🚀 DeepCAD RT Runners

Welcome to DeepCAD RT Runners! This project is a small UV-based solution that bundles all necessary dependencies for DeepCAD 1.2.0, along with convenient command-line scripts for configuration, training and prediction.

Key Features:

• 🔧 No manual Python environment setup required (just uv).
• 🖥️ Tested on Windows 11 and Debian Linux with CUDA-compatible GPUs.
• 📊 Seamless training and prediction workflows for denoising .tif movie files.

---

📋 Prerequisites

Before diving in, ensure you have:

• A system with a CUDA-compatible GPU (recommended for optimal performance).
• uv installed.
• Access to .tif movie files for training and testing.

---

🛠️ Installation

No installation needed! The project uses uvx to run commands directly from the GitHub repository. Your Python environment will be created and cached automatically.

---

📖 Usage

Follow these steps to train and test your models. Each step includes detailed instructions and examples.

1. 📝 Create Configuration Files

Generate local configuration files for training and testing.

Run the following command in your terminal:

uvx --from deepcadrt-run deepcadrt-config

This creates train_config.json and test_config.json in your current directory. Customize these files as needed (e.g., adjust parameters like patch size, number of epochs or learning rate).

Example Output:

• train_config.json: Default training settings.
• test_config.json: Default testing settings.

2. 🎯 Train Your Model

Prepare a folder containing your .tif movie files (e.g., data/my_movies/).

Edit train_config.json to match your requirements (leave dataset_path unchanged).

Run the training command:

uvx --from deepcadrt-run deepcadrt-train "mymovies" -c train_config.json

This will:

• Train a DeepCAD model on your data.
• Create a models/ folder with a subfolder named like mymovies_202310011155 (based on your data folder and current date).

Tips:

• Ensure your .tif files are properly formatted (e.g., 3D stacks).
• The patch size in the time dimension is smaller than the movie length
• Monitor GPU usage during training for performance.

3. 🔮 Predict and Denoise

Use your trained model to denoise new or existing data.

Edit test_config.json as needed (leave dataset_path and denoise_model unchanged).

Run the prediction command:

uvx --from deepcadrt-run deepcadrt-predict mymovies/ models/mymovies_202310011155 -c test_config.json

This will:

• Apply denoising to your movies.
• Save results in a results/ folder with a subfolder for your output.

Example:

• Input: Noisy .tif movies.
• Output: Denoised versions in results/. By default the model from the latest epoch is used.

---

❓ Troubleshooting

• CUDA Issues: Ensure your GPU drivers are up-to-date and compatible.
• Memory Errors: Reduce patch size or train_datasets_size in config files for large datasets.
• Command Not Found: Verify uv is installed and in your PATH.
• For more help, check the DeepCAD documentation or open an issue on this repo.

---

🤝 Contributing

We welcome contributions! Feel free to submit pull requests or report bugs via the GitHub repository.

---

📄 License

This project is licensed under MIT License. See the LICENSE file for details.