jannikend/birch-trees
收藏Hugging Face2026-03-27 更新2026-03-29 收录
下载链接：
https://hf-mirror.com/datasets/jannikend/birch-trees
下载链接
链接失效反馈官方服务：
资源简介：
---
license: cc-by-4.0
tags:
- tree-height-estimation
- species-identification
- remote-sensing
- forest-monitoring
- drone-imagery
- ecology
- uav
- vision
- fine-grained-classification
- biodiversity
pretty_name: BIRCH-Trees
size_categories:
- 10K<n<100K
source_datasets:
- Quebec Trees Dataset
- Barro Colorado Island Dataset
- Quebec Plantations Dataset
task_categories:
- image-classification
task_ids:
- multi-class-image-classification
---
# BIRCH-Trees 🌳🌲🌴
[](https://RolnickLab.github.io/DINOvTree)
[](https://arxiv.org/pdf/2603.23669)
[](https://arxiv.org/abs/2603.23669)
[](https://github.com/RolnickLab/DINOvTree)
[](https://huggingface.co/jannikend/dinovtree)
This is the official Hugging Face repository for the benchmark introduced in our paper:
**Estimating Individual Tree Height and Species from UAV Imagery**
Authors: [Jannik Endres](https://scholar.google.com/citations?user=imQyM0QAAAAJ&hl=en), [Etienne Laliberté](https://lefo.ca/), [David Rolnick](http://www.davidrolnick.com/), [Arthur Ouaknine](https://arthurouaknine.github.io/)
## 🧭 Table of Contents
1. [Benchmark Overview](#benchmark-overview)
2. [News](#news)
3. [Datasets](#datasets)
- [Quebec Trees](#datasets-qt)
- [BCI](#datasets-bci)
- [Quebec Plantations](#datasets-qp)
4. [Benchmark Structure](#benchmark-structure)
- [Quebec Trees](#structure-qt)
- [BCI](#structure-bci)
- [Quebec Plantations](#structure-qp)
5. [License](#license)
6. [Citation](#citation)
<h2 id="benchmark-overview">📝 Benchmark Overview</h2>
BIRCH-Trees (*Benchmark for Individual Recognition of Class and Height of Trees*) is the first benchmark for individual tree height and species estimation from tree-centered UAV images.
We formulate individual tree height estimation and species identification as regression and classification tasks, respectively. We provide high-resolution RGB drone images, alongside corresponding Digital Surface Models (DSMs), of tree canopies from three distinct environments:
* **Temperate forest:** Quebec Trees
* **Tropical forest:** Barro Colorado Island (BCI)
* **Boreal plantation:** Quebec Plantations
<h2 id="news">📰 News</h2>
- **26/03/2026:** Our paper is available on [arXiv](https://arxiv.org/abs/2603.23669).
<h2 id="datasets">📦 Datasets</h2>
Below is a summary of the three datasets that make up the BIRCH-Trees benchmark:
| Dataset | Environment | Resolution | Total Images | Splits (Train / Val / Test) | Classes | Mean ± Std of Height |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| **Quebec Trees** | Temperate Forest | 1.9 cm/px | 22.3K | 13.3K / 3.6K / 5.4K | 14 | 14.22 m ± 4.91 m |
| **BCI** | Tropical Forest | 4.0 cm/px | 2.0K | 1.4K / 0.3K / 0.3K | 21 | 29.05 m ± 6.59 m |
| **Quebec Plantations** | Boreal Plantation | 0.5 cm/px | 17.7K | 11.1K / 4.0K / 2.6K | 8 | 3.48 m ± 1.47 m |
<h3 id="dataset-qt">Quebec Trees</h3>
The Quebec Trees dataset ([Source](https://data.niaid.nih.gov/resources?id=zenodo_8148478)) contains images of temperate forest trees in Quebec, Canada. We adopt the class definitions from [Teng et al, 2025](https://arxiv.org/abs/2506.04970), excluding supercategories ("Acer", "Betula", "Magnoliopsida", "Pinopsida"), which reflect annotator uncertainty. After filtering, the dataset comprises 14 classes with a long-tailed distribution. Note that the classes "Populus" and "Picea" are genus-level and not species-level due to annotator uncertainty.
<h3 id="dataset-bci">BCI</h3>
The BCI dataset ([Source 1](https://smithsonian.figshare.com/articles/dataset/Barro_Colorado_Island_50-ha_plot_crown_maps_manually_segmented_and_instance_segmented_/24784053), [Source 2](https://smithsonian.dataone.org/view/doi%3A10.60635%2FC3F593)), captured in Panama, represents a complex tropical forest environment. Unlike [Teng et al, 2025](https://arxiv.org/abs/2506.04970), we exclude the "Other" class and any class with fewer than two samples in the validation or test sets to ensure reliable evaluation. The resulting 21 classes exhibit a long-tailed distribution. Classes correspond to families and not species on the BCI dataset.
<h3 id="dataset-qp">Quebec Plantations</h3>
The Quebec Plantations dataset ([Source](https://www.frdr-dfdr.ca/repo/dataset/9f10a155-c89f-43ee-9864-da28ca436af6)) comprises images from boreal plantations in Quebec, Canada. Plantations contain young trees in regular grids, so most images exclude neighboring trees. We exclude the "Other" class aggregating diverse species without consistent visual characteristics. The 8 remaining classes exhibit an imbalanced distribution, dominated by *Picea glauca* and *Pinus banksiana*.
<h2 id="benchmark-structure">📁 Benchmark Structure</h2>
Below is an overview of the benchmark structure, with descriptions for key files and directories:
```
birch-trees/
├── datasets/ # Datasets of the benchmark.
│ ├── bci/ # BCI dataset.
│ ├── quebec_plantations/ # Quebec Plantations dataset.
│ └── quebec_trees/ # Quebec Trees dataset.
└── README.md # Benchmark README file (this file)
```
<h3 id="structure-qt">Quebec Trees</h3>
<details>
<summary><b>Click to expand full Dataset Structure</b></summary>
```
quebec_trees/
├── 20210902_sblz1/ # --- ZONE 1 (Train, Valid, Test) ---
│ ├── 2021_09_02_sbl_z1_rgb/
│ │ ├── tiles/
│ │ │ ├── test/*.tif # Test RGB tiles
│ │ │ ├── train/*.tif # Train RGB tiles
│ │ │ └── valid/*.tif # Valid RGB tiles
│ │ └── *.json # COCO labels: test, train, valid
│ └── 20210902_sblz1_p4rtk_dsm/
│ ├── tiles/
│ │ ├── test/*.tif # Test DSM tiles
│ │ ├── train/*.tif # Train DSM tiles
│ │ └── valid/*.tif # Valid DSM tiles
│ └── *.json # COCO labels: test, train, valid
│
├── 20210902_sblz2/ # --- ZONE 2 (Train) ---
│ ├── 2021_09_02_sbl_z2_rgb/
│ │ ├── tiles/train/*.tif # Training RGB tiles
│ │ └── *train.json # COCO labels: train
│ └── 20210902_sblz2_p4rtk_dsm/
│ ├── tiles/train/*.tif # Training DSM tiles
│ └── *train.json # COCO labels: train
│
├── 20210902_sblz3/ # --- ZONE 3 (Test) ---
│ ├── 2021_09_02_sbl_z3_rgb/
│ │ ├── tiles/test/*.tif # Test RGB tiles
│ │ └── *test.json # COCO labels: test
│ └── 20210902_sblz3_p4rtk_dsm/
│ ├── tiles/test/*.tif # Test DSM tiles
│ └── *test.json # COCO labels: test
│
├── quebec_trees_categories.json # Global category definitions
├── stats_quebec_trees_by_id.json # Number of instances per id after exclusion
└── tree_allometry_regression_results_log_quebec_trees.csv # Parameters for allometric equations
```
</details>
<h3 id="structure-bci">BCI</h3>
<details>
<summary><b>Click to expand full Dataset Structure</b></summary>
```
bci/
├── bci_50ha_2022_09_29/ # --- BCI (Train, Valid, Test) ---
│ ├── bci_50ha_2022_09_29_dsm/
│ │ ├── tiles/
│ │ │ ├── test/*.tif # Test DSM tiles
│ │ │ ├── train/*.tif # Train DSM tiles
│ │ │ └── valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: test, train, valid
│ └── bci_50ha_2022_09_29_orthomosaic/
│ ├── tiles/
│ │ ├── test/*.tif # Test RGB tiles
│ │ ├── train/*.tif # Train RGB tiles
│ │ └── valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: test, train, valid
│
├── bci_categories.json # Global category definitions
├── stats_bci_by_id.json # Number of instances per id after exclusion
└── tree_allometry_regression_results_log_bci.csv # Parameters for allometric equations
```
</details>
<h3 id="structure-qp">Quebec Plantations</h3>
<details>
<summary><b>Click to expand full Dataset Structure</b></summary>
```
quebec_plantations/
├── 20230605_cbblackburn1_p1/ # --- SITE: Cbblackburn1 (Valid) ---
│ ├── 20230605_cbblackburn1_p1_dsm/
│ │ ├── tiles/valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: valid
│ └── 20230605_cbblackburn1_p1_rgb/
│ ├── tiles/valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: valid
│
├── 20230606_cbblackburn3_p1/ # --- SITE: Cbblackburn3 (Test) ---
│ ├── ..._dsm/
│ │ ├── tiles/test/*.tif # Test DSM tiles
│ │ └── *.json # COCO labels: test
│ └── ..._rgb/
│ ├── tiles/test/*.tif # Test RGB tiles
│ └── *.json # COCO labels: test
│
├── 20230606_cbblackburn4_p1/ # --- SITE: Cbblackburn4 (Test) ---
│ ├── ..._dsm/
│ │ ├── tiles/test/*.tif # Test DSM tiles
│ │ └── *.json # COCO labels: test
│ └── ..._rgb/
│ ├── tiles/test/*.tif # Test RGB tiles
│ └── *.json # COCO labels: test
│
├── 20230606_cbblackburn5_p1/ # --- SITE: Cbblackburn5 (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── 20230606_cbblackburn6_p1/ # --- SITE: Cbblackburn6 (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── 20230607_cbblackburn2_p1/ # --- SITE: Cbblackburn2 (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── 20230608_cbbernard1_p1/ # --- SITE: Cbbernard1 (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── 20230608_cbbernard2_p1/ # --- SITE: Cbbernard2 (Valid) ---
│ ├── ..._dsm/
│ │ ├── tiles/valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: valid
│ └── ..._rgb/
│ ├── tiles/valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: valid
│
├── 20230608_cbbernard3_p1/ # --- SITE: Cbbernard3 (Train, Valid, Test) ---
│ ├── 20230608_cbbernard3_p1_dsm/
│ │ ├── tiles/
│ │ │ ├── test/*.tif # Test DSM tiles
│ │ │ ├── train/*.tif # Train DSM tiles
│ │ │ └── valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: test, train, valid
│ └── 20230608_cbbernard3_p1_rgb/
│ ├── tiles/
│ │ ├── test/*.tif # Test RGB tiles
│ │ ├── train/*.tif # Train RGB tiles
│ │ └── valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: test, train, valid
│
├── 20230608_cbbernard4_p1/ # --- SITE: Cbbernard4 (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── 20230608_cbpapinas_p1/ # --- SITE: Cbpapinas (Valid, Test) ---
│ ├── ..._dsm/
│ │ ├── tiles/
│ │ │ ├── test/*.tif # Test DSM tiles
│ │ │ └── valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: test, valid
│ └── ..._rgb/
│ ├── tiles/
│ │ ├── test/*.tif # Test RGB tiles
│ │ └── valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: test, valid
│
├── 20230609_cbserpentin1_p1/ # --- SITE: Cbserpentin1 (Test) ---
│ ├── ..._dsm/
│ │ ├── tiles/test/*.tif # Test DSM tiles
│ │ └── *.json # COCO labels: test
│ └── ..._rgb/
│ ├── tiles/test/*.tif # Test RGB tiles
│ └── *.json # COCO labels: test
│
├── 20230609_cbserpentin2_p1/ # --- SITE: Cbserpentin2 (Valid) ---
│ ├── ..._dsm/
│ │ ├── tiles/valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: valid
│ └── ..._rgb/
│ ├── tiles/valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: valid
│
├── 20230712_afcagauthier_itrf20_p1/ # --- SITE: Afcagauthier (Valid, Test) ---
│ ├── ..._dsm/
│ │ ├── tiles/
│ │ │ ├── test/*.tif # Test DSM tiles
│ │ │ └── valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: test, valid
│ └── ..._rgb/
│ ├── tiles/
│ │ ├── test/*.tif # Test RGB tiles
│ │ └── valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: test, valid
│
├── 20230712_afcagauthmelpin_itrf20_p1/ # --- SITE: Afcagauthmelpin (Train, Valid, Test) ---
│ ├── ..._dsm/
│ │ ├── tiles/
│ │ │ ├── test/*.tif # Test DSM tiles
│ │ │ ├── train/*.tif # Train DSM tiles
│ │ │ └── valid/*.tif # Valid DSM tiles
│ │ └── *.json # COCO labels: test, train, valid
│ └── ..._rgb/
│ ├── tiles/
│ │ ├── test/*.tif # Test RGB tiles
│ │ ├── train/*.tif # Train RGB tiles
│ │ └── valid/*.tif # Valid RGB tiles
│ └── *.json # COCO labels: test, train, valid
│
├── 20230712_afcahoule_itrf20_p1/ # --- SITE: Afcahoule (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── 20230712_afcamoisan_itrf20_p1/ # --- SITE: Afcamoisan (Train) ---
│ ├── ..._dsm/
│ │ ├── tiles/train/*.tif # Train DSM tiles
│ │ └── *.json # COCO labels: train
│ └── ..._rgb/
│ ├── tiles/train/*.tif # Train RGB tiles
│ └── *.json # COCO labels: train
│
├── quebec_plantations_categories.json # Global category definitions
├── stats_quebec_plantations_by_id.json # Number of instances per id after exclusion
└── tree_allometry_regression_results_log_quebec_plantations.csv # Parameters for allometric equations
```
</details>
<h2 id="license">⚖️ License</h2>
The **BIRCH-Trees** benchmark is released under the [CC BY 4.0 License](https://creativecommons.org/licenses/by/4.0/).
The underlying source datasets maintain their original licenses:
* **Quebec Trees:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
* **BCI:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
* **Quebec Plantations:** [CC0 1.0 Universal](https://creativecommons.org/publicdomain/zero/1.0/)
<h2 id="citation">📚 Citation</h2>
If you find the BIRCH-Trees benchmark useful for your research, please consider citing our paper:
```bibtex
@article{endres2026treeheightspecies,
title = {Estimating Individual Tree Height and Species from UAV Imagery},
author = {Endres, Jannik and Lalibert{\'e}, Etienne and Rolnick, David and Ouaknine, Arthur},
journal = {arxiv:2603.23669 [cs.CV]},
year = {2026}
}
```
### Source Datasets
Depending on which subsets of the BIRCH-Trees benchmark you utilize, please ensure you also credit the original dataset creators by citing their respective works:
**Quebec Trees:**
```bibtex
@article{cloutier2024influence,
title = {Influence of temperate forest autumn leaf phenology on segmentation of tree species from UAV imagery using deep learning},
author = {Cloutier, Myriam and Germain, Micka{\"e}l and Lalibert{\'e}, Etienne},
journal = {Remote Sensing of Environment},
volume = {311},
pages = {114283},
year = {2024},
publisher = {Elsevier}
}
```
**BCI:**
```bibtex
@misc{Vasquez2023BCI,
author = {Vicente Vasquez and Katherine Cushman and Pablo Ramos and Cecilia Williamson and Paulino Villareal and Luisa Fernanda Gomez Correa and Helene C. Muller-Landau},
title = {Barro Colorado Island 50-ha Plot Crown Maps: Manually Segmented and Instance Segmented},
year = {2023},
doi = {10.25573/data.24784053.v2},
publisher = {Smithsonian Institution Figshare},
}
```
and
```bibtex
@misc{forestgeo_smithsonian_2024,
author = {{ForestGEO Smithsonian}},
title = {2023 high-resolution airborne {LiDAR} data for {Barro Colorado Island} and other {Smithsonian ForestGEO Sites} in {Central Panama}},
year = {2024},
doi = {10.60635/C3F593},
publisher = {Smithsonian Research Data Repository}
}
```
**Quebec Plantations:**
```bibtex
@misc{Lefebvre:2024,
author = {Lefebvre, Isabelle and Lalibert{\'e}, Etienne},
title = {UAV LiDAR, UAV Imagery, Tree Segmentations and Ground Measurements for Estimating Tree Biomass in Canadian (Quebec) Plantations},
year = {2024},
doi = {10.20383/103.0979},
publisher = {Federated Research Data Repository}
}
```
提供机构：
jannikend
