deadtrees.earth-aerial: A Multi-Resolution Aerial Image Dataset for Tree Cover and Mortality Detection

This repository is the official implementation of [deadtrees.earth-aerial: A Multi-Resolution Aerial Image Dataset for Tree Cover and Mortality Detection](Arxiv link will be shared soon).

Setup

# Clone the repository
git clone github.com/GeoSense-Freiburg/DTE-aerial

# Install and activate the conda environment
conda create -n dte python=3.10
conda activate dte
pip install -r requirements.txt

---

Download Benchmark Dataset

The benchmark dataset will be made publicly available upon release. Until then, reviewers can download it using a Harvard Dataverse API token.

curl -L -OJ \
-H "X-Dataverse-key: <YOUR_DATAVERSE_API_KEY>" \
"https://dataverse.harvard.edu/api/access/dataset/:persistentId/?persistentId=doi:10.7910/DVN/IYCUML"

unzip dataverse_files.zip

tar -xvf DTE-aerial-bench-tiles.tar
tar -xvf DTE-aerial-bench-masks.tar

rm dataverse_files.zip
rm DTE-aerial-bench-tiles.tar
rm DTE-aerial-bench-masks.tar

After extraction, the dataset should have the following structure:

DTE-Aerial-Data/
├── DTE-aerial-bench-meta.csv
├── tiles/
└── masks/

---

Download Pre-trained Model

The pretrained model is available on Hugging Face: DTE-aerial-model.

You can either download it programmatically using the Hugging Face Hub:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="ayushi3536/DTE-aerial-model",
    local_dir="DTE-aerial-model",
)

or clone the repository with Git LFS:

git lfs install
git clone https://huggingface.co/ayushi3536/DTE-aerial-model

---

Evaluation

update <input_dir> in ./config/evaluation.yml

python evaluation.py --cfg ./config/evaluation.yml --checkpoint <PATH_TO_CHECKPOINT>

Repository Structure

DTE-aerial/
├── train.py                 # Training entry point
├── evaluation.py            # Evaluation entry point
├── requirements.txt
├── README.md
│
├── config/
│   ├── train.yml
│   └── evaluation.yml
│
├── scripts/
│   └── data_download.py
│
└── src/
    ├── dataset/             # Dataset loading
    ├── model/               # Network architectures
    ├── loss/                # Loss functions
    └── utils/               # Utilities

Training

To train the models described in the paper:

1. Prepare a configuration file (see config/) by specifying the path to the dataset.
2. Run the following command:

python train.py --cfg ./config/<config_file>.yaml --output <output_path>

Pre-trained Models

All pre-trained models will be available soon

Results

Tree Mortality Segmentation (F1)

Model	Temp	Trop	Boreal	Drylands	5cm	10cm	20cm
DT-V1	0.50	0.61	0.40	0.55	0.54	0.47	0.38
MiT-B3	0.56	0.64	0.58	0.56	0.59	0.55	0.45
MiT-B1	0.54	0.65	0.57	0.57	0.58	0.53	0.42
U-Net (ResNet34)	0.52	0.63	0.53	0.59	0.57	0.52	0.42
M2F (Small)	0.52	0.65	0.57	0.58	0.58	0.54	0.44
DeepLabV3+ (R50)	0.50	0.63	0.54	0.56	0.56	0.49	0.40
DINOv2 (Base)	0.40	0.61	0.42	0.54	0.48	0.46	0.38

Tree Cover Segmentation (F1)

Model	Temp	Trop	Boreal	Drylands	5cm	10cm	20cm
OAM-TCD	0.86	0.91	0.88	0.86	0.88	0.89	0.87
MiT-B3	0.85	0.91	0.93	0.93	0.89	0.89	0.89
MiT-B1	0.85	0.91	0.93	0.92	0.89	0.89	0.88
U-Net (ResNet34)	0.85	0.91	0.92	0.92	0.89	0.89	0.88
M2F (Small)	0.85	0.91	0.93	0.92	0.89	0.89	0.88
DeepLabV3+ (R50)	0.84	0.91	0.93	0.92	0.89	0.89	0.87
DINOv2 (Base)	0.84	0.91	0.87	0.89	0.87	0.87	0.84

Citation

@article{sharma2026deadtrees,
  title={deadtrees. earth-aerial: A Multi-Resolution Aerial Image Dataset for Tree Cover and Mortality Detection},
  author={Sharma, Ayushi and Mosig, Clemens and Drees, Lukas and Soltani, Salim and Vajna-Jehle, Janusch and Sheppard, Aaron and Ahmadi, Belqis and Schmid, Jonathan and Neumeier, Paul and Jacobs, Nathan and others},
  journal={arXiv preprint arXiv:2605.19605},
  year={2026}
}