2021
DOI: 10.48550/arxiv.2101.08418
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Rethinking Semantic Segmentation Evaluation for Explainability and Model Selection

Yuxiang Zhang,
Sachin Mehta,
Anat Caspi

Abstract: Semantic segmentation aims to robustly predict coherent class labels for entire regions of an image. It is a scene understanding task that powers real-world applications (e.g., autonomous navigation). One important application, the use of imagery for automated semantic understanding of pedestrian environments, provides remote mapping of accessibility features in street environments. This application (and others like it) require detailed geometric information of geographical objects. Semantic segmentation is a … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1

Citation Types

0
4
0

Year Published

2022
2022
2023
2023

Publication Types

Select...
3
1

Relationship

0
4

Authors

Journals

citations
Cited by 4 publications
(4 citation statements)
references
References 27 publications
0
4
0
Order By: Relevance
“…Additionally, we employed the region-wise over-segmentation measure (ROM) and region-wise under-segmentation measure (RUM) recently proposed in [ 62 ]. ROM and RUM enable us to quantitatively measure the over- and under-segmentation characteristics of the models, providing a more objective evaluation compared to previous qualitative assessments.…”
Section: Methodsmentioning
confidence: 99%
See 1 more Smart Citation
“…Additionally, we employed the region-wise over-segmentation measure (ROM) and region-wise under-segmentation measure (RUM) recently proposed in [ 62 ]. ROM and RUM enable us to quantitatively measure the over- and under-segmentation characteristics of the models, providing a more objective evaluation compared to previous qualitative assessments.…”
Section: Methodsmentioning
confidence: 99%
“…We believe that the SBCB framework enables the backbone to learn and preserve boundary-aware features, which results in segmentation masks with better quality around the mask boundaries. In this section, we evaluate the effects of the SBCB framework in terms of overand under-segmentation using the recently proposed region-based over-segmentation measure (ROM) and region-based under-segmentation measure (RUM) [62]. The results are presented in Table 18, where lower ROM and RUM values indicate better segmentation quality, reflecting reduced over-and under-segmentation, respectively.…”
Section: Does Sbcb Improve Segmentation Around Boundaries?mentioning
confidence: 99%
“…In order to quantify the performance, appropriate evaluation measurements are required. However, it has been demonstrated that current measurements have limitations when covering certain edge cases like weak labels [ 4 , 5 , 6 , 7 , 8 ] when a metric is used in isolation. Weak labels, i.e., data with no area of interest in the segmentation mask, are the cases we want to focus on in this paper.…”
Section: Introductionmentioning
confidence: 99%
“…There are many evaluation metrics available [6][7][8] , making it di cult to determine which metrics to use when evaluating a segmentation pipeline on a particular data set. Distinct types of evaluation metrics have their own limitations 9 , making it essential to obtain results from multiple evaluation metrics to better understand cell segmentation quality. While the region of interest (ROI) level evaluation for segmentations can be used to calculate over and under segmentation 10,11 , the pixel level (PL) scores can be used to quantify the foreground and background detection 12 and ne grain differences between segmentations.…”
Section: Introductionmentioning
confidence: 99%