CuneiNormals
CuneiNormals is an image-to-image translation model based on the Pix2Pix architecture. It is trained to generate surface normal maps from input images of cuneiform tablets.
The model takes a conventional tablet image as input and predicts a corresponding normal map, where pixel values encode the estimated orientation of the tablet surface. These generated normal maps can make the three-dimensional structure of wedges and impressed signs more explicit than standard photographic images.
More information is available on the CuneiNormals GitHub repository
This is the FeatureMatched implementation of pix2pix.
Output
The model generates an RGB normal map corresponding to the input image. The RGB channels represent estimated surface-normal directions.
Intended use
This model is primarily intended for experimental and scholarly use. Results should be visually inspected before they are used in further analysis, publication, annotation, or automated recognition workflows.
Project context
This model was created in the context of the CUNE-IIIF-ORM project, an interdisciplinary consortium of historians, museum curators, digital humanities and heritage experts, digitization specialists, and computer scientists from KU Leuven, Ghent University, and the Royal Museums of Art and History.
The project aims to improve access to diverse cultural, scientific, and historical heritage collections held by the Royal Museums of Art and History.
This project is supported by experts at the Ghent Centre for Digital Humanities (GhentCDH), a core facility specializing in consultancy and support for Digital Humanities projects at Ghent University.
This model was trained by @y2gkcsdef
- Downloads last month
- 14