Code Downloads

Archives of stable Darwin source code releases can be downloaded from the following links:

release 1.9 [ download ]
release 1.8 [ download ]
release 1.7 [ .tar.gz ]
release 1.6.1 [ .tar.gz ], release 1.6 [ .tar.gz ]
release 1.5.1 [ .tar.gz ], release 1.5 [ .tar.gz ]
release 1.4 [ .tar.gz ]
release 1.3.2 [ .tar.gz ], release 1.3.1 [ .tar.gz ], release 1.3 [ .tar.gz ]
release 1.2 [ .tar.gz ]
release 1.1.3 [ .tar.gz ], release 1.1.2 [ .tar.gz ], release 1.1.1 [ .tar.gz ], release 1.1 [ .tar.gz ]
release 1.0.2 [ .tar.gz ], release 1.0.1 [ .tar.gz ], release 1.0 [ .tar.gz ]
release 0.9 [ .tar.gz ]
release 0.8.1 [ .tar.gz ], release 0.8 [ .tar.gz ]
release 0.7.1 [ .tar.gz ], release 0.7 [ .tar.gz ]
release 0.6.2 [ .tar.gz ], release 0.6.1 [ .tar.gz ], release 0.6 [ .tar.gz ]
release 0.5 [ .tar.gz ]
release 0.4 [ .tar.gz ]
release 0.3 [ .tar.gz ]

See Also: Linux Installation Instructions; Windows Installation Instructions; Mac OS X Installation Instructions

Pre-compiled Executables

Archives of pre-compiled Windows executables can be downloaded from the following links:

release 1.5 [ .zip ]
release 1.3.2 [ .zip ]

Note that these executables require the Microsoft Visual C++ 2010 runtime libraries. If you do not have Visual Studio 2010 installed on your computer you can install the runtime libraries by downloading the Microsoft Visual C++ 2010 Redistributable Package.

Pre-compiled Matlab mex files for various architectures can be downloaded from the following links:

release 1.5 [ .mexw32 | .mexw64 | .mexglx ]

Change Log

Release 1.10

Release 1.9

Replaced drwnInPaint class with drwnImageInPainter class and added inPaint application
Added function to read CIFAR-10 and CIFAR-100 style datasets (see http://www.cs.utoronto.ca/~kriz/cifar.html)
Added drwnMaskedPatchMatch, drwnBasicPatchMatch, drwnSelfPatchMatch and basicPatchMatch application
drwnPatchMatchGraph now allows multiple matches to the same image
Upgraded wxWidgets to 3.0.2 (problems on Mac OS X)
Switched Mac OS X compilation to libc++ instead of libstdc++
Added Python scripts for running experiments and regression tests
Refactored drwnGrabCutInstance class to support both GMM and colour histogram model
Added cacheSortIndex to drwnDecisionTree for trading-off speed versus memory usage
Added mexLoadPatchMatchGraph for loading drwnPatchMatchGraph objects into Matlab
Improved documentation, other bug fixes and performance improvements

Release 1.8

Added Superpixel Graph Label Transfer (nnGraph) Project project
Added Python scripts for automating some projects
Added ability to pre-process features on-the-fly with one drwnFeatureTransform when applying or learning another drwnFeatureTransform
Fixed race condition in Windows threading (thanks to Edison Guo)
Switched Windows and Linux to build against OpenCV 2.4.9
changed drwnMAPInference::inference to return upper and lower energy bounds
Added pruneRounds function to drwnBoostedClassifier
Added drwnSLICSuperpixels function
Added drwnIndexQueue class
mexLearnClassifier and mexAnalyseClassifier now support integer label types
Bug fix in mexSaveSuperpixels to support single channel

Release 1.7

Log file now shows the command line
Utility application added for viewing multi-class segmentation legend
Added LBP filter response features to multi-class segmentation model
Added drwnColourHistogram class
Added k-means segmentation method for creating superpixels
Application visualizeSuperpixels and mex routines for loading and saving superpixels
Improved mex parsing of Matlab objects to support more matrix types
Bug fix in drwnOptimizer (thanks to Subarna Tripathi)
Updated copyright notice to 2007-2015
Other bug fixes and performance improvements

Release 1.6.1

Maximum size of drwnShowDebuggingImage can be set from command line
Windows MSVC projects updated to link against OpenCV 2.4.6
Fixes for gcc 4.7 (thanks to Sarma Tangirala)
Bug fixes and performance improvements

Release 1.6

Changed vision code from OpenCV 1.x C API to OpenCV 2.x C++ API
Added drwnHistogram class by Jason Corso
Added separate EPSG, EPSF and EPSX parameters to drwnOptimizer and changed signature of solve function
Added "-outUnary" option to inferPixelLabels for writing out unary potentials
Improved Matlab mex interfaces
Added drwnFeatureTransformFactory and improved drwnFactory class
Added drwnLinearTransform class
Bug fixes and performance improvements

Release 1.5.1

Option in drwnPCA to not normalize output space
drwnKMeans supports weighted training
Bug fixes and performance improvements

Release 1.5

Win32 threading implementation (drwnThreadPool)
Added standard command line option for setting random seed
Made drwnPersistentStorage thread safe (on Linux and Mac OS X)
Added drwnAverageRegions function
Added fast superpixel code (drwnFastSuperpixels)
Implemented drwnPersistentRecord interface for drwnSuperpixelContainer
Enhanced drwnSuperpixelContainer with additional member functions
Added image inpainting routines (drwnInPaint)
Bug fixes and performance improvements

Release 1.4

Added a dense linear program solver (drwnLPSolver)
Added a sparse linear program solver (drwnSparseLPSolver)
Upgraded to Eigen 3.1.1
Added sparse dot product (drwnSparseVec::dot)
Changed drwnPersistentStorage to be thread safe
Improved installation documentation particularly for Windows
Bug fixes and performance improvements

Release 1.3.2

Improved performance of multi-class image segmentation by spatially sampling pixels during learning of unary terms. This results in about 3% higher accuracy on the MSRC dataset (see Multi-class Image Segmentation)

Release 1.3.1

Documentation bug fixes
Matlab mex interface improvements (see Matlab Interfaces)

Release 1.3

Added PatchMatch algorithm (Barnes et al., ECCV 2010) (see PatchMatchGraph Project)
Added PatchMatchGraph data structure and label transfer (Gould and Zhang, ECCV 2012) (see PatchMatchGraph Project)
Added Matlab mex interface to max-flow algorithm
XML code tidy up
Improved user documentation
Bug fixes and performance enhancements

Release 1.2

Improved user documentation
Switched to RapidXML for XML parsing
Added miscellaneous features: drwn::drwnRemoveFile, drwnDataset::subSample, drwnConnectedComponents
Bug fixes and performance enhancements

Release 1.1.3

Fixed bug in compressed drwnPersistentStorage

Release 1.1.2

Fixed bug in writing drwnCompositeClassifier

Release 1.1.1

Fixed bug in drwnPCA and drwnFisherLDA modules for non-zero mean data

Release 1.1

Changed from Eigen 2.0.17 to Eigen 3.0.5
Changed from OpenCV 2.1 to OpenCV 2.4
Added drwnKMeans clustering class
Experimental GUI now works on Linux, Mac OS X and Windows
Bug fixes and performance enhancements

Release 1.0.2

Fixed bug in matlab factor graph inference

Release 1.0.1

Mac OS X install fix
Windows debug mode compilation fix

Release 1.0

First major release