Implementation of Implementation of 'Semantic Texton Forests for Image Categorization and Segmentation by Jamie Shotton, Matthew Johnson, Roberto Cipolla CVPR 08, based on C# code provided by Matthew Johnson. Pdf
Only implements the first tree that provides pixel level class distribution and bags of semantic textons.
do_stf.m is the driver that runs the entire experiment from preprocessing, training and testing. config.m has all the configuration of the experiment. Please change the path to MSRC21 or any kind of dataset you want to use to train the forest. For MSRC21, I used the training/testing split from here.
to test each components seperately you can run do_train('config') do_test('config')
Bag of semantic texton forest (BoST) is an unnormalized histogram that denotes the nodes in the forest that was visited by an image region. Example of extracting BoST is at the bottom of do_stf.m.