Instance Segmentation

After installing conda, we need to configure Magneton environment.

Based on Linux machines with NVIDIA GPUs

Create a Virtual Environment

conda create -y -n magneton python=3.9
conda activate magneton

Install pytorch with right cuda version (for H100 or A100: usually 12.4):

conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia

We have made modifications based on pytc (version 0.1). This section covers the basics of how to download and install this version.

git clone https://github.com/kuan-lab/magneton.git
cd magneton
pip install --editable .

cd pytorch_connectomics
pip install --editable .

cd ../waterz
pip install --editable .

If error: “waterz/backend/types.hpp:3:10: fatal error: boost/multi_array.hpp: No such file or directory”

conda install boost

If error: ModuleNotFoundError: No module named ‘Cython’

pip install cython

If error: ‘PyDataType_ELSIZE’ was not declared in this scope

pip install --upgrade numpy

If error: monai 1.4.0 requires numpy<2.0, >=1.24 After waterz installed

pip install numpy==1.26.4

If error: no module named “mahotas”

pip install mahotas

If error: ImportError: … libstdc++.so.6: version `GLIBCXX_3.4.32’ not found ..

conda install -c conda-forge libstdcxx-ng