ForestSemantic-MS Dataset: multispectral LiDAR forest point clouds for fine-grained forest semantic segmentation
Authors/Creators
- 1. Fondazione Bruno Kessler
- 2. TU Wien
-
3.
Finnish Geospatial Research Institute
Description
Description
ForestSemantic-MS dataset contains multispectral (MS) LiDAR forest point clouds used in our paper titled 3D Forest Semantic Segmentation Using Multispectral LiDAR and 3D Deep Learning (DOI: 10.1007/s41064-025-00369-4). This dataset is collected by the helicopter-mounted HeliALS multispectral LiDAR system developed by the Finnish Geospatial Research Institute (FGI). The MS point clouds are manually annotated in six forest components: ground, low vegetation, trunk, branches, foliage, and woody debris. ForestSemantic-MS consists of six forest plots used to train (four plots) and evaluate (two plots) deep learning models we benchmarked in our paper.
Point cloud attributes
| Attribute | Description |
| SWIR | Normalized ([0-1]) reflectance at 1550 nm |
| NIR | Normalized ([0-1]) reflectance at 905 nm |
| Green | Normalized ([0-1]) reflectance at 532 nm |
| VI | Normalized ([0-1]) vegetation index (NDVI NIR-SWIR ) |
| semantic_GT | Semantic segmentation labels |
Labels
| Class | Name |
| 0 |
Ground |
| 1 |
Low vegetation |
| 2 |
Trunks |
| 3 |
Branches |
| 4 |
Foliage |
| 5 |
Woody debris |
ForestSemantic-MS vs. Other Public Datasets
|
FOR-Instance (Xiang et al., 2024) |
ForestSemantic (Liang et al., 2024) |
EvoMS (Ruoppa et al., 2025) |
ForestSemantic-MS (ours) |
|
|
Forest component detail level |
5 (Ground, low vegetation, stem, live branches, and dead branches) |
3 (Trunk, branches, and foliage) |
2 (Foliage and wood) |
6 (Ground, low vegetation, trunk, branches, foliage, and woody debris) |
|
Multispectral LiDAR |
✕ |
✕ |
✓ (1550 nm, 905 nm, and 532 nm) |
✓ (1550 nm, 905 nm, and 532 nm) |
|
Vegetation index |
✕ |
✕ |
✕ |
NDVI NIR-SWIR |
|
Platform |
ULS |
TLS |
ULS |
ULS |
|
Forest type(s) |
Boreal, temperate, alluvial, and eucalypt forests |
Boreal forests |
Boreal forests |
Boreal forests |
|
Geographical data coverage |
Norway, Austria, the Czech Republic, Australia, and New Zealand |
Finland |
Finland |
Finland |
|
Number of points
|
116,099,253 |
355,511,770 |
8,265,448 |
9,600,927 |
Citation
Any scientific publication using this dataset should cite the following paper and the dataset:
Takhtkeshha, N., Bocaux, L., Ruoppa, L., Remondino, F., Mandlburger, G., Kukko, A., & Hyyppä, J. (2025). 3D forest semantic segmentation using multispectral LiDAR and 3D deep learning. PFG – Journal of Photogrammetry Remote Sensing and Geoinformation Science. https://doi.org/10.1007/s41064-025-00369-4
735 multispectral LiDAR forest point clouds for fine-grained forest semantic segmentation [dataset], 2025a. doi:
736 10.5281/zenodo.17172162
Files
Files
(131.1 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:2c36cc760602d75fee4bd466f7da9a8f
|
7.8 MB | Download |
|
md5:1476d103d889dfdde238c76a9253e8ff
|
18.3 MB | Download |
|
md5:8d24c31df17ef58e58fd25f8bad23238
|
21.0 MB | Download |
|
md5:b24cf74f9feba6dce5b4014c95a6312c
|
42.6 MB | Download |
|
md5:5f360b14e2720482fe9916ca702973d0
|
16.1 MB | Download |
|
md5:b29fd542abbe35dd606793531d43ce12
|
25.3 MB | Download |
Additional details
Software
- Repository URL
- https://github.com/3DOM-FBK/3D-forest-semantic-segmentation
- Programming language
- Python
- Development Status
- Active