Deep MIB - segmentation using Deep Learning

The deep learning tool (Deep MIB) provides access to training of deep convolutional networks over the user data and utilization of those networks for image segmentation tasks.

Back to Index --> User Guide --> Menu --> Tools Menu

Overview
Example networks
Network panel
Directories and Preprocessing tab
Train tab
Predict tab
Options tab

Overview

For details of deep learning with DeepMIB please refer to the following tutorials:
The newest tutorial:

Deep-learning segmentation using 2.5D Depth-to-Colors workflow in MIB
Older tutorials:

DeepMIB: 2D U-net for image segmentation

DeepMIB: 3D U-net for image segmentation

DeepMIB, features and updates in MIB 2.80 (a recommended to see recommended workflow without preprocessing)

DeepMIB, 2D Patch-wise mode
Trained networks and examples:

Deep learning segmentation projects of FIB-SEM dataset of a U2-OS cell

The typical semantic segmentation workflow consists of two parts:

network training
image prediction

During network training, users specify the type of network architecture (the Network panel of Deep MIB) and provide images and ground truth labels (the Directories and Preprocessing tab). For training, the provided data will be split into two sets: one set to be used for the actual training (normally it contains most of the ground truth data) and another for validation. The network trains itself over the training set while checking its own performance using the validation set (the Training tab).
The pretrained network is saved to disk and can be distributed to predict unseen datasets (the Predict tab).
Please refer to the documentation below for details of various options available in DeepMIB

For the list of available workflows and networks jump to description of the Network panel

Example networks

Number of demo trainined DeepMIB projects are available for download and tests. Navigate to Menu->File->Example datasets->DeepMIB projects.
Detailed information about these projects is available from Menu-File

Network panel

This panel is used to select workflow and convolutional network architecture to be used during training

Details of the Network panel

Directories and Preprocessing tab

This tab allows choosing directories with images for training and prediction as well as various parameters used during image loading and preprocessing.

Details of the Directories and Preprocessing tab