Three LiDARs, Twenty-Five Datasets: Cross-Source Generalization in Tree-Species Recognition

Jae Joong Lee; Bedrich Benes

cs.CVOct 15, 2026

Three LiDARs, Twenty-Five Datasets: Cross-Source Generalization in Tree-Species Recognition

Authors: Jae Joong Lee, Bedrich Benes

Organizations: Purdue University, Department of Computer Science 305 N University St, West Lafayette, IN 47907, USA

Abstract

FOR-species20K is the largest public LiDAR benchmark for tree-species classification, pooling 25 independently collected terrestrial, mobile, and airborne laser-scanning datasets across 33 species. We introduce a multi-resolution, multi-view ConvNeXtV2 model that improves official hidden-test accuracy from 79.5% to 85.0%, ranking first at the time of submission. We then examine what this score measures. Because the hidden test is species-balanced, its overall accuracy is more comparable to a macro metric than to accuracy under the public set's long-tailed distribution. Moreover, the official random split places every source dataset in both training and evaluation. A logistic probe recovers source identity from out-of-fold model probabilities with 69% accuracy, while geometric features identify the source above 90% within two widely shared species; 18 of the 33 species occur in only one source. Under a source-species-cell holdout, evaluated on 5,566 trees from eight eligible species that retain the same LiDAR-type training support in other sources, the unweighted species mean falls from 82% to 36%, and tree-weighted accuracy falls from 95% to 60%. Count-matched retraining and experiments with two alternative backbones indicate that this gap is not explained by reduced training support alone, while an external PureForest experiment shows a smaller but consistent grouped-split effect. These results establish the 85% leaderboard score as strong in-distribution performance, but not as an estimate of source-disjoint generalization, motivating balanced metrics and source-aware evaluation. The code is available at: https://github.com/JaeLee18/FOR20K-lidar-tree-species

Figures & tables

Figure 1: Composition of FOR-species20K. (a) Trees per source dataset, colored by LiDAR type. (b) Species frequency on a log scale, with red markers for species that appear in only one dataset.

Figure 2: The 33 species of FOR-species20K. A labeled-set tree is shown per species as a raw point cloud with height-coded color.

Figure 3: Source-species holdout, one fold: 25 sources (held-out group outlined) $\times$ 33 species sorted by source count (divider before the 18 single-source species). Blue = training, red = held out and scored, dark = hidden sources’ single-source species.

Figure 4: Overview of the multi-view ConvNeXtV2. Seven orthographic renders are split into three view groups. Each group is processed by one ConvNeXtV2-Base backbone [ 51 ] , followed by group max-pooling, concatenation with tree height, and a classification head.

Method	Input	OA	Prec.	Rec.	F1
DetailView [ 11 ]	Img	79.5	82.3	76.7	78.0
YOLOv5	Img	77.9	84.2	75.0	77.3
SimpleView	Img	76.2	76.9	75.5	75.6
Ensemble PointNet++ [ 38 ]	PC	75.6	78.2	73.5	74.9
MinkNet [ 7 ]	PC	73.7	79.9	70.6	72.3
PointMixer	PC	71.1	74.4	65.5	71.1

Table 1: Official FOR-species20K leaderboard [ 34 ] and our multi-resolution ConvNeXtV2 method. All values are percentages, and higher is better. Bold marks the best score in each column.

Backbone	Rend.	OA	macro-F1	TLS	MLS	ULS
DenseNet-201 $\times$ 3 (baseline)	256	0.818	0.761	0.814	0.860	0.816
DINOv2-Base $\times$ 3	224	0.819	0.758	0.805	0.894	0.836
ConvNeXtV2-Base $\times$ 3 (ours)	256	0.886	0.843	0.879	0.934	0.894
ConvNeXtV2-Base $\times$ 3 (ours)	512	0.894	0.855	0.888	0.928	0.900
+ multi-resolution ensemble	256 $+$ 512	0.899	0.860	0.893	0.936	0.903

Table 2: Method ablation on the public labeled set with five-fold out-of-fold scores. Higher is better. The bottom rows show our ConvNeXtV2-Base models and the multi-resolution ensemble. Bold marks the best public labeled-set score. These scores are not directly comparable to the species-balanced hidden test in Table 1 .

Species	$n$	random	LSO	$\Delta$
Acer_pseudoplatanus	24	0.79	0.00	+0.79
Betula_pendula	210	0.91	0.25	+0.66
Fagus_sylvatica	44	0.93	0.34	+0.59
Quercus_robur	182	0.92	0.36	+0.57
Picea_abies	1983	0.95	0.55	+0.40
Pinus_sylvestris	3048	0.96	0.69	+0.27

Table 3: Cross-source accuracy drops on the same trees that are accurate under the random split. Each species keeps at least 50 training trees from the same LiDAR type outside the held-out dataset and has at least 20 held-out trees. The table reports random-split accuracy and accuracy after the dataset is held out on the same trees. A larger

\Delta

means a larger cross-source drop. Bold marks the pooled overall.

Figure 5: Cross-source evaluation exposes the accuracy hidden by the random split. Light markers show random-split accuracy on the same trees. Solid markers show accuracy after the dataset is held out. Rows are colored by LiDAR type and sorted by the number of training trees from the same LiDAR type outside the held-out dataset. Every row shifts left, even when the species remain in training.

Figure 6: Row-normalized confusion matrix among the co-occurring deciduous broadleaves. Rows are true species, and columns are predicted species.

Hyperparameter	Value
Backbone	ConvNeXtV2-Base (FCMAE pretrained)
Backbones	3 (one per view group)
Total params	264.7M
Loss	Logit-adjusted CE ( $\tau=1.0$ )
Optimizer	AdamW
Backbone LR	$3\times 10^{-5}$

Table A1: Training hyperparameters for the ConvNeXtV2 model.

Model	OA	Bal. acc.	Macro-F1	OA $-$ Macro-F1
ConvNeXtV2 (image)	0.886	0.846	0.843	+0.043
DenseNet-201 (image)	0.818	0.819	0.761	+0.057
DINOv2-Base (image)	0.819	0.759	0.758	+0.061
PointNet++ (point cloud)	0.573	0.433	0.460	+0.113
DGCNN (point cloud)	0.626	0.391	0.427	+0.199

Table A2: Natural-distribution and macro metrics on the public development set, using pooled five-fold out-of-fold predictions.

Source dataset	Plat.	$n$	Source acc.
Saarinen_2021	TLS	1958	0.999
Xi_2020b	TLS	278	0.993
FORinstance_SCION	ULS	97	0.990
Luck	TLS	280	0.982
Xi_2020a	TLS	488	0.980
BlueCat_a	TLS	3474	0.973

Table A3: Source-recognition accuracy per source dataset, pooled per platform, and overall by backbone, from a logistic source classifier on the model’s output probabilities. Chance is

1/25=0.040

and higher is better; bold marks the image-modality reference.

Fold	$n_{\text{val}}$	OA	macro-F1	TLS	MLS	ULS
0	3542	0.8905	0.8488	0.8783	0.9486	0.9076
1	3542	0.8862	0.8466	0.8805	0.9301	0.8890
2	3541	0.8870	0.8386	0.8775	0.9233	0.9048
3	3541	0.8820	0.8387	0.8804	0.9156	0.8735
4	3541	0.8865	0.8405	0.8770	0.9532	0.8929
Mean	–	0.8864	0.8427	0.8787	0.9342	0.8936

Table A4: ConvNeXtV2 out-of-fold accuracy for each fold. Higher is better, and bold marks the five-fold mean. Internal figures, not comparable to the balanced hidden test.

Species	Support	F1		$\Delta$
		DenseNet	ConvNeXtV2
P. sylvestris	3296	0.895	0.964	+0.069
F. sylvatica	2482	0.826	0.923	+0.098
P. abies	1983	0.924	0.951	+0.027
C. betulus	1243	0.648	0.720	+0.072
A. campestre	1240	0.743	0.791	+0.048

Table A5: Per-species F1 (higher is better) for DenseNet-201 and ConvNeXtV2-Base, sorted by training support.

\Delta

= ConvNeXtV2

-

DenseNet. Bold

\Delta

indicates a gain

>0.05

Figure A1 : Per-species F1 for DenseNet-201 (gray) and ConvNeXtV2 (blue), sorted by support (rarest at bottom). Dashed lines mark each model’s mean F1. Annotated counts on the right show training support for the rarest ten species.

Figure A2 : Row-normalized confusion matrix for all 33 species (5-fold OOF, $n{=}17{,}707$ ). Rows are the true species and columns the predicted; each cell is the within-row prediction rate. Cells $\geq 0.05$ are annotated. Species are in the same order as Table A5 .

$S\geq$	Cells	Patches	Random	Grouped	Drop
0	40	22993	0.936	0.815	+0.121
100	39	22276	0.942	0.841	+0.101
200	38	21974	0.941	0.846	+0.095
400	37	21877	0.942	0.848	+0.094
800	27	21211	0.946	0.854	+0.092

Table A7: Effect of minimum grouped-training support on the PureForest comparison. At threshold

S

, evaluation includes species-fold cells with at least

S

training patches in the grouped split. Random and Grouped are pooled out-of-fold accuracies on the retained patches; Drop

=

Random

-

Grouped.

Species	$n$	Random	Grouped	Drop
A. alba	811	0.691	0.004	+0.687
Q. robur	447	0.902	0.441	+0.461
P. nigra	3198	0.873	0.694	+0.179
Q. petraea	3071	0.967	0.788	+0.179
P. menziesii	530	0.930	0.781	+0.149
P. abies	4074	0.931	0.840	+0.091

Table A9: PureForest accuracy by species under random patch-level and forest-grouped five-fold evaluation. Scores are pooled out-of-fold accuracies on identical patches, and

n

is the patch count. Drop

=

Random

-

Grouped, computed before rounding; bold denotes the patch-weighted pooled row.

Tree Species Classification

Tree species classification identifies tree taxa from images, spectral measurements, 3D structure, and satellite or aerial observations.

See all 18 papers

Multispectral airborne laser scanning dataset for tree species classification: MS-ALS-SPECIES
Apr 27, 2026Matti Hyyppä, Klaara Salolahti, Eric Hyyppä +9Point Cloud Classification Tree Species Classification
Department of Remote Sensing and Photogrammetry, Finnish Geospatial Research Institute FGI, The National Land Survey of Finland, Vuorimiehentie 5, Espoo, FI-02150, Finland
FLORA: A deep learning approach to predict forest attributes from heterogeneous LiDAR data
Jun 30, 2026Emilie Vautier, Clément Mallet, Cédric VegaPoint Cloud Learning Remote Sensing
Univ Gustave Eiffel, Géodata Paris, IGN, LASTIG, F-77454 Marne-la-Vallée, France · Univ Gustave Eiffel, Géodata Paris, IGN, LIF, F-54000, Nancy, France · Université de Lorraine, Géodata Paris, IGN, LIF, F-54000 Nancy, France
The Role of Radiometric Features in Cross-Site Leaf-Wood Segmentation of LiDAR Point Clouds
Sep 18, 2026Roman Kaharlytskyi, Derek T. Robinson, Roberto Guglielmi3D Point Cloud Segmentation Remote Sensing
Dept. of Applied Mathematics, Faculty of Mathematics University of Waterloo Waterloo, Canada · Dept. of Geography and Environmental Management, Faculty of Environment University of Waterloo Waterloo, Canada

Apr 27, 2026J/K move · Enter open · S save

Multispectral airborne laser scanning dataset for tree species classification: MS-ALS-SPECIES

Matti Hyyppä, Klaara Salolahti, Eric Hyyppä, Xiaowei Yu+8

Department of Remote Sensing and Photogrammetry, Finnish Geospatial Research Institute FGI, The National Land Survey of Finland, Vuorimiehentie 5, Espoo, FI-02150, Finland

Point Cloud Classification Tree Species Classification Remote Sensing Airborne Laser Scanning

Open paper AskPDF arXiv

The shift from stand-level to individual-tree-level forest assessments supports improved species mapping and biodiversity monitoring, particularly in boreal ecosystems where tree species like aspen (Populus tremula L.) play a keystone role. Airborne laser scanning (ALS) is the standard for such inventories, but a major limitation for developing improved species classification methods is the small number of publicly available ALS datasets containing high-quality, field-validated reference data. Recently, multispectral ALS data has shown promise for tree species classification, but the progress is hindered by the lack of open multispectral ALS datasets with high-quality field reference data. This paper presents and details an open multispectral ALS dataset for tree species classification that was used before its public release for an international benchmarking study of machine learning and deep learning classification methods in a related publication by Taher et al.,(2026). The dataset comprises 6326 segment-level point clouds of individual trees representing nine species in southern Finland. The point cloud data has been acquired using two multispectral laser scanning systems each operating at three laser wavelengths: a helicopter-borne system (HeliALS) with a point density exceeding 1000 points\m2 and an Optech Titan system with approximately 35 points\m2. Furthermore, we present a crowdsourcing application that facilitates the collection of high-quality field reference data of tree species in an efficient and scalable manner. Our article showcases the versatility of the open dataset by presenting new analyses on species classification using multispectral data building upon the initial findings of Taher et al.,(2026).

Abstract

Tree Species Classification

Explore similar work

Multispectral airborne laser scanning dataset for tree species classification: MS-ALS-SPECIES

FLORA: A deep learning approach to predict forest attributes from heterogeneous LiDAR data

The Role of Radiometric Features in Cross-Site Leaf-Wood Segmentation of LiDAR Point Clouds