TOFMapper

Trained checkpoints for mapping and classifying Trees outside Forest (TOF) in high-resolution aerial RGB orthophotos, from the paper Mapping and classification of trees outside forests using deep learning (Science of Remote Sensing, 2026; preprint: arXiv:2510.25239).

• Code: https://github.com/Moerizzy/TOFMapper
• Training data: https://zenodo.org/records/17468859

Models

Model	File	Config	Size
FT-UNetFormer	ftunetformer.ckpt	config/tof/ftunetformer.py	1.5 GB
DC-Swin	dcswin.ckpt	config/tof/dcswin_small.py	1.1 GB
ABCNet	abcnet.ckpt	config/tof/abcnet.py	216 MB
LSKNet	unetformer_lsk_s.ckpt	config/tof/unetformer_lsk_s.py	231 MB
BANet	banet.ckpt	config/tof/banet.py	204 MB
U-Net	unet.ckpt	config/tof/unet.py	496 MB

Classes

Index	Class
0	Background
1	Forest
2	Patch
3	Linear
4	Tree

Results

Test-set results as reported in the TOFMapper repository. Best value per column in bold.

Model	mIoU	mF1	Forest IoU	Forest F1	Patch IoU	Patch F1	Linear IoU	Linear F1	Tree IoU	Tree F1
ABCNet	0.709	0.821	0.944	0.971	0.532	0.695	0.739	0.850	0.622	0.767
BANet	0.656	0.777	0.937	0.967	0.422	0.593	0.689	0.816	0.575	0.730
DC-Swin	0.714	0.824	0.949	0.974	0.570	0.726	0.751	0.858	0.586	0.727
FT-UNetFormer	0.739	0.843	0.952	0.975	0.606	0.754	0.774	0.872	0.626	0.770
LSKNet	0.697	0.812	0.943	0.971	0.545	0.705	0.718	0.836	0.582	0.736
U-Net	0.484	0.618	0.804	0.891	0.151	0.263	0.527	0.690	0.456	0.626

Usage

The files are PyTorch Lightning checkpoints and are loaded through the TOFMapper code, not through transformers. Set up the repository as described in its README, then download a checkpoint to the location its config expects (run from the Project folder that contains TOFMapper/). Example for FT-UNetFormer:

from huggingface_hub import hf_hub_download

name = "ftunetformer"  # or: dcswin, abcnet, unetformer_lsk_s, banet, unet
hf_hub_download(
    repo_id="Moerizzi/TOFMapper",
    filename=f"{name}.ckpt",
    local_dir=f"model_weights/tof/{name}",
)

Run inference on a folder of georeferenced RGB orthophotos with the matching config:

python TOFMapper/inference_with_overlap.py \
  -i path/to/images \
  -c TOFMapper/config/tof/ftunetformer.py \
  -o path/to/output \
  -st 256 -ps 1024 -b 2 -m 500 -kr 0.7

Predictions are written as GeoTIFF and converted to a shapefile.

Use an up-to-date checkout of the code. Versions of inference_with_overlap.py before 9 October 2026 swapped the red and blue channels at inference, which does not match how the models were trained.

Input

3-band RGB orthophotos (8 bit), normalised with ImageNet mean and standard deviation. The models were trained on 1024 x 1024 px patches.

License

GPL-3.0, the same as the TOFMapper code.

Citation

@article{LUCAS2026100510,
title = {Mapping and classification of trees outside forests using deep learning},
journal = {Science of Remote Sensing},
volume = {14},
pages = {100510},
year = {2026},
issn = {2666-0172},
doi = {https://doi.org/10.1016/j.srs.2026.100510},
url = {https://www.sciencedirect.com/science/article/pii/S2666017226001483},
author = {Moritz Lucas and Hamid Ebrahimy and Viacheslav Barkov and Ralf Pecenka and Kai-Uwe Kühnberger and Björn Waske}
}

Acknowledgement

TOFMapper builds on GeoSeg by Libo Wang.