featureforest

A napari plugin for segmentation using vision transformers' features.
We developed a napari plugin to train a Random Forest model using extracted embeddings of ViT models for input and just a few scribble labels provided by the user. This approach can do the segmentation of desired objects almost as well as manual segmentations but in a much shorter time with less manual effort.

---

Documentation¶

The plugin documentation is here.

Installation¶

It is highly recommended to use a python environment manager like conda to create a clean environment for installation.
You can install all the requirements using provided environment config files:

# for GPU
conda env create -f ./env_gpu.yml

# if you don't have a GPU
conda env create -f ./env_cpu.yml

Requirements¶

• python >= 3.9
• numpy
• opencv-python
• scikit-learn
• scikit-image
• matplotlib
• pyqt
• magicgui
• qtpy
• napari
• h5py
• pytorch=2.1.2
• torchvision=0.16.2
• timm
• pynrrd

If you want to install the plugin manually using GPU, please follow the pytorch installation instruction here.
For detailed napari installation see here.

Installing The Plugin¶

If you use the conda env.yml file, the plugin will be installed automatically. But in case you already have the environment setup, you can just install the plugin. First clone the repository:

git clone https://github.com/juglab/featureforest

Then run the following commands:

cd ./featureforest
pip install .

License¶

Distributed under the terms of the BSD-3 license, "featureforest" is free and open source software

Issues¶

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