GPU accelerated spot finding and localization for both skewed (diSPIM, lattice, OPM) and standard geometry (widefield, confocal) 3D microscopy data.
Create a python>=3.10 and CUDA >=11.2, <12.0 environment. We will use conda (or mamba if you prefer) for the CUDA and CuPy installation.
For Linux OS:
conda create -n spots3d python=3.10
conda activate spots3d
conda install -c conda-forge -c rapidsai -c nvidia cudatoolkit=11.8 cupy cucim=24.02 cuda-version=11.8 cudnn cutensor
For Windows OS:
conda create -n spots3d python=3.10
conda activate spots3d
conda install -c conda-forge -c nvidia cudatoolkit=11.8 cupy cuda-version=11.8 cudnn cutensor
On Windows OS, the required skimage portion of cuCIM will be built and installed during spots3d installation.
Then, build and install the qi2lab branch of Gpufit.
Finally, install spots3d
pip install spots3d@git+https://git@github.com/qi2lab/spots3d@main#egg=spots3d
An optional, but recommended napari plugin for spot localization, that uses SPOTS3D and the qi2lab branch of Gpufit, is found at napari-spot-detection.