Hieroglyphs AI: automatic Gardiner code recognition for Egyptian hieroglyphs

Hieroglyphs AI is a web application that identifies the Gardiner code of an Egyptian hieroglyph in an image. The user uploads an image, selects a sign, and receives the five most likely Gardiner codes with their probabilities. The tool is meant for Egyptologists and students who need a quick first identification of a sign without browsing a sign list by hand. The recogniser is a deep neural network trained on about 145,000 hieroglyphs in 709 Gardiner classes, labelled by hand by the author and checked against publications where available. The images come from stelae, temple and tomb walls and printed sign lists, from the Old Kingdom to the Ptolemaic period. The model was tested on 21 monuments (7,903 signs), mostly stelae, that were kept out of training completely. The correct code is ranked first in 89.3% of cases and is among the top five in 96.8% (measured with a model trained the same way without these monuments). Details, limitations and citation: see the attached PDF.