A dataset with $23\,870$ digital trajectories (i.e. time series) of handwritten lower- and uppercase Latin letters and Arabic numbers ($a$-$z$, $A$-$Z$, $0$-$9$), generated by $77$ experts using a Wacom Pen Tablet. An expert is considered a proficient user of the recorded symbols, in this case adult native German speakers.
DigiLetTs was created to extend the Omniglot dataset and contains five variants per character per subject to allow the quantification of intra-subject variability and to assess and account for individual writing styles. The determination and imitation of subject-dependent writing styles is introduced as a new task in this paper.
For more information about the dataset, please refer to the repository (Homepage button below).
Paper | Code | Results | Date | Stars |
---|