Supervised quality evaluation of binary partition trees for object segmentation

The binary partition tree (BPT) allows for the hierarchical representation of images in a multiscale way, by providing nodes corresponding to image regions. In particular, cuts BPT can be interpreted as segmentations associated image. Building an then constitutes relevant preliminary step optimization-based segmentation methods. A wide literature has been devoted construction BPTs, and their involvement such tasks. Comparatively, there exist few works dedicated evaluate quality i.e. ability allow further methods compute good results. We propose framework evaluating with respect object problem, one or several objects from This is supervised, since notion not only depending on application but also user’s objectives, expressed via chosen ground-truth metric. develop two sides within this framework. First, we intrinsic analysis, that relies structural coherence ground-truth. More precisely, what extent structure well-matching examples, set / combinatorial fashion. Second, extrinsic allowing user assess based metrics correspond desired properties subsequent segmentation. provide results whereas handling trade-off cardinality cuts.