Initial DDPM Code Commit for Sea Ice Forecasting (Integration in Progress)

[marianovitasari20]

Author: Maria Carolina Novitasari

This repository contains the implementation of a Denoising Diffusion Probabilistic Model (DDPM) to forecast sea ice concentration, conditioned on ERA5 reanalysis data. The model is trained and evaluated using OSISAF sea ice concentration observations as groundtruth. This work is still in progress, and changes are anticipated.

Current project structure

The current project structure is as follows:

-- notebooks/                            # Jupyter notebooks for training and experiments
   ├── 0_notebook_tf.ipynb               
   ├── 1_icenet_forecast_unet.ipynb      
   ├── 2_icenet_forecast_cgan.ipynb      
   ├── 3_icenet_forecast_DDPM.ipynb      
   └── ...                                # Additional experiments

-- models/                               # Model architectures
   ├── lit_diffusion.py                   # LightningModule for DDPM
   ├── unet_diffusion.py                  # U-Net backbone for diffusion
   ├── gaussian_diffusion.py              # Forward and reverse DDPM process
   └── __init__.py                        # Init file for module import

-- utils/                                # Helper functions and utilities
   ├── losses.py                          # Weighted MSE, BCEWithLogits, etc.
   ├── metrics.py                         # IceNetAccuracy, SIEError, etc.
   └── __init__.py                        # Init file for module import

-- trainers/                             # Training logic
   ├── train_diffusion.py                 # Training loop for DDPM
   └── __init__.py                        # Init file for module import

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB