Documentaton - BLOCKS - BLOCK_VISLOC

Visualize two-class localization (such as Graz-02) and find the precision=recall point.

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

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

Required input:

db

The database.

prediction

The output of the localization classifier.

Options:

bk.seg_prefix

The prefix path to the ground truth segmentations (Required).

bk.seg_ext

The extension of each ground truth segmentation (Required).

bk.cat_ids

A mapping from category names in the database to values in the ground truth segmentation images (Required).

bk.num_levels

The number of thresholds at which to evaluate performance. Default 40.

bk.levels

The actual thresholds to evalutate performance. Default [] uses num_levels to determine the thresholds automatically.

Fetchable attributes:

report

A structure containing a report on the results which has fields: seg_ids: The training segment ids. tp: The raw true positives. fp: The raw false positives. tn: The raw true negatives. fn: The raw false negatives. levels: The levels chosen. eer: The precision = recall point. precision: Precision for each level. recall: Recall for each leve.

eer_level

The precision = recall point.