NIPS2009_0084_slide - to directly minimize a...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
84. Maximin afFnity learning of image segmentation 1. Some local errors in boundary detection can have dramatic global consequences noisy image desired boundary predicted boundary (segmentation consistent) predicted boundary (segmentation inconsistent) catastrophic minor nuisance minor nuisance Boundary detectors for image segmentation are usually trained to minimize local boundary classiFcation error. But some local errors in boundary detection can have dramatic global consequences. We show how to train a boundary detector
Background image of page 1
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: to directly minimize a clustering-based global segmentation error, resulting in end-to-end segmentation learning . 2. Rand index measures global segmentation error Abstract end-to-end segmentation learning P P P lowest pixel minimum pixel along a path maximum over paths of the minimum pixel 3. Maximin paths are used to directly minimize global segmentation error 1 2 3 4 1 2 3 4 1’ 2’ 3’ 4’ 3 1’ 2’ 3’ 4’ merger split rand index groundtruth test all pairs!...
View Full Document

This note was uploaded on 02/12/2010 for the course COMPUTER S 10586 taught by Professor Jilinwang during the Fall '09 term at Zhejiang University.

Ask a homework question - tutors are online