napari-roxas-ai

A plugin that integrates the ROXAS AI analysis methods for quantitative wood anatomy in the napari platform.

Note: This plugin is currently under development and not yet stable nor officially released.

---

This napari plugin was generated with copier using the napari-plugin-template.

Installation

Environment setup

It's recommended to create a dedicated Python environment for napari-roxas-ai:

1. Install Miniconda or Miniforge if you don't have it already:

   • Miniconda Installation Guide
   • Miniforge Installer Repo

2. Create a new environment:

conda create -n roxas-ai python=3.12
conda activate roxas-ai

Installation

Install napari-roxas-ai via pip:

pip install napari-roxas-ai

Launching the plugin

Once installed, you can launch napari with the roxas-ai plugin (from within the activated environment):

napari

Verifying installation

To check if the plugin is working correctly:

1. Go to File > Open Sample > ROXAS AI in the napari interface.
2. The first time you open a sample, it may take some time as sample data and model weights are being downloaded. Progress will be logged in the terminal.
3. After the downloads, a sample made of three layers should open in the viewer

GPU Support

Model inference runs on the CPU by default. CPU inference works everywhere but is slow on large images; enabling GPU acceleration is strongly recommended if you have a compatible GPU.

On Linux and Windows systems (NVIDIA GPUs): The default installation ships a CPU-only build of PyTorch. To enable your NVIDIA GPU, reinstall PyTorch from the CUDA wheel index:

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128

This single command covers all recent NVIDIA GPUs (RTX 20-series and newer, including the RTX 50-series); the CUDA runtime is bundled with the wheels, so no separate CUDA toolkit installation is required. You only need an up-to-date NVIDIA driver (Linux, Windows). For older GPUs or drivers, pick the matching command from the PyTorch Installation Guide instead.

On macOS: CUDA is not available for macOS, however, PyTorch is supporting GPU acceleration for Apple Silicon devices (M1 and newer) via MPS. There is no need to install drivers or CUDA. Note that the MPS acceleration can currently only be leveraged for the cell segmentation model.

Finally, enable GPU support in the napari-roxas-ai settings within the napari interface. Go to Plugins > ROXAS AI > ZZ - Settings, open the processing section, tick try_to_use_gpu and try_to_use_autocast, and press Apply. The settings take effect immediately and no restart is needed, but a segmentation or measurement run that is already in progress keeps the settings it started with.

Contributing

Contributions are very welcome. Tests are automatically run with tox, please ensure the coverage at least stays the same before you submit a pull request.

Contributor Installation

In order to contribute to the development of the plugin the installation can be done as follows:

1. Create an environment

conda create -n roxas-ai python=3.12
conda activate roxas-ai

2. In the cloned / forked plugin directory, install the plugin dependencies

pip install -e .

3. Install the testing dependencies, as well as the napari plugin engine

pip install -e ".[testing]"
pip install npe2

4. Install pre-commit for quality checks

pip install pre-commit
pre-commit install

Documentation: Plugin Template and Development

You can find more information on the plugin template on the napari-plugin-template repository. You can find more information on plugin contributions and how to create plugins on the plugins section of the napari documentation.

License

Distributed under the terms of the GNU GPL v3.0 license, "napari-roxas-ai" is free and open source software

Issues

If you encounter any problems, please file an issue along with a detailed description.