meteahishali/ADA-Net-dataset

Attention-Guided Domain Adaptation Network (ADA-Net) =============================

This repository shares the data for ADA-Net: Attention-Guided Domain Adaptation Network with Contrastive Learning for Standing Dead Tree Segmentation Using Aerial Imagery and includes the annotated dataset for mapping standing dead trees. ADA-Nets are generic networks and they can be used in different domation adaptation and Image-to-Image translation problems. In this repository, we specifically focus on transforming multispectral remote sensing aerial images from USA sites into images resembling those from Finland. The tree annotations are provided at the individual tree level.

Usage

Please refer to: https://github.com/meteahishali/ADA-Net and https://huggingface.co/docs/datasets/loading#hdf5-files

<p align="center"> <img src="usa2finland.png" width="1000"/> </p> <p align="center"> <em>Dead tree segmentation results are given for both the original images and the generated ones obtained through different domain transformation approaches. The pretrained segmentation network is trained using images from Finland sites.</em> </p>

Citation

If you use method(s) and the dataset(s) provided in this repository, please cite the following paper:

M. Ahishali, A. U. Rahman, E. Heinaro, and S. Junttila, "ADA-Net: Attention-Guided Domain Adaptation Network with Contrastive Learning for Standing Dead Tree Segmentation Using Aerial Imagery," arXiv preprint arXiv:2504.04271, 2025.

@misc{ahishali2025adanet,
      title={ADA-Net: Attention-Guided Domain Adaptation Network with Contrastive Learning for Standing Dead Tree Segmentation Using Aerial Imagery}, 
      author={Mete Ahishali and Anis Ur Rahman and Einari Heinaro and Samuli Junttila},
      year={2025},
      eprint={2504.04271},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2504.04271}, 
}

Downloading the Dataset

We collect the dataset consisting of unpaired aerial multispectral image samples from the US [1] and Finland [2].

These datasets also consist of polygon annotations for standing dead trees annotated by our collaborator group of forest health experts. Note that we share only a small sub-set of the Finland data due to the extensive size of the whole annotated regions and the aerial imagery data.

Kaggle Dataset

Although we already provide direct ``.h5` files for the pre-processed data above, the full dataset with untiled image frames are available in the following Kaggle repository: https://www.kaggle.com/datasets/meteahishali/aerial-imagery-for-standing-dead-tree-segmentation. We share the RGB and NRG images in `.png`` format together with the corresponding ground-truth mask images for the USA data.

References

[1] "National Agriculture Imagery Program," https://naip-usdaonline.hub.arcgis.com/. \ [2] "National Land Survey of Finland," https://asiointi.maanmittauslaitos.fi/karttapaikka/tiedostopalvelu.