ASETS Matlab repository

License:

BSD (see license.md)

Developers:

• Martin Rajchl (@mrajchl), Imperial College London (UK)
• John SH. Baxter (@jshbaxter), Robarts Research Institute (CAN)
• Jing Yuan, Robarts Research Institute (CAN)

Features:

• Fast parallel continuous max flow solvers in 2D/3D

  • Binary max flow
  • Multi-region (Potts model, Ishikawa model, Hierarchical Max Flow)
  • In two different implementations (full flow and pseudo flow solvers)

• Implemented in multiple languages

  • Matlab/mex/C
  • Matlab/CUDA

• Tutorials

  • T01 Binary graph cuts
  • T02 Multi-region color image segmentation with the Potts model
  • T03 Using different max flow implementations (Matlab, C, CUDA)

• Application examples for (medical) image segmentation:

  • Interactive max flow graph cuts (3D)
  • Regularization of probabilistic label maps as in atlas-based segmentation (3D)
  • High-performance multi-phase levelsets (3D)
  • Post-processing of flawed manual segmentations with constrast sensitive regularization (2D)
  • L1 intensity segmentation (2D)

Overview of folder structure:

./: Compile scripts, readme, license and todo list
./applications: Contains examples of typical applications in image segmentation and analysis
./data: Example data to run the applications
./lib: Is created by compile.m and contains the compiled C/mex files
./maxflow: Optimization code in C/mex and Matlab
./tests: Test scripts to compare different implementations against each other
./tutorials: Contains available tutorials

Compile/Installation instructions:

To compile the C/mex code run:

compile.m

which creates the folder ./lib. For testing purposes run any script in ./tests.

Tests:

• Matlab 2014a, 64-bit Linux (Ubuntu 12.04 LTS)
• Matlab 2015a, 64-bit Windows 7
• Matlab 2012a, 32-bit WinXP