A new #Hieroglyphic release is out now. Thanks to the exciting work done by @bnyro,
Hieroglyphic can now also recognize @typst symbols (a modern alternative to
LaTeX). Hardware-acceleration will now be preferred over using the CPU, when
available, reducing power-consumption.
Head over to @flathub to download the latest version: https://flathub.org/en/apps/io.github.finefindus.Hieroglyphic
@FineFindus "Submit recognized drawings to Hieroglyphic's dataset".
I was wondering how this works. How does it know when it recognized something correctly? Does it do this if I actually copy a character?
And what's the risk of getting junk-data? Like will my awful writing make things worse?
Not criticizing, I'm just curious. It's a cool idea to collect this information for training on opt-in. :)
@sstendahl It's generally submitted when copying (there are some additional constraints, e.g. not on metered networks).
I generally trust people not to intentionally submit faulty data. Bad handwriting isn't an issue, since:
1) the data is cleaned up and downscaled before being saved/used for training
2) it's just training data with slightly more variance (which is infect helpful).
Happy to hear feedback about this, it's definitely something that could be improved.
FineFindus
Sjoerd Stendahl