Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections

Philipe Ambrozio Dias; Henry Medeiros

doi:10.1007/978-3-030-20890-5_9

Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections

Philipe Ambrozio Dias
, Henry Medeiros

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

16 Scopus citations

Abstract

The semantic segmentation produced by most state-of-the-art methods does not show satisfactory adherence to object boundaries. Methods such as fully-connected conditional random fields (CRFs) can significantly refine segmentation predictions. However, they rely on supervised parameter optimization that depends upon specific datasets and predictor modules. We propose an unsupervised method for semantic segmentation refinement that takes as input the confidence scores generated by a segmentation network and re-labels pixels with low confidence levels. More specifically, a region growing mechanism aggregates these pixels to neighboring areas with high confidence scores and similar appearance. To minimize the impact of high-confidence prediction errors, our algorithm performs multiple growing steps by Monte Carlo sampling initial seeds in high-confidence regions. Our method provides both running time and segmentation improvements comparable to state-of-the-art refinement approaches for semantic segmentation, as demonstrated by evaluations on multiple publicly available benchmark datasets.

Original language	English
Title of host publication	Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers
Editors	Greg Mori, C.V. Jawahar, Konrad Schindler, Hongdong Li
Publisher	Springer Verlag
Pages	131-146
Number of pages	16
ISBN (Print)	9783030208899
DOIs	https://doi.org/10.1007/978-3-030-20890-5_9
State	Published - 2019
Externally published	Yes
Event	14th Asian Conference on Computer Vision, ACCV 2018 - Perth, Australia Duration: Dec 2 2018 → Dec 6 2018

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11362 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	14th Asian Conference on Computer Vision, ACCV 2018
Country/Territory	Australia
City	Perth
Period	12/2/18 → 12/6/18

Funding

We acknowledge the support of USDA ARS agreement #584080-5-020, and of NVIDIA Corporation with the donation of the GPU used for this research.

Keywords

Instance semantic segmentation
Segmentation refinement
Unsupervised post-processing

Access to Document

10.1007/978-3-030-20890-5_9

Cite this

Dias, P. A., & Medeiros, H. (2019). Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections. In G. Mori, C. V. Jawahar, K. Schindler, & H. Li (Eds.), Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers (pp. 131-146). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11362 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-20890-5_9

Dias, Philipe Ambrozio ; Medeiros, Henry. / Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections. Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. editor / Greg Mori ; C.V. Jawahar ; Konrad Schindler ; Hongdong Li. Springer Verlag, 2019. pp. 131-146 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{77bd20bb1b1d48f59550f310fb17d9e8,

title = "Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections",

abstract = "The semantic segmentation produced by most state-of-the-art methods does not show satisfactory adherence to object boundaries. Methods such as fully-connected conditional random fields (CRFs) can significantly refine segmentation predictions. However, they rely on supervised parameter optimization that depends upon specific datasets and predictor modules. We propose an unsupervised method for semantic segmentation refinement that takes as input the confidence scores generated by a segmentation network and re-labels pixels with low confidence levels. More specifically, a region growing mechanism aggregates these pixels to neighboring areas with high confidence scores and similar appearance. To minimize the impact of high-confidence prediction errors, our algorithm performs multiple growing steps by Monte Carlo sampling initial seeds in high-confidence regions. Our method provides both running time and segmentation improvements comparable to state-of-the-art refinement approaches for semantic segmentation, as demonstrated by evaluations on multiple publicly available benchmark datasets.",

keywords = "Instance semantic segmentation, Segmentation refinement, Unsupervised post-processing",

author = "Dias, \{Philipe Ambrozio\} and Henry Medeiros",

note = "Publisher Copyright: {\textcopyright} 2019, Springer Nature Switzerland AG.; 14th Asian Conference on Computer Vision, ACCV 2018 ; Conference date: 02-12-2018 Through 06-12-2018",

year = "2019",

doi = "10.1007/978-3-030-20890-5\_9",

language = "English",

isbn = "9783030208899",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "131--146",

editor = "Greg Mori and C.V. Jawahar and Konrad Schindler and Hongdong Li",

booktitle = "Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers",

}

Dias, PA & Medeiros, H 2019, Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections. in G Mori, CV Jawahar, K Schindler & H Li (eds), Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11362 LNCS, Springer Verlag, pp. 131-146, 14th Asian Conference on Computer Vision, ACCV 2018, Perth, Australia, 12/2/18. https://doi.org/10.1007/978-3-030-20890-5_9

Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections. / Dias, Philipe Ambrozio; Medeiros, Henry.
Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. ed. / Greg Mori; C.V. Jawahar; Konrad Schindler; Hongdong Li. Springer Verlag, 2019. p. 131-146 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11362 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections

AU - Dias, Philipe Ambrozio

AU - Medeiros, Henry

PY - 2019

Y1 - 2019

N2 - The semantic segmentation produced by most state-of-the-art methods does not show satisfactory adherence to object boundaries. Methods such as fully-connected conditional random fields (CRFs) can significantly refine segmentation predictions. However, they rely on supervised parameter optimization that depends upon specific datasets and predictor modules. We propose an unsupervised method for semantic segmentation refinement that takes as input the confidence scores generated by a segmentation network and re-labels pixels with low confidence levels. More specifically, a region growing mechanism aggregates these pixels to neighboring areas with high confidence scores and similar appearance. To minimize the impact of high-confidence prediction errors, our algorithm performs multiple growing steps by Monte Carlo sampling initial seeds in high-confidence regions. Our method provides both running time and segmentation improvements comparable to state-of-the-art refinement approaches for semantic segmentation, as demonstrated by evaluations on multiple publicly available benchmark datasets.

AB - The semantic segmentation produced by most state-of-the-art methods does not show satisfactory adherence to object boundaries. Methods such as fully-connected conditional random fields (CRFs) can significantly refine segmentation predictions. However, they rely on supervised parameter optimization that depends upon specific datasets and predictor modules. We propose an unsupervised method for semantic segmentation refinement that takes as input the confidence scores generated by a segmentation network and re-labels pixels with low confidence levels. More specifically, a region growing mechanism aggregates these pixels to neighboring areas with high confidence scores and similar appearance. To minimize the impact of high-confidence prediction errors, our algorithm performs multiple growing steps by Monte Carlo sampling initial seeds in high-confidence regions. Our method provides both running time and segmentation improvements comparable to state-of-the-art refinement approaches for semantic segmentation, as demonstrated by evaluations on multiple publicly available benchmark datasets.

KW - Instance semantic segmentation

KW - Segmentation refinement

KW - Unsupervised post-processing

UR - https://www.scopus.com/pages/publications/85067355645

U2 - 10.1007/978-3-030-20890-5_9

DO - 10.1007/978-3-030-20890-5_9

M3 - Conference contribution

AN - SCOPUS:85067355645

SN - 9783030208899

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 131

EP - 146

BT - Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers

A2 - Mori, Greg

A2 - Jawahar, C.V.

A2 - Schindler, Konrad

A2 - Li, Hongdong

PB - Springer Verlag

T2 - 14th Asian Conference on Computer Vision, ACCV 2018

Y2 - 2 December 2018 through 6 December 2018

ER -

Dias PA, Medeiros H. Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections. In Mori G, Jawahar CV, Schindler K, Li H, editors, Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. Springer Verlag. 2019. p. 131-146. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-20890-5_9

Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this