Downstream Semantic Segmentation Model for Low-Level Surface Crack Detection
Table 3
Comparing our downstream models built upon DeepLabV3-ResNet101 with other works of deep-learning-based low-level crack detection on the same test datasets; the flexibility value of = 2 is used in all evaluations. While ODS F-measure is reported with its corresponding precision (P) and recall (R) as written underneath, OIS F-measure is reported together with its standard deviation (written with a leading ± symbol) showing the dispersion of the optimal image-scale F-measure in a particular test dataset. For each training strategy of us, the best (maximum) value among the four loss functions is highlighted in bold.