Adaptive & Fine-Grained Domain Adaptation for Pavement Crack Segmentation using YOLOv8 Learning Framework

In this study, the utilization of Roboflow's YOLOv8 architecture for crack segmentation is explored, focusing on evaluating the model's performance and readiness for real- world deployment. The methodology involves key steps such as data collection, pre processing, model training, evaluation, and deployment. Initially, a diverse dataset of images containing various types and severity of cracks is curated and annotated. The dataset undergoes pre processing to enhance image quality and standardize input dimensions for optimal training performance. Subsequently, the YOLOv8 architecture is employed for model training due to its efficiency in realtime object detection tasks and effectiveness in handling small objects like cracks. To evaluate the trained model, metrics such as precision, recall, F1score, and mean Intersection over Union (mIoU) are computed on a separate validation dataset to assess the model's ability to accurately segment cracks while minimizing false positives and false negatives.