2021
DOI: 10.48550/arxiv.2109.00456
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Weakly-Supervised Surface Crack Segmentation by Generating Pseudo-Labels using Localization with a Classifier and Thresholding

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1

Citation Types

0
3
0

Year Published

2022
2022
2022
2022

Publication Types

Select...
1
1

Relationship

1
1

Authors

Journals

citations
Cited by 2 publications
(3 citation statements)
references
References 0 publications
0
3
0
Order By: Relevance
“…This then yields a segmentation map as darker pixels are segmented as crack pixels. A similar approach is shown in [55], where a classifier is used to create a detection map as shown in Figure 4, which is then merged with a thresholding approach. Merging the detection map with the thresholding in this approach aims Fig.…”
Section: B Semi and Weakly Supervised Learningmentioning
confidence: 99%
See 1 more Smart Citation
“…This then yields a segmentation map as darker pixels are segmented as crack pixels. A similar approach is shown in [55], where a classifier is used to create a detection map as shown in Figure 4, which is then merged with a thresholding approach. Merging the detection map with the thresholding in this approach aims Fig.…”
Section: B Semi and Weakly Supervised Learningmentioning
confidence: 99%
“…Approaches that train on classification labels and then use thresholding to create a segmentation map [55], [82] S Training on coarse segmentation labels [83] Q Rotation analysis of cracks by rotating the input [39] Unsupervised S Transformation of an input image into latent space or frequency domain before reversing transforming it back into an image. The differences between the input and output then segment areas belonging to cracks or other anomalies.…”
mentioning
confidence: 99%
“…Because YOLOv5 [10] is known to provide a high accuracy with a low time complexity in the object detection research field, we use it as a baseline architecture of AugMoCrack. To address the issue of a lack of crack data in practical applications, we devised a crack data augmentation method based on Poisson blending [11] and a classifier of crack and non-crack patches [12]. We also enforce frequency augmentation in the spectral domain of an AugMoCrack network.…”
mentioning
confidence: 99%