FAST is a real-time self-supervised denoising framework for fluorescence neural imaging, achieving enhanced image quality in low-SNR scenarios through spatiotemporal joint optimization. (https://doi.org/10.21203/rs.3.rs-6101322/v1)

• 🚀 Real-Time Processing: >1000 fps denoising (hardware-dependent)
• 🤖 Self-Supervised Learning: Eliminates need for clean ground truth
• 🔄 Spatiotemporal Optimization: Frame-multiplexing enhances SNR
• 📊 High Adaptability: Suitable for various fluorescence imaging data

• Python 3.9+
• PyTorch 2.x (CUDA version aligned with your hardware)
• CUDA-capable GPU (recommended)

• Windows 10: NVIDIA RTX 3060, RTX 3090, RTX 4070
• Ubuntu 22.04: NVIDIA RTX A6000
• CUDA Version: 12.4

numpy==1.24.1
torch==2.5.1
torchvision==0.20.1
torchaudio==2.5.1
scikit-image==0.24.0
tqdm==4.66.5
pyqt5==5.15.7
pyqt5-qt5==5.15.2
pyqt5-sip==12.11.0
csbdeep==0.8.1

# Create and activate environment
conda create -n FAST python=3.9
conda activate FAST
pip install -r requirements.txt

# Training mode
python main.py --config_path "./params.json"

# Testing mode (with pretrained model)
python main.py --config_path "./checkpoint/model_name/config.json" --test_path "./data/test/test_dir"

# Launch training GUI
python Train_GUI.py

# Launch testing GUI
python Test_GUI.py

FAST/
├── checkpoint/         # Model checkpoints
│   └── model_name
├── data/              # Data directory
│   ├── test/          # Testing data
│   └── train/         # Training data
├── datasets/          # Dataset processing
│   ├── dataAug.py     # Data augmentation
│   ├── data_process.py
│   └── dataset.py     # Dataset classes
├── environment.yml    # Environment configuration
├── FAST_logo.png     # Project logo
├── log.txt           # Runtime logs
├── main.py           # Main entry point
├── models/           # Model architectures
│   ├── baseLayers.py
│   ├── loss/         # Loss functions
│   │   └── loss.py
│   └── Unet_Lite.py  # Main model
├── params.json       # Configuration file
├── result/           # Output results
│   └── model_name
├── Test_GUI.py       # GUI for testing
├── test_in_gui.py
├── test.py          # Testing script
├── Train_GUI.py      # GUI for training
├── train_in_gui.py
├── train.py         # Training script
└── utils/           # Utility functions
    ├── config.py    # Configuration utils
    ├── fileSplit.py
    ├── general.py   # General utilities
    └── __init__.py

Customize model parameters by modifying params.json:

{
    "data_extension": "tif",
    "epochs": 100,
    "miniBatch_size": 4,
    "lr": 0.0001,
    "weight_decay": 0.9,
    "gpu_ids": "0",
    "train_frames": 2000,
    "data_type": "3D",
    "denoising_strategy": "FAST",
    "seed": 123,
    "save_freq": 25,
    "clip_gradients": 20.0,
    "num_workers": 0,
    "batch_size": 1
}

1. Prepare Training Data
   Place one or more .tif/.tiff files (each should be an xy-t stack, ideally >1000 frames per file) into a new folder under ./data/train/, e.g., ./data/train/MyTrainDataset/.

2. Edit Configuration
   Open params.json and set "train_folder": "path/to/MyTrainDataset". Adjust other parameters as needed—see the Configuration section for details on each option.

3. Start Training
   Run the following command to train the model:

   python main.py --config_path "./params.json"

   After training, model weights and the config file will be saved in ./checkpoint/model_name/.

1. Prepare Test Data
   Place one or more .tif/.tiff files (xy-t stacks to be denoised) into a folder under ./data/test/, e.g., ./data/test/MyTestDataset/.

2. Run Inference
   Use the following command to denoise your data with the trained model:

   python main.py --config_path "./checkpoint/model_name/config.json" --test_path "./data/test/MyTestDataset"

   After completion, you will find the processed results in the ./result/model_name directory.

We welcome contributions, particularly:

1. 🐛 Bug reports and fixes
2. ✨ New feature proposals and implementations
3. 📚 Documentation improvements
4. 🎨 Code optimizations

• Use UpperCamelCase for class names
• Use lowercase_with_underscores for functions and variables
• Include docstrings for core functions
• Follow PEP8 standards (validate using flake8)

This project is licensed under the GNU General Public License v3.0. This means you are free to:

• ✅ Use
• ✅ Modify
• ✅ Distribute

But you must:

• ⚠️ Disclose source
• ⚠️ Include original copyright
• ⚠️ Use the same license

See LICENSE file for full text.

• 📧 Email: yiqunwang22@fudan.edu.cn
• 🌐 Project Page: GitHub Repository

If you use FAST in your research, please cite our paper:

@article{wang2024real,
    title={Real-time self-supervised denoising for high-speed fluorescence neural imaging},
    author={Yiqun Wang and Others},
    journal={https://doi.org/10.21203/rs.3.rs-6101322/v1},
    year={2025}
}