First off, thank you for considering contributing to this project! This repository represents a starting point for cross-sensor transfer learning in cloud removal, and community contributions are highly encouraged.
Because this project establishes a baseline for LISS-IV using Sentinel-2 proxy training, there is plenty of room for further development:
- Fine-Tuning on Real Data: If you have access to true paired (clear/cloudy) LISS-IV imagery, you can fork this repository, load our pretrained weights, and run a few-shot fine-tuning pass to perfectly bridge the 10m vs 5.8m spatial resolution gap.
- Architecture Tweaks: Feel free to experiment with different layer grafting strategies or modifications to the UNet backbone.
- Evaluation Pipelines: Build more robust evaluation scripts or integrate different radiometric calibration equations.
- Fork the Repository: Create your own copy of the project.
- Clone Locally:
git clone https://github.com/YOUR_USERNAME/liss-cloud-removal.git - Create a Branch:
git checkout -b feature/amazing-new-idea - Experiment & Build: Use your own compute to push the limits of what this model can do!
- Open a Pull Request: If you build something great that improves the core pipeline, open a PR. Please ensure your code is clean and includes comments explaining any new logic.
By contributing, you agree that your contributions will be licensed under its MIT License.