Cracks are the most important source of information about the damage that occurs to unreinforced masonry piers under seismic actions. To predict the structural state of unreinforced masonry piers after an earthquake, research models have been developed to quantify important features of crack patterns. One of the most used crack features is the width, but this can be influenced by several parameters such as the axial load ratio, the shear span ratio, and the loading protocol, which have not been fully studied in previous research studies. In this study, we use experimental data to investigate the evolution of cracking in stone masonry piers during the application of cyclic shear–compression loading. The data consists of gray-scale images taken during quasi-static shear–compression tests performed on six plastered rubble-stone masonry walls subjected to constant axial force and cycles of increasing drift demand. Through the combined use of digital image correlation and a pre-trained deep learning model, crack pixels are identified, post-processed, and quantified based on their width. The dependency of the crack width on the axial load ratio, the shear span ratio, and the loading protocol at the peak force and ultimate drift limit states of the piers is clarified by a displacement vector field analysis, histogram of the crack width, and the concentration of deformation in the cracks. We show that, as opposed to flexural cracks, diagonal shear cracks do not fully close when moving from the applied drift demand to the residual drift measured upon removal of the lateral load. Furthermore, we provide the maximum residual crack width at peak force and ultimate drift limit states. This study will improve the decision making abilities of future models used to quantify earthquake-induced damage to stone masonry buildings.