Welcome to the official repository for SEA-ViT — a cutting-edge deep learning model designed for predicting sea surface currents using Vision Transformers and bidirectional Gated Recurrent Units. This repository contains code, pretrained models, and documentation for exploring our innovative approach to ocean forecasting.
SEA-ViT integrates the Vision Transformer (ViT) architecture with bidirectional Gated Recurrent Units (GRUs) to forecast sea surface currents (U, V). Developed by Teerapong Panboonyuen, this model leverages high-frequency radar (HF) data to capture spatio-temporal dependencies and provide accurate predictions for maritime navigation, environmental monitoring, and climate analysis.
Key Contributions:
For a detailed discussion on the model and methodology, visit SEA-ViT Overview.
Our experiments demonstrate SEA-ViT’s efficacy in forecasting sea surface currents with high precision. The model is well-suited for applications requiring detailed and accurate oceanographic predictions.
Results Summary:
To get started, clone the repository and install the required dependencies:
git clone https://github.com/kaopanboonyuen/gistda-ai-sea-surface-currents.git
cd gistda-ai-sea-surface-currents
pip install -r requirements.txt
Detailed usage instructions and example code are provided in the docs directory. For questions or contributions, please refer to the contributing guidelines.
If you use SEA-ViT in your research, please cite our work:
@article{panboonyuen2024SEA-ViT,
title={SEA-ViT: Sea Surface Currents Forecasting using Vision Transformers and Bidirectional GRUs},
author={Teerapong Panboonyuen},
year={2024},
url={https://github.com/kaopanboonyuen/gistda-ai-sea-surface-currents}
}
For further inquiries, reach out to:
Thank you for visiting our repository! We hope SEA-ViT proves valuable in advancing sea surface current forecasting and oceanographic research.