What’s the point: Semantic segmentation with point supervision

Amy Bearman, Olga Russakovsky, Vittorio Ferrari, Li Fei-Fei

Research output: Chapter in Book/Report/Conference proceedingConference contribution

594 Scopus citations

Abstract

The semantic image segmentation task presents a trade-off between test time accuracy and training time annotation cost. Detailed per-pixel annotations enable training accurate models but are very timeconsuming to obtain; image-level class labels are an order of magnitude cheaper but result in less accurate models. We take a natural step from image-level annotation towards stronger supervision: we ask annotators to point to an object if one exists. We incorporate this point supervision along with a novel objectness potential in the training loss function of a CNN model. Experimental results on the PASCAL VOC 2012 benchmark reveal that the combined effect of point-level supervision and objectness potential yields an improvement of 12.9% mIOU over image-level supervision. Further, we demonstrate that models trained with pointlevel supervision are more accurate than models trained with image-level, squiggle-level or full supervision given a fixed annotation budget.

Original languageEnglish (US)
Title of host publicationComputer Vision - 14th European Conference, ECCV 2016, Proceedings
EditorsBastian Leibe, Jiri Matas, Nicu Sebe, Max Welling
PublisherSpringer Verlag
Pages549-565
Number of pages17
ISBN (Print)9783319464770
DOIs
StatePublished - 2016
Externally publishedYes
Event14th European Conference on Computer Vision, ECCV 2016 - Amsterdam, Netherlands
Duration: Oct 8 2016Oct 16 2016

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9911 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference14th European Conference on Computer Vision, ECCV 2016
Country/TerritoryNetherlands
CityAmsterdam
Period10/8/1610/16/16

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science

Keywords

  • Data annotation
  • Semantic segmentation
  • Weak supervision

Fingerprint

Dive into the research topics of 'What’s the point: Semantic segmentation with point supervision'. Together they form a unique fingerprint.

Cite this