Semantic segmentation is the task of assigning a class to every pixel of an image, widely used to automatically locate objects in the context of computer vision applications, such as autonomous vehicles, robotics, agriculture, gaming, and medical imaging. Deep Neural Network models like the Convolutional Neural Networks (CNN) are suitable to this extent. Among the plethora of models, the UNet model is widely adopted in bio-medical imaging. The segmentation using CNNs is efficiently performed using GPU accelerators. FPGA devices are also emerging as novel technologies, especially for performing inferences, promising higher energy efficiency and lower latency solutions. In this contribution, we assess the use of FPGA-based accelerators for the inference task using the UNet model. The calcium segmentation in the cardiac aortic valve computer tomography scans is devised as a benchmark application. In particular, we show how to port and deploy a CNN model on such devices and compare the accuracy, throughput, and energy efficiency benchmarking with recent CPUs and GPUs.