Goal: the purpose of this work is to generate 3d segmentation of synapses on hair cells in the cochlea. Current training has focused on pre-synaptic (CTBP2) segmentation of inner hair cells only, but can be expanded to include post-synaptic signals and outer hair cells. Current training data also includes volumes with varying resolution from varying microscope modalities (laser scanning confocal, spinning disk confocal, and airyscan).
- Data conversion: data is converted to zarr using scripts in 01_data
- Training and prediction: training and prediction scripts for 2D, 2.5D, and 3D models in 02_train. (Validation statistics also recorded at time of prediction.)
-
conda create -n synapses python=3.9
-
conda activate synapses
-
conda install pytorch=1.12 torchvision torchaudio cudatoolkit=10.2 -c pytorch
-
python -m pip install zarr scikit-image neuroglancer imagecodecs tensorboard tensorboardX pandas
-
python -m pip install git+https://github.com/funkey/gunpowder.git
-
python -m pip install git+https://github.com/funkelab/funlib.learn.torch.git
-
python -m pip install git+https://github.com/funkelab/funlib.show.neuroglancer.git
-
python -m pip install git+https://github.com/funkey/waterz.git