Pith. sign in

REVIEW 3 major objections 5 minor 64 references

Pretraining on 75 hectares of synthetic forest scans lets a single real plot match full-scale training.

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

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

T0 review · deepseek-v4-flash

2026-08-04 16:56 UTC pith:PJGNA7EB

load-bearing objection A genuinely useful synthetic-data pipeline for forest LiDAR segmentation, but the 'one real plot suffices' claim is quantitatively under-supported by single-run F1 scores. the 3 major comments →

arxiv 2509.11201 v1 pith:PJGNA7EB submitted 2025-09-14 cs.CV

Scaling Up Forest Vision with Synthetic Data

Yihang She , Andrew Blake , David Coomes , Srinivasan Keshav This is my paper
classification cs.CV
keywords synthetic dataLiDAR simulationtree instance segmentationforest point cloudsfew-shot learningprocedural foliage generationUAV laser scanningsim-to-real transfer
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper claims that synthetic 3D forest scans, generated from procedurally grown game-engine forest scenes and rendered through physics-based LiDAR simulation, can substitute for almost all real labelled data in tree-segmentation models. Pretraining on the paper's 75-hectare synthetic corpus and then fine-tuning on one real plot of under 0.1 hectare yields segmentations competitive with models trained on the full 2.79-hectare real benchmark. Training from scratch on the same few plots collapses in accuracy, so the gain is attributed to the synthetic pretraining. The paper further claims, through ablations, that the transfer works because of three factors: physics-based simulation, scene diversity, and dataset scale.

Core claim

A 3D instance-segmentation network pretrained on 11 synthetic forest scenes (187 plots, 75 hectares) and fine-tuned on a single real plot recovers nearly all the accuracy of training on all real plots: one coniferous region reaches 77.1 F1 versus 79.8 with full real data, and one deciduous region reaches 69.3 versus 69.9. Training from scratch on the corresponding few-shot subsets collapses to 50.5 and 31.6. Ablations show F1 rising from 5.4 to 42.6 when LiDAR-simulated points replace raw mesh nodes, to 81.5 when scene diversity is added, and to 87.0 when dataset scale increases.

What carries the argument

The load-bearing mechanism is a fully automated synthetic-data pipeline: procedural foliage generation grows large forest stands from a small set of tree meshes; a Python bridge exports each tree with a unique instance ID; and a ray-tracing LiDAR simulator performs a virtual UAV survey, producing point clouds whose density and occlusion patterns match real UAV laser scans. This combination supplies the physics, diversity, and scale that the ablations rank as the three drivers of successful sim-to-real transfer.

Load-bearing premise

The premise that game-engine procedural forests, built from a limited set of tree meshes, reproduce the structural statistics of real forests in the ways that matter for learning tree segmentation—canopy overlap, branch occlusion, understory density, and stem visibility—carries the whole transfer result.

What would settle it

Apply the same one-plot fine-tuning protocol to an independent real UAV-LiDAR dataset that was not part of the paper's experiments, covering different species and scanner settings. If the fine-tuned F1 falls back to within a few points of training from scratch, rather than within a few points of the full-data result, the transfer claim fails.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • A new forest region can reach near-full-data segmentation accuracy with a single annotated plot, cutting annotation cost and time by orders of magnitude.
  • Fine-tuning a pretrained model needs 60 epochs versus 120 from scratch, so synthetic pretraining also cuts deployment compute in half.
  • Scene diversity produces larger gains than simply adding more plots per scene, pointing dataset-construction effort toward new forest types rather than more samples of the same type.
  • The released pipeline lets researchers generate unlimited labelled forest scans for new regions or sensors, including terrestrial, mobile, and airborne LiDAR.
  • The transfer holds for both coniferous and deciduous forests, showing the learned representation is not locked to one forest type.

Where Pith is reading between the lines

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

  • If the scaling trend continues past 75 hectares, the synthetic-data approach should combine well with larger transformer-style segmentation models, since those architectures keep improving with data volume.
  • The reusability argument implies that a library of tree meshes digitized from a few field plots can be procedurally recombined to generate training data for any new region, amortizing each annotated campaign many times over.
  • The paper explicitly limits its synthetic labels to binary tree/non-tree and reports an unexplained drop on one deciduous test region; extending synthetic annotations to leaf-wood or five-class labels is the natural experiment to see whether the remaining gap closes.
  • The same synthetic corpus could serve as a balanced evaluation benchmark for comparing tree-segmentation algorithms across forest types, since real benchmarks are heavily skewed toward coniferous plots.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces a synthetic data generation pipeline for 3D forest point clouds, combining Unreal Engine procedural foliage generation with the HELIOS LiDAR simulator. The authors generate a 75-hectare, 12-scene annotated dataset, pretrain a simplified version of the ForAINet segmentation model on it, and then fine-tune on small subsets of the real FOR-Instance dataset. The central claim is that fine-tuning the pretrained model on a single real plot of less than 0.1 hectare yields segmentations competitive with a model trained on the full-scale real data. Additional experiments attribute the synthetic-data benefit to three factors: physics-based LiDAR simulation, scene diversity, and dataset scale. Mixed training with synthetic and real data is also evaluated and honestly reported as matching, but not surpassing, prior state-of-the-art.

Significance. If the central claim holds, this is a substantial contribution: the dataset and pipeline could materially reduce the annotation burden for individual-tree segmentation, and the paper identifies concrete factors that make synthetic forest data useful. The manuscript is strengthened by the use of an external benchmark (FOR-Instance) and an independent algorithm (ForAINet), by a relatively clean ablation design for the diversity-versus-scale comparison, and by the open-sourcing of the pipeline and dataset. The mixed-training section is notably honest in reporting parity rather than gains. However, the headline equivalence claim currently rests on single-run point estimates without uncertainty quantification, which is a load-bearing weakness. The external-validity assumption about procedural forest scenes representing real forest structure is also asserted rather than demonstrated.

major comments (3)
  1. [§4.1, Figures 12/20/21] Every reported F1/precision/recall value is a single point estimate from one training run. The abstract and §4.1 claim that fine-tuning on one real plot is 'competitive' with training on full-scale Real6,42; the supporting gaps are as small as 0.6 points (Deciduous-R: 69.3 vs 69.9) and as large as 6.3 points (Coniferous-C: 87.1 vs 93.4). Training is stochastic (TreeMix3D random replacement, random cylinder sampling, network initialization and optimizer noise), so the parity claim is not assessable without multiple seeds or repeated trials. Please report mean ± std over at least 3–5 seeds, or rephrase the claim as 'not distinguishable' only if statistically supported. This is load-bearing because the headline conclusion is an equivalence claim.
  2. [§4.2, Figure 16/22] The 'physics' factor compares Nodal4,20 (tree-mesh nodal points, 100–200 pts/m²) with Sim4,20 (HELIOS LiDAR simulation, >1000 pts/m²). These differ not only in ray-tracing physics but also in point density, occlusion patterns, and sampling distribution. The large F1 gains could be due to density or coverage rather than physically correct LiDAR returns. To support the statement that 'physics-based simulation is critical,' the ablation should control point density (e.g., densify nodal points to matched density, or decimate simulated clouds) and ideally hold scene geometry fixed while varying the forward model.
  3. [§3.1 and §5.1, Figure 19] The general sim-to-real transfer claim rests on the assumption that UE procedural forests are structurally representative of real forests. This is asserted via 'broadly consistent with ecological theory' rather than demonstrated quantitatively. The Deciduous3 scene, for example, uses 17 base meshes for 2,053 trees, and the dataset is built from a small set of gaming assets. I recommend adding a quantitative comparison of structural statistics (e.g., tree density, canopy overlap, stem visibility, branch occlusion) or point-cloud distributions between synthetic scenes and FOR-Instance regions, and/or evaluating on an additional real dataset such as Wytham Woods. Without this, the results on five regions may be dataset-specific rather than a general capability.
minor comments (5)
  1. [§3.3] The semantic remapping text reads: 'tree: woody branches, live branches, low-vegetation. non-tree: low-vegetation, ground.' This assigns low-vegetation to both classes. Presumably the tree class should be stem, woody branches, and live branches. Please correct and confirm the binary mapping actually used in training.
  2. [§3.3] The list of coniferous regions is given as 'Coniferous-N, Coniferous-S, Coniferous-S'; the second should be Coniferous-C.
  3. [§4.1] The statement that extreme few-shot training 'drops just a couple of percentage points on F1' understates the Coniferous-C result, where the gap is 6.3 points (87.1 vs 93.4). Please qualify the claim with region-specific gaps.
  4. [§4.2] The text says 'Sim 11,20 adds three deciduous and four coniferous scenes not in Sim4,20 and Sim4,20' — this should presumably read 'Sim11,55' and the sentence is garbled. Please correct.
  5. [§4.3, Figure 18] The Deciduous-T discrepancy (47.4 vs 69.4 reported in Xiang et al.) is attributed to 'random seed effects or the small size of the test set,' but no supporting evidence is provided. This is another manifestation of the missing uncertainty quantification; at minimum, acknowledge the inconsistency more explicitly.

Circularity Check

0 steps flagged

No significant circularity: the sim-to-real claim is tested on an external benchmark (FOR-Instance) and no fitted quantity is repackaged as a prediction.

full rationale

The paper's derivation chain is empirical: Unreal Engine scenes combined with HELIOS LiDAR simulation produce the synthetic dataset Sim11,187; ForAINet is pretrained on it, fine-tuned on small FOR-Instance subsets, and evaluated on held-out FOR-Instance test plots. No equation defines the reported F1 scores in terms of the synthetic-data inputs, no parameter is fitted to the test metric, and the test benchmark (Puliti et al.) and base algorithm (Xiang et al.) are external to the authors. The only notable self-citation is the SPREAD dataset (Feng et al. 2025), from which five scenes are adapted; that citation is data provenance and is not used to justify the central sim-to-real claim, which is instead supported by the present experiments on real data. The paper's own caveats—e.g., Section 4.3 attributing the Deciduous-T drop possibly to 'random seed effects or the small size of the test set,' and Section 5.2 calling binary labels 'a limitation of our generated data'—are statistical or scope limitations, not circularity. The ablations on physics/diversity/scale are partially confounded (adding scenes also increases data volume), and results are single runs without error bars, but these are correctness/robustness concerns rather than reductions of outputs to inputs by construction. No circular step is present, so the circularity score is 0.

Axiom & Free-Parameter Ledger

2 free parameters · 6 axioms · 0 invented entities

No free parameter is fitted to the target F1 metric: the pipeline inputs (APFG parameters, survey settings) are hand-chosen but anchored to ecological plausibility and real RIEGL VUX-1 survey practice, and the training schedules follow the prior ForAINet work. The load-bearing premises are all domain assumptions about sim-to-real validity: game-engine forests stand in for real canopy structure, HELIOS ray-tracing reproduces real scan statistics, per-tree mesh IDs match the annotator notion of an individual tree, and binary tree/non-tree labels capture the task. No invented entities. The least-supported premise is that APFG-generated scenes are ecologically representative; the evidence for it is indirect (transfer results on five real regions).

free parameters (2)
  • APFG procedural foliage parameters (overstory and understory) = Overstory: density 3.0, collision 250-300 cm, shade 400-500 cm; understory: density 2.0, collision 50-100 cm (Fig. 3)
    Hand-chosen values for seed density, collision/shade radii, spread distance and variance, number of steps, and max age. They determine the structure and density of every synthetic forest scene. They are justified by ecological plausibility rather than derived from real forest data or fitted to the target metric.
  • Virtual UAV-LiDAR survey parameters = RIEGL VUX-1 UAV, 100 kHz, 15 returns, criss-cross, 20 m spacing, 5 m/s, 60 m altitude (120 m Redwood)
    Scanner model, pulse frequency, flight pattern, spacing, speed, and altitude are chosen to match real RIEGL VUX-1 UAV surveys as in Puliti et al. (2023). These are reasonable inputs anchored to real practice, not fitted to the outcome; the Redwood altitude is adjusted for tree size.
axioms (6)
  • domain assumption Game-engine forests generated by Unreal Engine's procedural foliage algorithm, from a limited set of tree assets, are structurally representative of real forests (canopy overlap, branch occlusion, understory density) for learning transfer.
    Load-bearing premise of the whole sim-to-real pipeline. Section 3.1 asserts APFG is 'broadly consistent with ecological theory' citing Halle and Oldeman (1970), Pacala et al. (1996), and Purves et al. (2007), but these are separate ecological models; no reference shows UE's collision/shade radius mechanics instantiate them. Evidence is indirect: few-shot gains on five real regions in Section 4.1.
  • domain assumption HELIOS ray-tracing LiDAR simulation faithfully reproduces the relevant statistics of real UAV laser scans (point density, multi-return, occlusion), sufficient for representation learning.
    Section 3.1 uses HELIOS voxel ray-tracing with RIEGL VUX-1 settings. The paper's own physics ablation (Sim4,20 vs Nodal4,20, Figure 16) gives supporting evidence but is confounded with point density (100-200 vs >1000 pts/m2), so fidelity is not isolated.
  • domain assumption Per-tree unique IDs assigned to game-engine tree meshes correspond to the same 'individual tree' notion that FOR-Instance annotators label; trained instance discrimination transfers across this gap.
    Section 3.1: each exported tree object gets a unique ID. In synthetic scenes instances are separated by construction, while real scans require grouping interleaved crowns; the paper does not discuss whether this makes synthetic instance learning easier.
  • domain assumption Pretraining on synthetic data followed by fine-tuning on real data transfers (domain randomization / meta-learning assumption).
    Section 1 cites Tobin et al. (2017), Prakash et al. (2019), and Finn et al. (2017). The paper's few-shot results are evidence for this premise, but there is no control pretrained on large-scale real data, so the attribution to synthetic pretraining is by design rather than by direct comparison.
  • domain assumption The binary tree/non-tree remapping of FOR-Instance labels preserves the instance segmentation task well enough for the central claims.
    Section 3.3 remaps five semantic classes to two for consistency with synthetic labels. The remapping text is internally inconsistent (low-vegetation listed under both tree and non-tree). The paper notes in Section 5.2 that the two-label setting causes 'some performance loss' relative to five-class learning.
  • domain assumption Point density above 1000 pts/m2 is adequate for effective tree segmentation learning.
    Section 3.2, citing Wielgosz et al. (2024). Standard empirical assumption in the field, used to justify the tiling/thresholding of synthetic plots.

pith-pipeline@v1.3.0-alltime-deepseek · 20486 in / 26685 out tokens · 270007 ms · 2026-08-04T16:56:51.564560+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Scaling Up Forest Vision with Synthetic Data." pith.science (2026). https://pith.science/paper/PJGNA7EB

@misc{pith2026250911201,
  author       = {Pith},
  title        = {Pith review of: Scaling Up Forest Vision with Synthetic Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PJGNA7EB}},
  note         = {Machine review of arXiv:2509.11201}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Accurate tree segmentation is a key step in extracting individual tree metrics from forest laser scans, and is essential to understanding ecosystem functions in carbon cycling and beyond. Over the past decade, tree segmentation algorithms have advanced rapidly due to developments in AI. However existing, public, 3D forest datasets are not large enough to build robust tree segmentation systems. Motivated by the success of synthetic data in other domains such as self-driving, we investigate whether similar approaches can help with tree segmentation. In place of expensive field data collection and annotation, we use synthetic data during pretraining, and then require only minimal, real forest plot annotation for fine-tuning. We have developed a new synthetic data generation pipeline to do this for forest vision tasks, integrating advances in game-engines with physics-based LiDAR simulation. As a result, we have produced a comprehensive, diverse, annotated 3D forest dataset on an unprecedented scale. Extensive experiments with a state-of-the-art tree segmentation algorithm and a popular real dataset show that our synthetic data can substantially reduce the need for labelled real data. After fine-tuning on just a single, real, forest plot of less than 0.1 hectare, the pretrained model achieves segmentations that are competitive with a model trained on the full scale real data. We have also identified critical factors for successful use of synthetic data: physics, diversity, and scale, paving the way for more robust 3D forest vision systems in the future. Our data generation pipeline and the resulting dataset are available at https://github.com/yihshe/CAMP3D.git.

Figures

Figures reproduced from arXiv: 2509.11201 by Andrew Blake, David Coomes, Srinivasan Keshav, Yihang She.

Figure 1
Figure 1. Figure 1: Individual tree segmentation is essential for deriving key forest properties in ecology. This figure shows examples of metrics derived from individual tree segmentation: canopy height and diameter at breast height for biomass and carbon estimation, and crown area, crown volume, and species composition for understanding forest structure, biodiversity, and ecosystem functions. The choice of algorithm plays a… view at source ↗
Figure 2
Figure 2. Figure 2: Photorealistic forest scenes can be generated, at large scale, with a limited set of tree models, using procedural foliage generation. All forest scenes we have collected from Unreal Engine were generated using this algorithm. Although developed and used mainly in the game industry, this algorithm has roots in ecological theory on forest growth [Pacala et al., 1996, Purves et al., 2007]. With our generatio… view at source ↗
Figure 4
Figure 4. Figure 4: Synthetic data generation pipeline, written in Python, and fully automated to execute each step in this figure. The pipeline processes a forest scene from Unreal Engine by (a) adding files to Blender, (b) customizing instance and semantic labels, (c) planning the UAV flight path, setting up a virtual laser scanner, and (d) structuring files for LiDAR survey. It completes post-processing to produce a machin… view at source ↗
Figure 5
Figure 5. Figure 5: Our virtual survey resembles a real world UAV mission. The RIEGL VUX-1 UAV scanner is widely used. For Redwood scene we set the Relative Altitude as 120m because of the tree’s large size. All parameters are customizable in our data pipeline. Parameter Value Scanner Model RIEGL VUX-1 UAV Pulse Frequency 100 kHz Max Returns (100 kHz) 15 Flight Pattern Criss-cross Flight Spacing 20.0 m Flight Speed 5.0 m/s Re… view at source ↗
Figure 6
Figure 6. Figure 6: Synthetic data covers 12 forest scenes for a wide range of forest types, totalling 75 hectares. Forest scenes in the first block are adapted from the SPREAD dataset [Feng et al., 2025]. The generated data is tiled into roughly 50m*50m plots and split into train, test, and validation sets. Scene Name Tree Species Plots points/m² Train Val Test Deciduous1 Birch 17 4 4 2092.687 Deciduous2 Beech, Oak 17 4 4 14… view at source ↗
Figure 7
Figure 7. Figure 7: The synthetic data is diverse, physics-based, and available at large scale. Using Unreal Engine, large-scale, photorealistic forest scenes are created using gaming assets and procedural foliage. Then the pipeline simulates laser scans to produce synthetic point clouds with instance labels (shown here in colours). The 50m x 50m tiled plots are ready for machine learning. 8 [PITH_FULL_IMAGE:figures/full_fig… view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the FOR-Instance data set from [Puliti et al., 2023]. NIBIO2 is not part of the FOR-Instance dataset but is included in the training set in [Xiang et al., 2024]. Note that RMIT and TUWIEN are deciduous forests and the rest are coniferous forests. Except for NIBIO and NIBIO2, the rest of the regions only have 1 or 2 training plots available. Scene Name Original Name Tree Species Plots Point Dens… view at source ↗
Figure 9
Figure 9. Figure 9: Annotating the FOR-Instance dataset [Puliti et al., 2023] took two skilled annotators six months, in addition to the time spent on data collection. It covers 3 coniferous scenes and 2 deciduous scenes totalling 2.79 hectares, and Coniferous-N2 is a supplementary dataset fromXiang et al. [2024]. Annotated plots from each region are illustrated. FOR-Instance contains both instance and semantic segmentation l… view at source ↗
Figure 10
Figure 10. Figure 10: The panoptic tree segmentation algorithm [Xiang et al., 2024] A adapted for this study. Panoptic segmentation addresses instance and semantic segmentation simultaneously. A uses the Minkowski Engine, a 3D CNN, as its backbone followed by semantic and instance branches. In our experiments, we simplify semantic prediction from five classes to binary: tree and non-tree. To train the model, 8m radius cylinder… view at source ↗
Figure 11
Figure 11. Figure 11: Real datasets used in few shot learning experiments. The subscripts denote the number of scenes and the number of plots. Besides the full dataset Real6,42, we organised few-shot subsets for coniferous (Real-C) and deciduous forests (Real-D). We report cylinder sample counts from each subset and their percentages of the full dataset. Dataset Scenes Plots Samples Percentage (%) Real6,42 Coniferous-N, Conife… view at source ↗
Figure 12
Figure 12. Figure 12: Pretraining on synthetic data, and using minimal real data fine-tuning, performs competitively with a much larger, labelled, real dataset. F1 scores improve significantly for small, labelled coniferous regions when synthetic (mixed deciduous/coniferous) data is available for pre-training. Deciduous plots show a similar pattern, with the same pretraining data. Test Pretrain Train/Tune Experiment Real Data … view at source ↗
Figure 13
Figure 13. Figure 13: Pretraining on synthetic data enables the algorithm to learn representations like those from real data, with fine-tuning on just a single forest plot. We present a sample from Coniferous-N, where embeddings and points shifted by offsets serve as the foundation, for instance segmentation. Both embeddings and points are colourized with ground-truth labels and projected in 2D using UMAP [McInnes et al., 2018… view at source ↗
Figure 20
Figure 20. Figure 20: ) Even extreme few-shot training, using only 3.1% of the samples, is competitive, dropping [PITH_FULL_IMAGE:figures/full_fig_p012_20.png] view at source ↗
Figure 14
Figure 14. Figure 14: Synthetic datasets in our ablation study analyze three factors: simulation physics (Nodal4,20 vs. Sim4,20), scene diversity (Sim4,20 vs. Sim11,55), and dataset scale (Sim11,55 vs. Sim11,187).Nodal points derive from tree mesh nodes, while Sim points come from LiDAR simu￾lations. Both Sim4,20 and Sim11,55 cover 30% of available plots (5 plots per scene) from their scenes. Dataset Scenes Plots LiDAR Nodal4,… view at source ↗
Figure 15
Figure 15. Figure 15: Nodal points from tree meshes provide a simple way to obtain point clouds. A large ash tree model from the Deciduous3 scene is shown. A rendered tree model consists of textures and a static mesh. The static mesh, made of vertices and edges, directly yields nodal points as point clouds. LiDAR simulation point clouds are based on UAV laser scans and are denser than these nodal points. 13 [PITH_FULL_IMAGE:f… view at source ↗
Figure 16
Figure 16. Figure 16: shows the results for Coniferous-C. More detail can be found in [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Synthetic data helps the algorithm learn disentangled representations. A Coniferous-C data sample with embeddings and offset-shifted points, is colorized with ground truth labels and projected into 2D using UMAP [McInnes et al., 2018]. Compared with training at nodal points, disentanglement is enhanced as we add: LiDAR simulation; diverse scenes, especially newly included coniferous ones; more simulated d… view at source ↗
Figure 18
Figure 18. Figure 18: Our model trained on mixed data achieves results comparable to state-of-the￾art, without exceeding it. F1-scores for mixed training, results in Xiang et al. [2024], and our reproduction of Xiang et al. [2024], are similar across regions. Deciduous-T is an exception where both our result and the reproduction are notably lower than reported by Xiang et al. [2024]. Metrics are computed at the plot level by a… view at source ↗
Figure 19
Figure 19. Figure 19: A structurally diverse forest scene can be created using a small set of tree models. The Deciduous3 composition uses 2,053 tree meshes within a 250m x 250m plot, derived from 17 basic meshes with various canopy levels, as shown in this table. Mesh Name Count Percentage (%) TreesAshLarge_A 60 2.92 TreesAshLarge_B 43 2.09 TreesAshLarge_C 42 2.05 TreesAshMedium_B 343 16.71 TreesAshMedium_C 180 8.77 TreesAshS… view at source ↗
Figure 20
Figure 20. Figure 20: The algorithm pretrained on our synthetic data can achieve competitive performance, with fine-tuning only on a few real forest plots. This table presents the results for coniferous regions. The improvement of F1 score is significant compared to training on these few-shot dataset from scratch, indicating that our synthetic data contains useful prior knowledge for the algorithm to learn in order to generali… view at source ↗
Figure 22
Figure 22. Figure 22: Three factors plays significant roles in generating strong synthetic data from our pipeline, as revealed in the zero-shot experiment. (1) Physics-based simulation, by comparing improvement of F1-score from the algorithm trained on LiDAR simulation (Sim4,20) against nodal points from tree meshes Nodal4,20. (2) More diverse scenes, by comparing the improvement from training on Sim4,20 against Sim11,55. Both… view at source ↗
Figure 23
Figure 23. Figure 23: shows the extended dataset with leaf and wood labels, the most fine-grained annotations achievable with current tree assets. These labels enable the leaf–wood separation task, which future researchers can use to train forest vision systems for quantitative studies of forest structural dynamics [PITH_FULL_IMAGE:figures/full_fig_p025_23.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

64 extracted references · 9 linked inside Pith

  1. [1]

    The future of tropical forests under the united nations sustainable development goals

    Latha Swamy, Erika Drazen, Whitney R Johnson, and Jacob J Bukoski. The future of tropical forests under the united nations sustainable development goals. Journal of sustainable forestry, 37 0 (2): 0 221--256, 2018

    2018

  2. [2]

    The shape of trees: Reimagining forest ecology in three dimensions with remote sensing

    Emily Rebecca Lines, Fabian J \"o rg Fischer, Harry Jon Foord Owen, and Tommaso Jucker. The shape of trees: Reimagining forest ecology in three dimensions with remote sensing. Journal of Ecology, 110 0 (8): 0 1730--1745, 2022

    2022

  3. [3]

    Close-range remote sensing of forests: The state of the art, challenges, and opportunities for systems and data acquisitions

    Xinlian Liang, Antero Kukko, Ivan Balenovi \'c , Ninni Saarinen, Samuli Junttila, Ville Kankare, Markus Holopainen, Martin Mokro s , Peter Surov \`y , Harri Kaartinen, et al. Close-range remote sensing of forests: The state of the art, challenges, and opportunities for systems and data acquisitions. IEEE Geoscience and Remote Sensing Magazine, 10 0 (3): 0...

    2022

  4. [4]

    The global ecosystem dynamics investigation: High-resolution laser ranging of the earth’s forests and topography

    Ralph Dubayah, James Bryan Blair, Scott Goetz, Lola Fatoyinbo, Matthew Hansen, Sean Healey, Michelle Hofton, George Hurtt, James Kellner, Scott Luthcke, et al. The global ecosystem dynamics investigation: High-resolution laser ranging of the earth’s forests and topography. Science of remote sensing, 1: 0 100002, 2020

    2020

  5. [5]

    Global canopy height regression and uncertainty estimation from gedi lidar waveforms with deep ensembles

    Nico Lang, Nikolai Kalischek, John Armston, Konrad Schindler, Ralph Dubayah, and Jan Dirk Wegner. Global canopy height regression and uncertainty estimation from gedi lidar waveforms with deep ensembles. Remote Sensing of Environment, 268: 0 112760, 2022

    2022

  6. [6]

    Area-based vs tree-centric approaches to mapping forest carbon in southeast asian forests from airborne laser scanning data

    David A Coomes, Michele Dalponte, Tommaso Jucker, Gregory P Asner, Lindsay F Banin, David FRP Burslem, Simon L Lewis, Reuben Nilus, Oliver L Phillips, Mui-How Phua, et al. Area-based vs tree-centric approaches to mapping forest carbon in southeast asian forests from airborne laser scanning data. Remote Sensing of Environment, 194: 0 77--88, 2017

    2017

  7. [7]

    Benchmarking tree species classification from proximally-sensed laser scanning data: introducing the for-species20k dataset

    Stefano Puliti, Emily R Lines, Jana M \"u llerov \'a , Julian Frey, Zoe Schindler, Adrian Straker, Matthew J Allen, Lukas Winiwarter, Nataliia Rehush, Hristina Hristova, et al. Benchmarking tree species classification from proximally-sensed laser scanning data: introducing the for-species20k dataset. arXiv preprint arXiv:2408.06507, 2024

    Pith/arXiv arXiv 2024

  8. [8]

    Monitoring ash dieback (hymenoscyphus fraxineus) in british forests using hyperspectral remote sensing

    Aland HY Chan, Chloe Barnes, Tom Swinfield, and David A Coomes. Monitoring ash dieback (hymenoscyphus fraxineus) in british forests using hyperspectral remote sensing. Remote Sensing in Ecology and Conservation, 7 0 (2): 0 306--320, 2021

    2021

  9. [9]

    o r \"a l \

    Xinlian Liang, Yunsheng Wang, Jiri Py \"o r \"a l \"a , Matti Lehtom \"a ki, Xiaowei Yu, Harri Kaartinen, Antero Kukko, Eija Honkavaara, Aimad EI Issaoui, Olli Nevalainen, et al. Forest in situ observations using unmanned aerial vehicle as an alternative of terrestrial measurements. Forest ecosystems, 6 0 (1): 0 1--16, 2019

    2019

  10. [10]

    Towards a worldwide wood economics spectrum

    Jerome Chave, David Coomes, Steven Jansen, Simon L Lewis, Nathan G Swenson, and Amy E Zanne. Towards a worldwide wood economics spectrum. Ecology letters, 12 0 (4): 0 351--366, 2009

    2009

  11. [11]

    Crown plasticity and competition for canopy space: a new spatially implicit model parameterized for 250 north american tree species

    Drew W Purves, Jeremy W Lichstein, and Stephen W Pacala. Crown plasticity and competition for canopy space: a new spatially implicit model parameterized for 250 north american tree species. PloS one, 2 0 (9): 0 e870, 2007

    2007

  12. [12]

    Global atmospheric methane uptake by upland tree woody surfaces

    Vincent Gauci, Sunitha Rao Pangala, Alexander Shenkin, Josep Barba, David Bastviken, Viviane Figueiredo, Carla Gomez, Alex Enrich-Prast, Emma Sayer, Tain \'a Stauffer, et al. Global atmospheric methane uptake by upland tree woody surfaces. Nature, 631 0 (8022): 0 796--800, 2024

    2024

  13. [13]

    Instance segmentation of individual tree crowns with yolov5: A comparison of approaches using the forinstance benchmark lidar dataset

    Adrian Straker, Stefano Puliti, Johannes Breidenbach, Christoph Kleinn, Grant Pearse, Rasmus Astrup, and Paul Magdon. Instance segmentation of individual tree crowns with yolov5: A comparison of approaches using the forinstance benchmark lidar dataset. ISPRS Open Journal of Photogrammetry and Remote Sensing, 9: 0 100045, 2023

    2023

  14. [14]

    Accurate tropical forest individual tree crown delineation from aerial rgb imagery using mask r-cnn

    James GC Ball, Sebastian HM Hickman, Tobias D Jackson, Xian Jing Koay, James Hirst, William Jay, M \'e laine Aubry-Kientz, Gr \'e goire Vincent, and David A Coomes. Accurate tropical forest individual tree crown delineation from aerial rgb imagery using mask r-cnn. bioRxiv, pages 2022--07, 2022

    2022

  15. [15]

    Point2tree (p2t)—framework for parameter tuning of semantic and instance segmentation used with mobile laser scanning data in coniferous forest

    Maciej Wielgosz, Stefano Puliti, Phil Wilkes, and Rasmus Astrup. Point2tree (p2t)—framework for parameter tuning of semantic and instance segmentation used with mobile laser scanning data in coniferous forest. Remote Sensing, 15 0 (15): 0 3737, 2023

    2023

  16. [16]

    Automated forest inventory: analysis of high-density airborne lidar point clouds with 3d deep learning

    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: 0 114078, 2024

    2024

  17. [17]

    Segmentanytree: A sensor and platform agnostic deep learning model for tree segmentation using laser scanning data

    Maciej Wielgosz, Stefano Puliti, Binbin Xiang, Konrad Schindler, and Rasmus Astrup. Segmentanytree: A sensor and platform agnostic deep learning model for tree segmentation using laser scanning data. arXiv preprint arXiv:2401.15739, 2024

    Pith/arXiv arXiv 2024

  18. [18]

    The bitter lesson

    Richard Sutton. The bitter lesson. Incomplete Ideas (blog), 13 0 (1): 0 38, 2019

    2019

  19. [19]

    a fer, Lukas Winiwarter, Nina Kra s ovec, Fabian E Fassnacht, and Bernhard H \

    Hannah Weiser, Jannika Sch \"a fer, Lukas Winiwarter, Nina Kra s ovec, Fabian E Fassnacht, and Bernhard H \"o fle. Individual tree point clouds and tree measurements from multi-platform laser scanning in german forests. Earth System Science Data, 14 0 (7): 0 2989--3012, 2022

    2022

  20. [20]

    Laser scanning reveals potential underestimation of biomass carbon in temperate forest

    Kim Calders, Hans Verbeeck, Andrew Burt, Niall Origo, Joanne Nightingale, Yadvinder Malhi, Phil Wilkes, Pasi Raumonen, Robert GH Bunce, and Mathias Disney. Laser scanning reveals potential underestimation of biomass carbon in temperate forest. Ecological Solutions and Evidence, 3 0 (4): 0 e12197, 2022

    2022

  21. [21]

    For-instance: a uav laser scanning benchmark dataset for semantic and instance segmentation of individual trees

    Stefano Puliti, Grant Pearse, Peter Surov \`y , Luke Wallace, Markus Hollaus, Maciej Wielgosz, and Rasmus Astrup. For-instance: a uav laser scanning benchmark dataset for semantic and instance segmentation of individual trees. arXiv preprint arXiv:2309.01279, 2023

    Pith/arXiv arXiv 2023

  22. [22]

    Dataset meta-level and statistical features affect machine learning performance

    Shahadat Uddin and Haohui Lu. Dataset meta-level and statistical features affect machine learning performance. Scientific Reports, 14 0 (1): 0 1670, 2024

    2024

  23. [23]

    Real-time human pose recognition in parts from single depth images

    Jamie Shotton, Andrew Fitzgibbon, Mat Cook, Toby Sharp, Mark Finocchio, Richard Moore, Alex Kipman, and Andrew Blake. Real-time human pose recognition in parts from single depth images. In CVPR 2011, pages 1297--1304. Ieee, 2011

    2011

  24. [24]

    Haar: Text-conditioned generative model of 3d strand-based human hairstyles

    Vanessa Sklyarova, Egor Zakharov, Otmar Hilliges, Michael J Black, and Justus Thies. Haar: Text-conditioned generative model of 3d strand-based human hairstyles. arXiv preprint arXiv:2312.11666, 2023

    Pith/arXiv arXiv 2023

  25. [25]

    Virtual worlds as proxy for multi-object tracking analysis

    Adrien Gaidon, Qiao Wang, Yohann Cabon, and Eleonora Vig. Virtual worlds as proxy for multi-object tracking analysis. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4340--4349, 2016

    2016

  26. [26]

    Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding

    Mike Roberts, Jason Ramapuram, Anurag Ranjan, Atulit Kumar, Miguel Angel Bautista, Nathan Paczan, Russ Webb, and Joshua M Susskind. Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10912--10922, 2021

    2021

  27. [27]

    Synthetic datasets for autonomous driving: A survey

    Zhihang Song, Zimin He, Xingyu Li, Qiming Ma, Ruibo Ming, Zhiqi Mao, Huaxin Pei, Lihui Peng, Jianming Hu, Danya Yao, et al. Synthetic datasets for autonomous driving: A survey. IEEE Transactions on Intelligent Vehicles, 9 0 (1): 0 1847--1864, 2023

    2023

  28. [28]

    Unreal engine environment generation (pcg): a comparative overview of existing tools

    Nora Holmberg. Unreal engine environment generation (pcg): a comparative overview of existing tools. 2025

    2025

  29. [29]

    Tree detection and diameter estimation based on deep learning

    Vincent Grondin, Jean-Michel Fortin, François Pomerleau, and Philippe Giguère. Tree detection and diameter estimation based on deep learning. Forestry: An International Journal of Forest Research, 96 0 (2): 0 264–276, October 2022. ISSN 1464-3626. doi:10.1093/forestry/cpac043. URL http://dx.doi.org/10.1093/forestry/cpac043

    doi:10.1093/forestry/cpac043 2022

  30. [30]

    M2fnet: Multi-modal forest monitoring network on large-scale virtual dataset

    Yawen Lu, Yunhan Huang, Su Sun, Tansi Zhang, Xuewen Zhang, Songlin Fei, and Victor Chen. M2fnet: Multi-modal forest monitoring network on large-scale virtual dataset. arXiv preprint arXiv:2402.04534, 2024

    Pith/arXiv arXiv 2024

  31. [31]

    Spread: A large-scale, high-fidelity synthetic dataset for multiple forest vision tasks

    Zhengpeng Feng, Yihang She, and Srinivasan Keshav. Spread: A large-scale, high-fidelity synthetic dataset for multiple forest vision tasks. Ecological Informatics, 87: 0 103085, 2025

    2025

  32. [32]

    Infinite nature: Perpetual view generation of natural scenes from a single image

    Andrew Liu, Richard Tucker, Varun Jampani, Ameesh Makadia, Noah Snavely, and Angjoo Kanazawa. Infinite nature: Perpetual view generation of natural scenes from a single image. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 14458--14467, 2021

    2021

  33. [33]

    Virtual laser scanning with helios++: A novel take on ray tracing-based simulation of topographic full-waveform 3d laser scanning

    Lukas Winiwarter, Alberto Manuel Esmor \' s Pena, Hannah Weiser, Katharina Anders, Jorge Mart \' nez S \'a nchez, Mark Searle, and Bernhard H \"o fle. Virtual laser scanning with helios++: A novel take on ray tracing-based simulation of topographic full-waveform 3d laser scanning. Remote Sensing of Environment, 269: 0 112772, 2022

    2022

  34. [34]

    Unsupervised semantic and instance segmentation of forest point clouds

    Di Wang. Unsupervised semantic and instance segmentation of forest point clouds. ISPRS Journal of Photogrammetry and Remote Sensing, 165: 0 86--97, 2020

    2020

  35. [35]

    Domain adaptation of deep neural networks for tree part segmentation using synthetic forest trees

    Mitch Bryson, Ahalya Ravendran, Celine Mercier, Tancred Frickey, Sadeepa Jayathunga, Grant Pearse, and Robin JL Hartley. Domain adaptation of deep neural networks for tree part segmentation using synthetic forest trees. ISPRS Open Journal of Photogrammetry and Remote Sensing, 14: 0 100078, 2024

    2024

  36. [36]

    Domain randomization for transferring deep neural networks from simulation to the real world

    Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Wojciech Zaremba, and Pieter Abbeel. Domain randomization for transferring deep neural networks from simulation to the real world. In 2017 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 23--30. IEEE, 2017

    2017

  37. [37]

    Structured domain randomization: Bridging the reality gap by context-aware synthetic data

    Aayush Prakash, Shaad Boochoon, Mark Brophy, David Acuna, Eric Cameracci, Gavriel State, Omer Shapira, and Stan Birchfield. Structured domain randomization: Bridging the reality gap by context-aware synthetic data. In 2019 International Conference on Robotics and Automation (ICRA), pages 7249--7255. IEEE, 2019

    2019

  38. [38]

    Model-agnostic meta-learning for fast adaptation of deep networks

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In International conference on machine learning, pages 1126--1135. PMLR, 2017

    2017

  39. [39]

    Advancing the understanding of fine-grained 3d forest structures using digital cousins and simulation-to-reality: Methods and datasets

    Jing Liu, Duanchu Wang, Haoran Gong, Chongyu Wang, Jihua Zhu, and Di Wang. Advancing the understanding of fine-grained 3d forest structures using digital cousins and simulation-to-reality: Methods and datasets. arXiv preprint arXiv:2501.03637, 2025

    Pith/arXiv arXiv 2025

  40. [40]

    A high-resolution canopy height model of the earth

    Nico Lang, Walter Jetz, Konrad Schindler, and Jan Dirk Wegner. A high-resolution canopy height model of the earth. Nature Ecology & Evolution, pages 1--12, 2023

    2023

  41. [41]

    3d modeling and reconstruction of plants and trees: A cross-cutting review across computer graphics, vision, and plant phenotyping

    Fumio Okura. 3d modeling and reconstruction of plants and trees: A cross-cutting review across computer graphics, vision, and plant phenotyping. Breeding Science, 72 0 (1): 0 31--47, 2022

    2022

  42. [42]

    Enhancing uav--sfm 3d model accuracy in high-relief landscapes by incorporating oblique images

    Paul Ryan Nesbit and Christopher H Hugenholtz. Enhancing uav--sfm 3d model accuracy in high-relief landscapes by incorporating oblique images. Remote Sensing, 11 0 (3): 0 239, 2019

    2019

  43. [43]

    Technological advances in close range sensors and methodological complexities for measuring forest structure and disturbance

    W Flynn. Technological advances in close range sensors and methodological complexities for measuring forest structure and disturbance. 2024

    2024

  44. [44]

    Towards accurate instance segmentation in large-scale lidar point clouds

    Binbin Xiang, Torben Peters, Theodora Kontogianni, Frawa Vetterli, Stefano Puliti, Rasmus Astrup, and Konrad Schindler. Towards accurate instance segmentation in large-scale lidar point clouds. arXiv preprint arXiv:2307.02877, 2023

    Pith/arXiv arXiv 2023

  45. [45]

    a , Harri Kaartinen, Matti Lehtom \

    Xinlian Liang, Juha Hyypp \"a , Harri Kaartinen, Matti Lehtom \"a ki, Jiri Py \"o r \"a l \"a , Norbert Pfeifer, Markus Holopainen, G \'a bor Brolly, Pirotti Francesco, Jan Hackenberg, et al. International benchmarking of terrestrial laser scanning approaches for forest inventories. ISPRS journal of photogrammetry and remote sensing, 144: 0 137--179, 2018

    2018

  46. [46]

    Large-area virtual forests from terrestrial laser scanning data

    Kim Calders, Andrew Burt, Niall Origo, Mathias Disney, J Nightingale, Pasi Raumonen, and Philip Lewis. Large-area virtual forests from terrestrial laser scanning data. In 2016 IEEE international geoscience and remote sensing symposium (IGARSS), pages 1765--1767. IEEE, 2016

    2016

  47. [47]

    Carla: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. Carla: An open urban driving simulator. In Conference on robot learning, pages 1--16. PMLR, 2017

    2017

  48. [48]

    Airsim-w: A simulation environment for wildlife conservation with uavs

    Elizabeth Bondi, Debadeepta Dey, Ashish Kapoor, Jim Piavis, Shital Shah, Fei Fang, Bistra Dilkina, Robert Hannaford, Arvind Iyer, Lucas Joppa, et al. Airsim-w: A simulation environment for wildlife conservation with uavs. In Proceedings of the 1st ACM SIGCAS Conference on Computing and Sustainable Societies, pages 1--12, 2018

    2018

  49. [49]

    Deep learning with simulated laser scanning data for 3d point cloud classification

    Alberto M Esmor \' s, Hannah Weiser, Lukas Winiwarter, Jose C Cabaleiro, and Bernhard H \"o fle. Deep learning with simulated laser scanning data for 3d point cloud classification. ISPRS Journal of Photogrammetry and Remote Sensing, 215: 0 192--213, 2024

    2024

  50. [50]

    a fer, Hannah Weiser, Lukas Winiwarter, Bernhard H \

    Jannika Sch \"a fer, Hannah Weiser, Lukas Winiwarter, Bernhard H \"o fle, Sebastian Schmidtlein, and Fabian Ewald Fassnacht. Generating synthetic laser scanning data of forests by combining forest inventory information, a tree point cloud database and an open-source laser scanning simulator. Forestry: An International Journal of Forest Research, 96 0 (5):...

    2023

  51. [51]

    Implications of 3d forest stand reconstruction methods for radiative transfer modeling: A case study in the temperate deciduous forest

    Chang Liu, Kim Calders, F \'e licien Meunier, JP Gastellu-Etchegorry, J Nightingale, M Disney, N Origo, W Woodgate, and Hans Verbeeck. Implications of 3d forest stand reconstruction methods for radiative transfer modeling: A case study in the temperate deciduous forest. Journal of Geophysical Research: Atmospheres, 127 0 (14): 0 e2021JD036175, 2022

    2022

  52. [52]

    Essay on the architecture and dynamics of growth of tropical trees

    Francis Hall \'e and Roelof AA Oldeman. Essay on the architecture and dynamics of growth of tropical trees. 1970

    1970

  53. [53]

    Forest models defined by field measurements: estimation, error analysis and dynamics

    Stephen W Pacala, Charles D Canham, John Saponara, John A Silander Jr, Richard K Kobe, and Eric Ribbens. Forest models defined by field measurements: estimation, error analysis and dynamics. Ecological monographs, 66 0 (1): 0 1--43, 1996

    1996

  54. [54]

    Semantic classification in uncolored 3d point clouds using multiscale features

    Michael Neumann, Dorit Borrmann, and Andreas N \"u chter. Semantic classification in uncolored 3d point clouds using multiscale features. In International Conference on Intelligent Autonomous Systems, pages 342--359. Springer, 2022

    2022

  55. [55]

    Pointgroup: Dual-set point grouping for 3d instance segmentation

    Li Jiang, Hengshuang Zhao, Shaoshuai Shi, Shu Liu, Chi-Wing Fu, and Jiaya Jia. Pointgroup: Dual-set point grouping for 3d instance segmentation. In Proceedings of the IEEE/CVF conference on computer vision and Pattern recognition, pages 4867--4876, 2020

    2020

  56. [56]

    4d spatio-temporal convnets: Minkowski convolutional neural networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4d spatio-temporal convnets: Minkowski convolutional neural networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3075--3084, 2019

    2019

  57. [57]

    Umap: Uniform manifold approximation and projection for dimension reduction

    Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426, 2018

    Pith/arXiv arXiv 2018

  58. [58]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk \"u hler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 42 0 (4): 0 1--14, 2023

    2023

  59. [59]

    Towards multimodal open-set domain generalization and adaptation through self-supervision

    Hao Dong, Eleni Chatzi, and Olga Fink. Towards multimodal open-set domain generalization and adaptation through self-supervision. In European Conference on Computer Vision, pages 270--287. Springer, 2024

    2024

  60. [60]

    Multimodal cross-domain few-shot learning for egocentric action recognition

    Masashi Hatano, Ryo Hachiuma, Ryo Fujii, and Hideo Saito. Multimodal cross-domain few-shot learning for egocentric action recognition. In European Conference on Computer Vision, pages 182--199. Springer, 2024

    2024

  61. [61]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4840--4851, 2024

    2024

  62. [62]

    Forestformer3d: A unified framework for end-to-end segmentation of forest lidar 3d point clouds

    Binbin Xiang, Maciej Wielgosz, Stefano Puliti, Kamil Kr \'a l, Martin Kr u c ek, Azim Missarov, and Rasmus Astrup. Forestformer3d: A unified framework for end-to-end segmentation of forest lidar 3d point clouds. arXiv preprint arXiv:2506.16991, 2025

    Pith/arXiv arXiv 2025

  63. [63]

    Scaling vision transformers

    Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, and Lucas Beyer. Scaling vision transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12104--12113, 2022

    2022

  64. [64]

    Weighing trees with lasers: advances, challenges and opportunities

    Mathias I Disney, Matheus Boni Vicari, Andrew Burt, Kim Calders, Simon L Lewis, Pasi Raumonen, and Phil Wilkes. Weighing trees with lasers: advances, challenges and opportunities. Interface Focus, 8 0 (2): 0 20170048, 2018

    2018