Data and code for: "From underwater to drone: a novel multi-scale knowledge distillation approach for coral reef monitoring"

# Deposit Structure

## **DATA**

### **Input**  
This folder contains all the input files used in the study:  
- **Aerial orthophotos**  
- **Aerial annotations on the test set**  
- **Aerial annotated tiles**  
- **Underwater fine-scale predictions**  

### **Output**  
This folder contains all the output files generated in the study:  

- **Aerial tiles**:  
  - Annotated tiles used for AI model training are stored in the `annotated_images_png` folder.  
  - Unannotated tiles used for AI model testing are located in the `unlabeled_images_png` folder.  

- **Aerial annotations**:  
  - Annotations derived from fine-scale predictions are saved as spatialized `.csv` files for easy integration and analysis.  

- **Aerial predictions**:  
  - AI model predictions on unlabeled tiles are organized within the `IA` subfolder and provided in two formats:  
    - A spatialized `.csv` file for data analysis.  
    - GIS-compatible `.tif` layers for seamless integration into geographic information systems.  

---

## **CODE**

- **DinoVdeau**:  
  This folder contains the code used to train the underwater (fine-scale) deep-learning model and the aerial (medium-scale) deep-learning model.  

- **Drone-upscaling**:  
  This folder contains the code used to generate aerial annotations from underwater predictions.  

- **predict_drone**:  
  This folder contains the code to make predictions with the aerial model on a test set or on unlabeled images.