Crack Semantic Segmentation With Mask2Former

The objective of crack detection is to identify any defects present on the surfaces of various physical structures. This task can be approached in two ways: bounding box detection and semantic segmentation. In this study, we focus on a method based on semantic segmentation that can provide per-pixel classification results. We applied Mask2Former, a method known for its state-of-the-art performance in semantic segmentation, for crack detection. We conducted experiments using various crack datasets, and the results highlight the need for enhanced performance in non-crack detection to achieve improved results.