Improve model card: add metadata, paper link, and repository link
#1
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
pipeline_tag: image-feature-extraction
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# EarthBridge: MaRS-Base Encoder
|
| 7 |
+
|
| 8 |
+
This repository contains a pre-trained MaRS encoder (Swin Transformer v2), a component of the **EarthBridge** translation framework. EarthBridge was developed for the 4th Multi-modal Aerial View Image Challenge — Translation (MAVIC-T) to perform high-fidelity image-to-image translation among Electro-Optical (EO), Infrared (IR), and Synthetic Aperture Radar (SAR) sensors.
|
| 9 |
+
|
| 10 |
+
- **Paper:** [EarthBridge: A Solution for 4th Multi-modal Aerial View Image Challenge Translation Track](https://huggingface.co/papers/2603.06753)
|
| 11 |
+
- **Repository:** [Bili-Sakura/EarthBridge-Preview](https://github.com/Bili-Sakura/EarthBridge-Preview)
|
| 12 |
+
|
| 13 |
+
## Model Description
|
| 14 |
+
|
| 15 |
+
The model utilizes the `Swinv2Model` architecture. Within the EarthBridge framework, these encoders are used for representation alignment and feature extraction to handle the distinct geometric and electromagnetic characteristics of different aerial imaging modalities. The EarthBridge framework achieved second place on the MAVIC-T leaderboard with a composite score of 0.38.
|
| 16 |
+
|
| 17 |
+
## Citation
|
| 18 |
+
|
| 19 |
+
```bibtex
|
| 20 |
+
@article{earthbridge2025,
|
| 21 |
+
title={EarthBridge: A Solution for 4th Multi-modal Aerial View Image Challenge Translation Track},
|
| 22 |
+
author={Li, Bo and others},
|
| 23 |
+
journal={arXiv preprint arXiv:2603.06753},
|
| 24 |
+
year={2025}
|
| 25 |
+
}
|
| 26 |
+
```
|