Neuroproof Tool

I2G Tool: NeuroProof

Description

Neuroproof is an image segmentation software tool developed by the FlyEM Project at Janelia Farm Research Campus. (https://github.com/janelia-flyem/NeuroProof) Neuroproof allows for agglomerating over-segmented EM volumes. It is best used in conjunction with the GALA tool to provide accurate dense segmentations.

How to build

1. cd saber/saber/i2g/neuroproof
2. docker build -t aplbrain/neuroproof .

Features

Our dockerized NeuroProof tool allows for training and deployment and agglomeration classifier.

The training mode (mode: 0) requires the following inputs:

• over-segmented (watershed) volume file as HDF5 or numpy array (ws_file)
• respective groundtruth file as HDF5 or numpy array (gt_file)
• prediction file that contains multiple channels of class predictions* (pred_file)
• (Optional) Number of iterations for training (num_iterations: 1)
• (Optional) Whether or not to train for a mitochondria label (use_mito: 0)

The training mode outputs an agglomeration classifier (as an HDF5 File).

The deployment mode (mode: 1) requires the following inputs:

• over-segmented (watershed) volume file as HDF5 or numpy array (ws_file)
• prediction file that contains multiple channels of class predictions* (pred_file)
• agglomeration classifier as an HDF5 file (class_file)

The deployment mode outputs a segmented volume as a numpy file.

Our current tool comes pre-packaged with an agglomeration classifier trained on the Kasthuri AC4 Dataset.

Tip: Binary prediction files can be generated with the membrane detection tool.