Volumetric bias in segmentation and reconstruction : Secrets and solutions
(2016) 15th IEEE International Conference on Computer Vision, ICCV 2015 11-18-December-2015. p.1769-1777- Abstract
Many standard optimization methods for segmentation and reconstruction compute ML model estimates for ap- pearance or geometry of segments, e.g. Zhu-Yuille [23], Torr [20], Chan-Vese [6], GrabCut [18], Delong et al. [8]. We observe that the standard likelihood term in these formu- lations corresponds to a generalized probabilistic K-means energy. In learning it is well known that this energy has a strong bias to clusters of equal size [11], which we express as a penalty for KL divergence from a uniform distribution of cardinalities. However, this volumetric bias has been mostly ignored in computer vision. We demonstrate signif- icant artifacts in standard segmentation and reconstruction methods due to this bias. Moreover, we propose... (More)
Many standard optimization methods for segmentation and reconstruction compute ML model estimates for ap- pearance or geometry of segments, e.g. Zhu-Yuille [23], Torr [20], Chan-Vese [6], GrabCut [18], Delong et al. [8]. We observe that the standard likelihood term in these formu- lations corresponds to a generalized probabilistic K-means energy. In learning it is well known that this energy has a strong bias to clusters of equal size [11], which we express as a penalty for KL divergence from a uniform distribution of cardinalities. However, this volumetric bias has been mostly ignored in computer vision. We demonstrate signif- icant artifacts in standard segmentation and reconstruction methods due to this bias. Moreover, we propose binary and multi-label optimization techniques that either (a) remove this bias or (b) replace it by a KL divergence term for any given target volume distribution. Our general ideas apply to continuous or discrete energy formulations in segmenta- tion, stereo, and other reconstruction problems.
(Less)
- author
- Boykov, Yuri ; Isack, Hossam ; Olsson, Carl LU and Ayed, Ismail Ben
- organization
- publishing date
- 2016-02-17
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- Proceedings - 2015 IEEE International Conference on Computer Vision, ICCV 2015
- volume
- 11-18-December-2015
- article number
- 7410563
- pages
- 9 pages
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- 15th IEEE International Conference on Computer Vision, ICCV 2015
- conference location
- Santiago, Chile
- conference dates
- 2015-12-11 - 2015-12-18
- external identifiers
-
- wos:000380414100198
- scopus:84973931568
- ISBN
- 9781467383912
- DOI
- 10.1109/ICCV.2015.206
- language
- English
- LU publication?
- yes
- id
- 632a92cc-920b-48d2-96d0-a523a8a2b1bc
- date added to LUP
- 2017-02-13 14:18:43
- date last changed
- 2025-01-07 06:42:53
@inproceedings{632a92cc-920b-48d2-96d0-a523a8a2b1bc,
abstract = {{<p>Many standard optimization methods for segmentation and reconstruction compute ML model estimates for ap- pearance or geometry of segments, e.g. Zhu-Yuille [23], Torr [20], Chan-Vese [6], GrabCut [18], Delong et al. [8]. We observe that the standard likelihood term in these formu- lations corresponds to a generalized probabilistic K-means energy. In learning it is well known that this energy has a strong bias to clusters of equal size [11], which we express as a penalty for KL divergence from a uniform distribution of cardinalities. However, this volumetric bias has been mostly ignored in computer vision. We demonstrate signif- icant artifacts in standard segmentation and reconstruction methods due to this bias. Moreover, we propose binary and multi-label optimization techniques that either (a) remove this bias or (b) replace it by a KL divergence term for any given target volume distribution. Our general ideas apply to continuous or discrete energy formulations in segmenta- tion, stereo, and other reconstruction problems.</p>}},
author = {{Boykov, Yuri and Isack, Hossam and Olsson, Carl and Ayed, Ismail Ben}},
booktitle = {{Proceedings - 2015 IEEE International Conference on Computer Vision, ICCV 2015}},
isbn = {{9781467383912}},
language = {{eng}},
month = {{02}},
pages = {{1769--1777}},
publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}},
title = {{Volumetric bias in segmentation and reconstruction : Secrets and solutions}},
url = {{http://dx.doi.org/10.1109/ICCV.2015.206}},
doi = {{10.1109/ICCV.2015.206}},
volume = {{11-18-December-2015}},
year = {{2016}},
}