Datasets:
Commit ·
d6caaf8
1
Parent(s): 62314c2
Add dataset card and metadata for Thalia (#1)
Browse files- Add dataset card and metadata for Thalia (eff4943832678374250ae3796ec33fc8f1b124e8)
Co-authored-by: Niels Rogge <[email protected]>
README.md
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
task_categories:
|
| 5 |
+
- image-classification
|
| 6 |
+
- image-segmentation
|
| 7 |
+
pretty_name: Thalia
|
| 8 |
+
tags:
|
| 9 |
+
- earth-science
|
| 10 |
+
- remote-sensing
|
| 11 |
+
- volcano-monitoring
|
| 12 |
+
- insar
|
| 13 |
+
- multi-modal
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# Thalia: A Global, Multi-Modal Dataset for Volcanic Activity Monitoring
|
| 17 |
+
|
| 18 |
+
[**Paper**](https://huggingface.co/papers/2505.17782) | [**GitHub**](https://github.com/Orion-AI-Lab/Thalia) | [**Interactive Demo (Colab)**](https://colab.research.google.com/drive/1NeVtXEqrAawe0ICw1prMJSuFdHPHOqlg)
|
| 19 |
+
|
| 20 |
+
Thalia is a global, multi-modal dataset for volcanic activity monitoring through Satellite-based Interferometric Synthetic Aperture Radar (InSAR) imagery. Building upon the Hephaestus dataset, Thalia provides higher-resolution, multi-source, and multi-temporal data in a machine-learning-ready format.
|
| 21 |
+
|
| 22 |
+
## Dataset Overview
|
| 23 |
+
|
| 24 |
+
Thalia consists of 38 spatiotemporal datacubes covering 7 years of volcanic activity. It integrates:
|
| 25 |
+
- **Georeferenced InSAR imagery**: 100m Ground Sample Distance (GSD) with physically interpretable pixel values.
|
| 26 |
+
- **Topographic Data**: Digital Elevation Model (DEM).
|
| 27 |
+
- **Atmospheric Variables**: Data used to account for signal delays that can mimic ground deformation.
|
| 28 |
+
- **Rich Annotations**: Includes expert labels for deformation type (sill, dyke, mogi, spheroid, earthquake), intensity level (low, medium, high), and volcanic activity phase (rest, unrest, rebound).
|
| 29 |
+
- **Text Descriptions**: Descriptive text for each sample, enabling language-based modeling.
|
| 30 |
+
|
| 31 |
+
## Tasks & Benchmark
|
| 32 |
+
|
| 33 |
+
Thalia supports two primary computer vision tasks for both single-image and time-series inputs:
|
| 34 |
+
|
| 35 |
+
### Image Classification
|
| 36 |
+
Evaluated using architectures such as:
|
| 37 |
+
- ResNet, MobileNet v3, EfficientNet v2, ConvNeXt, and ViT.
|
| 38 |
+
|
| 39 |
+
### Semantic Segmentation
|
| 40 |
+
Evaluated using:
|
| 41 |
+
- DeepLab v3, UNet, and SegFormer.
|
| 42 |
+
|
| 43 |
+
## Data Split
|
| 44 |
+
|
| 45 |
+
The dataset follows a temporal split to ensure robust evaluation:
|
| 46 |
+
- **Training**: 01/2014 – 05/2019
|
| 47 |
+
- **Validation**: 06/2019 – 12/2019
|
| 48 |
+
- **Testing**: 01/2020 – 12/2021
|
| 49 |
+
|
| 50 |
+
## Citation
|
| 51 |
+
|
| 52 |
+
If you use this dataset in your research, please cite the following paper:
|
| 53 |
+
|
| 54 |
+
```bibtex
|
| 55 |
+
@article{papadopoulos2025thalia,
|
| 56 |
+
title={Thalia: A Global, Multi-Modal Dataset for Volcanic Activity Monitoring},
|
| 57 |
+
author={Papadopoulos, Nikolas and Bountos, Nikolaos Ioannis and Sdraka, Maria and Karavias, Andreas and Papoutsis, Ioannis},
|
| 58 |
+
journal={arXiv preprint arXiv:2505.17782},
|
| 59 |
+
year={2025}
|
| 60 |
+
}
|
| 61 |
+
```
|