Efficient monitoring of construction progress is crucial for the timely and smooth completion of construction projects. It enables project progress to be documented, the schedule to be monitored and timely completions to be ensured. However, traditional manual methods are often inefficient and prone to errors, which has increased interest in automated solutions that utilize computer vision. This paper presents an instance segmentation model for detecting formwork elements on construction sites, which has the potential to support automated progress control in cast-in-place concrete formwork construction methods. Using a dataset consisting of both images collected on a construction site and publicly available images, over 6,600 formwork instances were annotated. This dataset is used to train, evaluate and compare different YOLOv8 and YOLOv9 models using cross-validation. These models achieved precision and recall rates of over 80% and 65% respectively and segmented individual formwork elements using masks, distinguishing between formwork for wall and internal corners. The best model, which showed consistent performance in cross-validation, achieved a mean average precision (mAP) of 76.2% and the current application opportunity is explained using real construction site images.