---
license: odbl
size_categories:
- 10M<n<100M
pretty_name: TerraMesh-Masks
tags:
- Earth observation
- Zero-Shot Segmentation
task_categories:
- image-feature-extraction
library_name: webdataset
# viewer: false
configs:
- config_name: default
  data_files:
  - split: train
    path: "train_metadata.parquet"
  - split: validation
    path: "val_metadata.parquet"
---

[![arXiv](https://img.shields.io/badge/arXiv-coming_soon-b31b1b?logo=arxiv)](https://arxiv.org/)
[![Code](https://img.shields.io/badge/GitHub-TerraMesh_Masks-0F62FE?logo=github)](https://github.com/IBM/TerraMesh-Masks)
[![HuggingFace](https://img.shields.io/badge/Hugging_Face-TerraMesh_Masks_Eval-FFD21E?logo=huggingface)](https://huggingface.co/ibm-esa-geospatial/TerraMesh-Masks-Eval)

# TerraMesh-Masks
 
TerraMesh-Masks is a dataset for open-vocabulary segmentation of satellite imagery. This dataset provides binary segmentation masks with captions that extend the samples from [TerraMesh](https://huggingface.co/ibm-esa-geospatial/TerraMesh).
We also provide an human-verfied evaluation benchmark, called [TerraMesh-Masks-Eval](https://huggingface.co/ibm-esa-geospatial/TerraMesh-Masks-Eval).

Examples from the training subset: 

![train_examples.png](https://github.com/IBM/TerraMesh-Masks/raw/main/assets/train_examples.png)

## Usage

Download the data loading code from [GitHub](https://github.com/IBM/TerraMesh-Masks) and install requirements with `pip install -r requirements.txt`. For development, you can stream data directly from Hugging Face.

Load training samples with:
```python
from torch.utils.data import DataLoader
from terramesh.masks_dataset import build_terramesh_masks_dataset

dataset = build_terramesh_masks_dataset(
    urls="https://huggingface.co/datasets/ibm-esa-geospatial/TerraMesh-Masks/resolve/main/val/[BINARY_MASK,S2RGB]/majortom_shard_{000001...000793}.tar",
    batch_size=4,
    shardshuffle=100,
)

dataloader  = DataLoader(dataset, batch_size=None)
batch = next(iter(dataloader))
```

## Download

Please note that the full dataset requires 18TB or storage and input images are provided by https://huggingface.co/ibm-esa-geospatial/TerraMesh.

```shell
# Download image data
hf download ibm-esa-geospatial/TerraMesh --repo-type dataset --local-dir data/TerraMesh

# Download binary mask data
hf download ibm-esa-geospatial/TerraMesh-Masks --repo-type dataset --local-dir data/TerraMesh
```

If you like to download only a subset of the data, you can specify it with `--include`.
```shell
# Only download a single modality (e.g., S2L2A)
hf download ibm-esa-geospatial/TerraMesh --repo-type dataset --include "*/S2L2A/*" --local-dir data/TerraMesh

# Only download val data from TerraMesh-Masks
hf download ibm-esa-geospatial/TerraMesh-Masks --repo-type dataset --include "val/*" --local-dir data/TerraMesh
```

## Citation

Preprint for TerraMesh-Masks is coming soon!

TerraMesh:

```bibtex
@article{blumenstiel2025terramesh,
  title={Terramesh: A planetary mosaic of multimodal earth observation data},
  author={Blumenstiel, Benedikt and Fraccaro, Paolo and Marsocci, Valerio and Jakubik, Johannes and Maurogiovanni, Stefano and Czerkawski, Mikolaj and Sedona, Rocco and Cavallaro, Gabriele and Brunschwiler, Thomas and Bernabe-Moreno, Juan and others},
  journal={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  year={2025},
}
```

## Licenses

The mask annotations are released under ODbL, following the OpenStreetMap license.

The image data in TerraMesh is released under CC-BY-SA 4.0. 

## Acknowledgements

TerraMesh is part of the **FAST‑EO** project funded by the European Space Agency Φ‑Lab.

The binary annotations are based on the base layer provided by [Overture Maps](https://overturemaps.org/) and from [OpenStreetMap](https://www.openstreetmap.org/copyright) tags.