Concrete damage detection, traditionally a manual and labor-intensive process, poses challenges regarding accuracy, efficiency, and safety. The emergence of deep learning technologies offers a promising alternative. These algorithms can automate the detection process, increase accuracy, and reduce inspection time and resources. Compared to deep learning concepts of classification and object detection, the semantic segmentation concept can identify the presence of concrete damage and also pinpoint the detailed location. Thus, the focus of this work is to compare and evaluate the performance of several various leading deep learning segmentation models, such as U-Net, DeepLab, and K-Net, etc., in the context of concrete damage detection. The study involves training these models on a comprehensive dataset comprising images of concrete cracks. A comparative analysis including the effectiveness and performance of these algorithms applied to concrete damage detection has been provided, thereby contributing to the advancement of automated and reliable inspection techniques. Our findings provide valuable insights into the suitability of each algorithm for different inspection scenarios, considering factors like the extent of damage, the urgency of inspection, and resource availability. Finally, this work not only highlights the potential of deep learning segmentation algorithms in revolutionizing concrete damage detection but also offers a critical evaluation of their comparative strengths and limitations.