WV-Net: A SAR Wave-mode Foundation Model

This repository contains the code and model weights accompanying AIES manuscript AIES-D-25-0003 and AGU presentation IN54B-03: "WV-Net: A SAR Wave-mode Foundation Model".

Features:

• Pre-trained Model Weights: Easily download and use our trained WV-Net model.
• Inference Demo: A simple Jupyter notebook demonstrating how to load the model and perform inference on sample data.

Roadmap:

• Reproducible codebase: Essential code for self-supervised training and evaluation.
• Additional pretrained models: More popular vision backbones pretrained used WV-Net recipe.

Getting started:

1. Clone repo:

   git clone https://github.com/hawaii-ai/WV-Net.git
   cd WV-Net

2. Set up your python environment. Setting up a virtual environment using venv or conda is recommended:

   python -m venv venv
   source venv/bin/activate # On Windows: `venv\Scripts\activate`

3. Install dependencies:

   pip install -r requirements.txt

4. Install the wvnet package containing reusable code:

   pip install -e .

5. Navigate to the notebooks directory and open a jupyter server:

   cd notebooks
   jupyter notebook

Model weights:

Pre-trained model weights are located in the model_weights/ directory. More will be added.

• wvnet_resnet50_weights.pt: standard, torchvision compatible, ResNet50 weights.