SPDiff

This repository contains the code for the paper "SPDiff: Semantic Proposal-guided Diffusion for Reliable Asymmetric Tropical Cyclone Estimation," currently under review at KDD AI4S 2027.

SPDiff is a conditional diffusion model for tropical cyclone intensity and asymmetric wind-radii estimation. The model uses multi-modal TC data to estimate:

• Maximum sustained wind (msw)
• Minimum sea-level pressure (mslp)
• Radius of maximum wind (rmw)
• Quadrant wind radii for 34-, 50-, and 64-knot winds

Project Structure

SPDiff_Code/
├── env/
│   ├── Config.py                  # Paths and experiment settings
│   ├── Dataset.py                 # Dataset loading and preprocessing
├── utils/
│   ├── Condition_Encoder.py       # Conditional feature encoder
│   ├── Diffusion.py               # Diffusion process
│   └── TwoDenoiseModel.py         # Denoising network
├── TrainNetworkHelper_TwoTG.py    # Training utilities
├── Train_TwoTGCD_Conf.py          # Training entry point
└── Test_TwoTGCD_Conf.py           # Evaluation entry point

Requirements

• Python 3.9+
• PyTorch
• torchvision
• NumPy
• Matplotlib
• tqdm
• einops
• OpenAI CLIP

Install the main dependencies with:

pip install torch torchvision numpy matplotlib tqdm einops
pip install git+https://github.com/openai/CLIP.git