TCP-Diffusion

This repository contains the testing code and a subset of the data for the TCP-Diffusion project.

🔧 Environment Setup

Configure the necessary environment. Recommended versions:

• Python 3.7
• PyTorch ≥ 1.10 with CUDA 10.2

---

📦 Download Model and Dataset Subset

Download the pre-trained model and the 2020 subset of the dataset via the following link:

Model

Dataset

Note:

• Only data from the year 2020 is included.

---

📁 Dataset Preparation

1. Extract the downloaded dataset to a path of your choice, e.g., your_data_path.
2. Edit line 17 of the file: video_diffusion_pytorch/rainfall_dataset_ICML.py
3. Replace the default path with your actual dataset location.

---

📥 Model Placement

Place the pre-trained model model-55.pt into the following directory: results/TCP_ICML_Test/

---

🚀 Run Evaluation

To run the evaluation, use the following command:

python evaluation_new_F4noinit_E1_ifscat_0321_ICML.py ICMLtest \
    --save results/TCP_ICML_Test \
    --output_frames 4 \
    --train_batch_size 16 \
    --test_epoch 55 \
    --timesteps 200 \
    --new_split \
    --multi_modals tquvz_t2m_sst_msl_topo_ifs \
    --input_transform_key loge \
    --loss_type l2

🖼️ Output

A subset of visualization results will be saved to: results/TCP_ICML_Test/ICML_under_review_55/predictions_55/