Documentaton - BLOCKS - BLOCK_TRAIN_SEL

Superpixel CRF, as proposed in Fulkerson et. al 2009.

BK = BLOCK_TEST_SEGCRF() Initializes the block with the default options.

BK = BLOCK_TEST_SEGCRF(BK) Executes the block with options and inputs BK.

Required inputs:

db

The database.

hist

Segment histograms.

Options:

bk.hists_per_im

The number of histograms to select from each training image. If hists_per_cat is set, this has no effect. Default 50

bk.hists_per_cat

The number of histograms to select per category. Default [] uses hist_per_im.

bk.seg_neighbors

The size of the superpixel neighborhood. Default 0.

bk.rand_seed

Set the random seed before proceeding. Default [] does not change the random seeds.

Fetchable attributes:

train_ids

Returns [train_ids labels]. train_ids is a Nx2 vector, where the first column denotes the seg_id and the second column denotes the superpixel.