This is the code that implements the method described in the paper Supervised Feature Learning for Curvilinear Structure Segmentation
It runs on MATLAB and contains Linux 64-bit mex files.
You may need to compile libteem for some of the mex files to run.
The code with the 2D implementation of our approach, containing the DRIVE dataset as well, can be downloaded HERE
How to use
To get started, uncompress the downloaded file, go to the uncompressed folder and open MATLAB.
To train a model on the DRIVE dataset, run KernelBoost.m. It will automatically train a classifier, and then apply it to the test images in the DRIVE dataset. The final results can be found in the results/ folder.
Important note: the DRIVE dataset included contains eroded masks for both train and test, generated from the masks in the vanilla DRIVE dataset, to ignore pixels outside the region of interest of each image.