dive-wsu / dl_bigneuron

caffe for 3D Bio-medical image segmentation, particularly for BigNeuron Project

DL_BigNeuron

This code is modifided from master version of caffe(https://github.com/BVLC/caffe), particularlly for 3D neuron image segmentaion in BigNeuron Project (https://http://alleninstitute.org/bigneuron/about/)

Functionalities:

• 1. Fully Convolutional networks with 3D data segmentataion.
• 2 Fast image based predictions.
• 3. Selecting random patchs from 3d Volum to build batch for networks training, eliminating the need to create patch files for training.
• 4. Enabling the sampling balanced calsses for batch training by setting probability of each class in prototxt file
• 5. Setting weighted loss for each class
• 6. Including avaible inception-residual-fully convolutional archetecture, made particularlly for 3D neuron image segmentaion