Installation Guide

This guide will help you install U-FISH on your system.

Requirements

Python 3.9 or higher
pip package manager

Basic Installation

The simplest way to install U-FISH is via pip:

pip install ufish

This will install U-FISH with CPU support for inference.

GPU Support

For Inference

To enable GPU acceleration for inference, install the GPU version of ONNX Runtime:

pip install onnxruntime-gpu

For Training

If you plan to train or fine-tune models, you’ll need PyTorch with CUDA support:

# For CUDA 11.8
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

# For CUDA 12.1
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Visit the PyTorch official website for more installation options.

Windows with AMD GPU

If you’re using Windows with an AMD GPU, install the DirectML backend:

pip install torch-directml

Development Installation

To install U-FISH for development:

Clone the repository:

git clone https://github.com/GangCaoLab/U-FISH.git
cd U-FISH

Install in development mode:
```
pip install -e .
```
Install development dependencies:
```
pip install -e ".[dev]"
```

Optional Dependencies

Napari Plugin

To use the Napari plugin for visualization:

pip install napari[all]
pip install napari-ufish

Large-scale Data Support

For working with large-scale data formats:

# For Zarr support
pip install zarr

# For OME-Zarr support
pip install ome-zarr

# For N5 support
pip install pyn5

Verifying Installation

To verify that U-FISH is installed correctly:

import ufish
print(ufish.__version__)

You can also check the CLI:

ufish --help

Troubleshooting

Common Issues

ImportError: No module named ‘ufish’

Make sure you’ve activated the correct Python environment where U-FISH is installed.
CUDA out of memory errors

Try reducing the batch size or image size, or use CPU inference instead.
ONNX Runtime errors

Ensure you have the correct version of onnxruntime or onnxruntime-gpu installed.

Getting Help

If you encounter any issues:

Check the GitHub issues
Create a new issue with detailed information about your problem
Include your system information (OS, Python version, GPU model if applicable)