This paper presents research on the application of 3D technologies to improve the documentation of large scale monuments for conservation as well as for the study of their visual imagery and narratives. The paper concentrates on the case of the Trajan Column cast at the Victoria and Albert (V\&A) museum in London, UK. The paper presents two contributions: i) lessons learned from the digitisation of a section of the Trajan’s Column cast at the V\&A; and ii) a novel method to automate the analysis of the semantic shapes carved on the monument. Given the prominence of human figures throughout the column, we focus on detecting the various faces for all the figures in the column as an important first step. These faces however are reliefs, and so do not follow the same geometry as normal 3D faces. Based on the observation that they look visually similar to normal faces, we take an image-based approach that renders the column, and trains image-based face detectors. We experiment with both deep learning and traditional methods, find that deep learning based methods are substantially better than traditional methods, and training with shading images estimated using intrinsic image decomposition produces the best results, due to the similarity of their characteristics to the rendered column images. The detected faces are finally mapped to the 3D column model. The initial results suggests that the digital 3D models and novel methods based on machine learning can provide useful tools for heritage professionals to deal with the large-scale challenges presented by such large monuments.