linear-time sequence modeling with selective state spaces. In: First conference on language modeling (2024)

    2024

  26. [26]

    In: NeurIPS (2025)

    Gu, J., Stevens, S., Campolongo, E.G., Thompson, M.J., Zhang, N., Wu, J., Kopanev, A., Mai, Z., White, A.E., Balhoff, J., Dahdul, W., Rubenstein, D., Lapp, H., Berger-Wolf, T., Chao, W.L., Su, Y.: BioCLIP 2: Emergent properties from scaling hierarchical contrastive learning. In: NeurIPS (2025)

    2025

  27. [27]

    ISPRS Journal of Photogrammetry and Remote Sensing178, 112–123 (2021)

    Hao, Z., Lin, L., Post, C.J., Mikhailova, E.A., Li, M., Chen, Y., Yu, K., Liu, J.: Automated tree-crown and height detection in a young forest plantation using mask region-based convolutional neural network (mask r-cnn). ISPRS Journal of Photogrammetry and Remote Sensing178, 112–123 (2021)

    2021

  28. [28]

    In: CVPR

    Hatamizadeh, A., Kautz, J.: Mambavision: A hybrid mamba-transformer vision backbone. In: CVPR. pp. 25261–25270 (2025)

    2025

  29. [29]

    In: ICCV

    He, K., Gkioxari, G., Dollár, P., Girshick, R.: Mask r-cnn. In: ICCV. pp. 2961–2969 (2017)

    2017

  30. [30]

    In: CVPR

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: CVPR. pp. 770–778 (2016)

    2016

  31. [31]

    arXiv:1606.08415 [cs.LG] (2016)

    Hendrycks, D.: Gaussian error linear units (gelus). arXiv:1606.08415 [cs.LG] (2016)

    Pith/arXiv arXiv 2016

  32. [32]

    Global change biology 28(17), 5254–5268 (2022) Estimating Individual Tree Height and Species from UAV Imagery 17

    Jucker, T., Fischer, F.J., Chave, J., Coomes, D.A., Caspersen, J., Ali, A., Loub- ota Panzou, G.J., Feldpausch, T.R., Falster, D., Usoltsev, V.A., et al.: Tallo: A global tree allometry and crown architecture database. Global change biology 28(17), 5254–5268 (2022) Estimating Individual Tree Height and Species from UAV Imagery 17

    2022

  33. [33]

    In: CVPR

    Kendall, A., Gal, Y., Cipolla, R.: Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In: CVPR. pp. 7482–7491 (2018)

    2018

  34. [34]

    In: NeurIPS (2012)

    Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep con- volutional neural networks. In: NeurIPS (2012)

    2012

  35. [35]

    bioRxiv (2025).https://doi.org/10.1101/2025.09.02.673753

    Laliberté, E., Caron-Guay, A., Le Falher, V., Tougas, G., Muller-Landau, H.C., Rivas-Torres, G., Walla, T.R., Baudchon, H., Hernandez, M., Buenaño, A., Weber, A., Chambers, J.Q., Inuma, J.C., Araúz, F., Valdes, J., Hernández, A., Brassfield, D., Sérgio, P., Vasquez, V., Simonetti, A., Marra, D.M., Vasconcelos, C., Vaca, J.F., Rivadeneyra, G., Illanes, J.,...

    work page doi:10.1101/2025.09.02.673753 2025

  36. [36]

    Nature Ecology & Evolution7(11), 1778–1789 (2023)

    Lang, N., Jetz, W., Schindler, K., Wegner, J.D.: A high-resolution canopy height model of the earth. Nature Ecology & Evolution7(11), 1778–1789 (2023)

    2023

  37. [37]

    Remote Sensing of Environment233, 111347 (2019)

    Lang, N., Schindler, K., Wegner, J.D.: Country-wide high-resolution vegetation height mapping with sentinel-2. Remote Sensing of Environment233, 111347 (2019)

    2019

  38. [38]

    Mémoire de maîtrise, Université de Montréal, Montréal, Québec (Jul 2025).https://doi.org/10.71781/20561

    Lefebvre, I.: Suivi de la séquestration de CO2 aérien dans des sites en reforestation à l’aide de drones. Mémoire de maîtrise, Université de Montréal, Montréal, Québec (Jul 2025).https://doi.org/10.71781/20561

    work page doi:10.71781/20561 2025

  39. [39]

    https://doi.org/10.20383/103.0979

    Lefebvre, I., Laliberté, E.: Uav lidar, uav imagery, tree segmentations and ground measurements for estimating tree biomass in canadian (quebec) plantations (2024). https://doi.org/10.20383/103.0979

    work page doi:10.20383/103.0979 2024

  40. [40]

    Lefort, T., Affouard, A., Charlier, B., Lombardo, J.C., Chouet, M., Goëau, H., Salmon, J., Bonnet, P., Joly, A.: Cooperative learning of pl@ ntnet’s artificial intelligence algorithm: How does it work and how can we improve it? Methods in Ecology and Evolution (2024)

    2024

  41. [41]

    PNAS nexus2(2023)

    Li, S., Brandt, M., Fensholt, R., Kariryaa, A., Igel, C., Gieseke, F., Nord-Larsen, T., Oehmcke, S., Carlsen, A.H., Junttila, S., et al.: Deep learning enables image- based tree counting, crown segmentation, and height prediction at national scale. PNAS nexus2(2023)

    2023

  42. [42]

    Remote Sensing 14(13), 3035 (2022)

    Li, Y., Chai, G., Wang, Y., Lei, L., Zhang, X.: ACE R-CNN: An attention comple- mentary and edge detection-based instance segmentation algorithm for individual tree species identification using uav rgb images and lidar data. Remote Sensing 14(13), 3035 (2022)

    2022

  43. [43]

    Drones 9(4), 296 (2025)

    Li, Z., Yu, S., Ye, Q., Zhang, M., Yin, D., Zhao, Z.: Tree species classification using uav-based rgb images and spectral information on the loess plateau, china. Drones 9(4), 296 (2025)

    2025

  44. [44]

    In: ICCV

    Lin, T.Y., Goyal, P., Girshick, R., He, K., Dollár, P.: Focal loss for dense object detection. In: ICCV. pp. 2980–2988 (2017)

    2017

  45. [45]

    In: CVPR

    Liu, S., Johns, E., Davison, A.J.: End-to-end multi-task learning with attention. In: CVPR. pp. 1871–1880 (2019)

    2019

  46. [46]

    In: NeurIPS

    Liu, Y., Tian, Y., Zhao, Y., Yu, H., Xie, L., Wang, Y., Ye, Q., Jiao, J., Liu, Y.: Vmamba: Visual state space model. In: NeurIPS. pp. 103031–103063 (2024)

    2024

  47. [47]

    In: CVPR

    Liu,Z.,Hu,H.,Lin,Y.,Yao,Z.,Xie,Z.,Wei,Y.,Ning,J.,Cao,Y.,Zhang,Z.,Dong, L., et al.: Swin transformer v2: Scaling up capacity and resolution. In: CVPR. pp. 12009–12019 (2022)

    2022

  48. [48]

    In: CVPR

    Liu, Z., Mao, H., Wu, C.Y., Feichtenhofer, C., Darrell, T., Xie, S.: A convnet for the 2020s. In: CVPR. pp. 11976–11986 (2022)

    2022

  49. [49]

    In: ICLR (2017) 18 J

    Loshchilov, I., Hutter, F.: SGDR: Stochastic gradient descent with warm restarts. In: ICLR (2017) 18 J. Endreset al

    2017

  50. [50]

    In: ICLR (2019)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: ICLR (2019)

    2019

  51. [51]

    In: ICML

    Nair, V., Hinton, G.E.: Rectified linear units improve restricted boltzmann ma- chines. In: ICML. pp. 807–814 (2010)

    2010

  52. [52]

    The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences42, 475–481 (2019)

    Natesan, S., Armenakis, C., Vepakomma, U.: Resnet-based tree species classifica- tion using uav images. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences42, 475–481 (2019)

    2019

  53. [53]

    Journal of Unmanned Vehicle Systems8(4), 310–333 (2020)

    Natesan, S., Armenakis, C., Vepakomma, U.: Individual tree species identification using dense convolutional network (densenet) on multitemporal rgb images from uav. Journal of Unmanned Vehicle Systems8(4), 310–333 (2020)

    2020

  54. [54]

    Remote Sensing12(7), 1070 (2020)

    Nezami, S., Khoramshahi, E., Nevalainen, O., Pölönen, I., Honkavaara, E.: Tree species classification of drone hyperspectral and rgb imagery with deep learning convolutional neural networks. Remote Sensing12(7), 1070 (2020)

    2020

  55. [55]

    TMLR (2024)

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al.: Dinov2: Learning robust visual features without supervision. TMLR (2024)

    2024

  56. [56]

    Science333(6045), 988–993 (2011)

    Pan, Y., Birdsey, R.A., Fang, J., Houghton, R., Kauppi, P.E., Kurz, W.A., Phillips, O.L., Shvidenko, A., Lewis, S.L., Canadell, J.G., Ciais, P., Jackson, R.B., Pacala, S.W., McGuire, A.D., Piao, S., Rautiainen, A., Sitch, S., Hayes, D.: A large and persistent carbon sink in the world’s forests. Science333(6045), 988–993 (2011). https://doi.org/10.1126/sci...

    doi:10.1126/science.1201609 2011

  57. [57]

    Advances in neural information processing sys- tems32(2019)

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: PyTorch: An imperative style, high- performance deep learning library. Advances in neural information processing sys- tems32(2019)

    2019

  58. [58]

    In: ICML

    Pauls, J., Zimmer, M., Kelly, U.M., Schwartz, M., Saatchi, S., Ciais, P., Pokutta, S., Brandt, M., Gieseke, F.: Estimating canopy height at scale. In: ICML. pp. 39972–39988 (2024)

    2024

  59. [59]

    In: ICML (2025)

    Pauls, J., Zimmer, M., Turan, B., Saatchi, S., Ciais, P., Pokutta, S., Gieseke, F.: Capturing temporal dynamics in large-scale canopy tree height estimation. In: ICML (2025)

    2025

  60. [60]

    Plants 12(21), 3710 (2023)

    Phillips, O.L.: Sensing forests directly: The power of permanent plots. Plants 12(21), 3710 (2023)

    2023

  61. [61]

    ISPRS Annals of the Photogram- metry, Remote Sensing and Spatial Information Sciences10, 1089–1096 (2023)

    Pierdicca, R., Nepi, L., Mancini, A., Malinverni, E., Balestra, M.: Uav4tree: Deep learning-based system for automatic classification of tree species using rgb optical images obtained by an unmanned aerial vehicle. ISPRS Annals of the Photogram- metry, Remote Sensing and Spatial Information Sciences10, 1089–1096 (2023)

    2023

  62. [62]

    Remote Sensing of Environment280, 113143 (2022)

    Qin, H., Zhou, W., Yao, Y., Wang, W.: Individual tree segmentation and tree species classification in subtropical broadleaf forests using uav-based lidar, hyper- spectral, and ultrahigh-resolution rgb data. Remote Sensing of Environment280, 113143 (2022)

    2022

  63. [63]

    Computers and Electronics in Agriculture240, 111162 (2026)

    Que, H., Gao, H., Shan, W., Liu, M., An, J., Deng, F., Feng, S., Yang, X., Mu, L.: FM-SAM: individual tree crown delineation and classification based on segmen- tation anything model (sam) and yolov10 in uav imagery for forest monitoring. Computers and Electronics in Agriculture240, 111162 (2026)

    2026

  64. [64]

    Environmental Data Science4, e37 (2025).https://doi.org/ 10.1017/eds.2025.10013

    Ramesh, V., Ouaknine, A., Rolnick, D.: Tree semantic segmentation from aerial image time series. Environmental Data Science4, e37 (2025).https://doi.org/ 10.1017/eds.2025.10013

    doi:10.1017/eds.2025.10013 2025

  65. [65]

    In: International Conference on Medical image computing and computer-assisted intervention

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedi- cal image segmentation. In: International Conference on Medical image computing and computer-assisted intervention. pp. 234–241 (2015) Estimating Individual Tree Height and Species from UAV Imagery 19

    2015

  66. [66]

    ISPRS Journal of Photogrammetry and Remote Sensing170, 205–215 (2020)

    Schiefer, F., Kattenborn, T., Frick, A., Frey, J., Schall, P., Koch, B., Schmidtlein, S.: Mapping forest tree species in high resolution uav-based rgb-imagery by means of convolutional neural networks. ISPRS Journal of Photogrammetry and Remote Sensing170, 205–215 (2020)

    2020

  67. [67]

    IEEE Transactions on Geo- science and Remote Sensing (2025)

    Shen, Y., Xiao, L., Chen, J., Du, Q., Ye, Q.: Learning cross-task features with mamba for remote sensing image multi-task prediction. IEEE Transactions on Geo- science and Remote Sensing (2025)

    2025

  68. [68]

    arXiv:2508.10104 [cs.CV] (2025)

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khalidov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., Massa, F., Haziza, D., Wehrstedt, L., Wang, J., Darcet, T., Moutakanni, T., Sentana, L., Roberts, C., Vedaldi, A., Tolan, J., Brandt, J., Couprie, C., Mairal, J., Jégou, H., Labatut, P., Bojanowski, P.: DINOv3. arXiv:2508.10104...

    Pith/arXiv arXiv 2025

  69. [69]

    In: 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS)

    Srivastava, S., Volpi, M., Tuia, D.: Joint height estimation and semantic labeling of monocular aerial images with cnns. In: 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS). pp. 5173–5176 (2017)

    2017

  70. [70]

    In: CVPR

    Stevens, S., Wu, J., Thompson, M.J., Campolongo, E.G., Song, C.H., Carlyn, D.E., Dong, L., Dahdul, W.M., Stewart, C., Berger-Wolf, T., et al.: BioCLIP: A vision foundation model for the tree of life. In: CVPR. pp. 19412–19424 (2024)

    2024

  71. [71]

    Forests10(11), 1047 (2019)

    Sun, Y., Huang, J., Ao, Z., Lao, D., Xin, Q.: Deep learning approaches for the mapping of tree species diversity in a tropical wetland using airborne lidar and high-spatial-resolution remote sensing images. Forests10(11), 1047 (2019)

    2019

  72. [72]

    In: NeurIPS (2025)

    Teng, M., Ouaknine, A., Laliberté, E., Bengio, Y., Rolnick, D., Larochelle, H.: Bringing SAM to new heights: Leveraging elevation data for tree crown segmenta- tion from drone imagery. In: NeurIPS (2025)

    2025

  73. [73]

    Remote Sensing of Environment300, 113888 (2024)

    Tolan,J.,Yang,H.I.,Nosarzewski,B.,Couairon,G.,Vo,H.V.,Brandt,J.,Spore,J., Majumdar, S., Haziza, D., Vamaraju, J., et al.: Very high resolution canopy height maps from rgb imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar. Remote Sensing of Environment300, 113888 (2024)

    2024

  74. [74]

    arXiv:2502.14786 [cs.CV] (2025)

    Tschannen, M., Gritsenko, A., Wang, X., Naeem, M.F., Alabdulmohsin, I., Parthasarathy, N., Evans, T., Beyer, L., Xia, Y., Mustafa, B., Hénaff, O., Harmsen, J., Steiner, A., Zhai, X.: SigLIP 2: Multilingual vision-language encoders with im- proved semantic understanding, localization, and dense features. arXiv:2502.14786 [cs.CV] (2025)

    Pith/arXiv arXiv 2025

  75. [75]

    In: ICML (2025)

    Tseng, G., Fuller, A., Reil, M., Herzog, H., Beukema, P., Bastani, F., Green, J.R., Shelhamer, E., Kerner, H., Rolnick, D.: Galileo: Learning global & local features of many remote sensing modalities. In: ICML (2025)

    2025

  76. [76]

    24784053.v2

    Vasquez,V.,Cushman,K.,Ramos,P.,Williamson,C.,Villareal,P.,Correa,L.F.G., Muller-Landau, H.C.: Barro colorado island 50-ha plot crown maps: Manually segmented and instance segmented (2023).https://doi.org/10.25573/data. 24784053.v2

    work page doi:10.25573/data 2023

  77. [77]

    In: NeurIPS (2017)

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. In: NeurIPS (2017)

    2017

  78. [78]

    Remote Sensing of Environment305, 114099 (2024)

    Wagner, F.H., Roberts, S., Ritz, A.L., Carter, G., Dalagnol, R., Favrichon, S., Hirye, M.C., Brandt, M., Ciais, P., Saatchi, S.: Sub-meter tree height mapping of california using aerial images and lidar-informed u-net model. Remote Sensing of Environment305, 114099 (2024)

    2024

  79. [79]

    Endreset al

    Wagner, F.H., Dalagnol, R., Carter, G., Hirye, M.C.M., Gill, S., Le, S.B.T., Favri- chon, S., Keller, M., Ometto, J.P.H.B., Alves, L., Creze, C., George-Chacón, S.P., Li, S., Liu, Z., Mullissa, A.G., Yang, Y., dos Santos, E.G., Worden, S.R., 20 J. Endreset al. Brandt, M., Ciais, P., Hagen, S.C., Saatchi, S.: High resolution tree height map- ping of the am...

    Pith/arXiv arXiv 2025

  80. [80]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing17, 11632–11654 (2024)

    Wang, D., Zhang, J., Xu, M., Liu, L., Wang, D., Gao, E., Han, C., Guo, H., Du, B., Tao, D., et al.: Mtp: Advancing remote sensing foundation model via multitask pretraining. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing17, 11632–11654 (2024)

    2024

Showing first 80 references.