CNN-based Super-Resolution for Ocean Current Data

This project demonstrates how to use a basic Convolutional Neural Network (CNN) for Super-Resolution (SR) reconstruction of ocean current data (e.g., u_bar and v_bar). The project includes a complete end-to-end example, a practice template for hands-on experience, and data processing scripts.

✨ Features

• End-to-End Workflow: Provides a complete workflow from data loading, model training, and validation to testing and visualization in a Jupyter Notebook (Task1_CNN_SuperResolution_example.ipynb).
• Basic CNN Model: Implements a simple yet effective CNN model tailored for the super-resolution task.
• Result Visualization: Generates comparison plots to visually showcase the differences between Low-Resolution input (LQ), High-Resolution ground truth (GT), and the model's Super-Resolution output (SR).
• Model & Prediction Saving: Automatically saves the best-performing model based on the validation set and can save partial predictions as .npy files for further analysis.
• Hands-on Practice: Includes a practice notebook (Task1_CNN_SuperResolution_practice.ipynb) to encourage users to replicate the entire process on the v_bar dataset.
• Data Utility: Contains a helper script (data_demo.py) to create smaller data subsets from the full dataset, facilitating quick debugging and experiments.

📂 Project Structure

.
├── data/
│   ├── u_bar/              # U-component ocean current data
│   │   ├── ubar_hr_train.npy
│   │   ├── ubar_sr_input_train.npy
│   │   └── ... (val, test datasets)
│   └── v_bar/              # V-component ocean current data
│       ├── vbar_hr_train.npy
│       └── ... (val, test datasets)
├── sr_cnn_results_ubar_demo/ # Output directory for the example code
│   ├── best_model.pth
│   ├── test_visualization.png
│   └── validation_npy_predictions/
│       └── ...
├── data_demo.py            # Script for creating a small-scale demo dataset
├── Task1_CNN_SuperResolution_example.ipynb # Core example: Complete SR workflow using u_bar data
├── Task1_CNN_SuperResolution_practice.ipynb # Practice task: Replicate the SR workflow on v_bar data
└── README.md               # This document

⚙️ Requirements

Please ensure you have the following Python libraries installed. You can install them using pip:

pip install torch numpy matplotlib

🚀 Usage Guide

1. Run the Core Example

Open and run Task1_CNN_SuperResolution_example.ipynb directly in a Jupyter environment.

• This notebook will use the data from the data/u_bar directory.
• It will perform model training, validation, and testing.
• All outputs (best model, visualization images, predicted .npy files) will be saved in the sr_cnn_results_ubar_demo/ directory.

2. Complete the Practice Task

Open Task1_CNN_SuperResolution_practice.ipynb and, following the prompts and referencing the example code, complete the code to build a new super-resolution task for the v_bar dataset.

3. Create a Demo Dataset (Optional)

If you want to use a smaller dataset for quick testing, you can run data_demo.py.

1. Open the data_demo.py file.
2. Modify the DEMO_SAMPLE_COUNT variable to define the number of samples for each dataset.
3. Modify the file paths in file_processing_list to ensure they point to your original data and desired output locations.
4. Run the script from your terminal:

   python data_demo.py

📊 Output

After running Task1_CNN_SuperResolution_example.ipynb, the following will be generated in the sr_cnn_results_ubar_demo/ directory:

• best_model.pth: The model weights file with the lowest loss on the validation set.
• test_visualization.png: A comparison plot with 5 subplots showing LQ, GT, SR, and the errors between them.
• validation_npy_predictions/: A subdirectory containing several .npy files, which save the LQ, GT, and SR data for some validation samples.

Getting Started

1. Clone/Download the Project: Get a local copy of this project.
2. Prepare the Data: Ensure the data/ directory contains the u_bar and v_bar data. If the data is missing, you may need to run a data preparation script or place it manually.
3. Install Dependencies: pip install torch numpy matplotlib.
4. Run the Example: Open and run Task1_CNN_SuperResolution_example.ipynb from start to finish to understand the entire workflow and see the results.
5. Hands-on Practice: Try to complete the exercises in Task1_CNN_SuperResolution_practice.ipynb to solidify your understanding.

📝 Notes

• Device Configuration: The code will automatically detect and prioritize using a CUDA-enabled GPU. If no GPU is available, it will fall back to CPU execution. Training on a CPU can be very time-consuming.
• Path Configuration: All paths in the notebooks and scripts are relative. Please ensure you run them from the project's root directory, or you may need to adjust the file paths.
• Data Format: This project uses the .npy file format for data storage. The data is assumed to be a 3D array of shape (num_samples, height, width), where height and width are the dimensions of the ocean current data fields.