A model that could turn "offline" handwriting (the ink on the page) into "online" (order and timing of the strokes) I think could be really useful for a histori

A model that could turn "offline" handwriting (the ink on the page) into "online" (order and timing of the strokes) I think could be really useful for a historical HTR pipeline... but ultimately, we need end to end. Why is historical HTR so neglected in all multi-task model evaluation benchmarks? There are millions of un-indexed handwritten historical documents which could give us a so much better understanding of our recent past. For that matter, it could give models much better understanding of our recent past.