Neuron segmentation and synapse detection using PyTorch
⚠️ This package is still under development, it was customized to process a Electron Microscopy volume although it could be potentially generalizable to other biological image volumes. Use at your own risk.
- Synapse detection
- Neuron boundary detection
- Semantic segmentation
- Image denoising by image inpainting
- Training using whole terabyte or even petabyte of image volume.
- Training using multiple version of image datasets as data augmentation.
- Data augmentation without zero-filling. The loaded patches before transformation/augmentation could be larger than the final training patch size. This size expansion is determined according to the transformation used. This careful adjustment of patch size ensures that there is no zero padding in the transformation and the distribution is closer to original dataset.
python setup.py install
We provide command line tool after installation. Check out the options provided:
neutrain-XXX --help
For example, to perform boundary detection training, we can use the following command:
neutrain-affs -c config.yaml
This package is built upon the following packages:
The development is supported by Flatiron Institute, a division of the Simons Foundation.