Description
Reassembly of 3D fragmented objects from a collection of hundreds of randomly mixed fragments is a problem that arises in several applied disciplines, such as archaeology, failure analysis, paleontology, etc. In this talk we will walk through the pipeline of 3D data generation in archaeological studies, from pre-processing of images, moving from 2D to 3D space, and finally the training and evaluation of generative adversarial networks in Python for 3D meshes corresponding to Iberian vessels. We will report several python libraries (scikit-image, pytorch, visdom, etc.) and how they are used in this particular pipeline. The main goal of augmenting our dataset in 3D is to perform fragment part identification and vessel reconstruction.