We've started rolling out a new beta version 3.2.98 on our beta channel on Play; we're worried enough that it might randomly crash on people that we haven't even distributed it to all testers yet, but will over the course of the next few days.
The big change here is that we're finally doing our long-awaited roll-out of new algorithms for OCR and handwriting; "New OCR" is now the default OCR interface, as is a new modern machine-learning-based OCR algorithm. A downside of newer OCR algorithms is that while they're generally more accurate, and better at pulling text out of weird places and distinguishing it from other random stuff, they aren't great at matching up detected characters to precise pixel locations in an image, so with the new algorithm, live OCR shows the detected characters above the box instead of overlaying it, and still/screen OCR covers characters with an opaque white rectangle rather than a transparent green overlay.
We also added a new 'snap OCR' screen, which is basically just 'pause video and then tap on characters to look them up.' But honestly, now that the text detection is more accurate we feel like this will be better than 'live OCR' for most of the applications for which people were using live OCR, and might even make 'live OCR' a hidden / off-by-default thing at some point depending on how people respond to 'snap OCR.'
Experimental handwriting has been available in various builds before, the only difference is that now it's the default.
You can go back to the old OCR interface and algorithm and the old handwriting algorithm via Settings / OCR and Settings / Input Methods / Handwriting. We're planning to phase out the old algorithms for new users over the next few months, but to continue allowing people who already paid for OCR or 'enhanced handwriting' (or a bundle, as all of them include enhanced handwriting) the option to use the old algorithms if they prefer them.
We also updated our PDF decoder to use the open-source PDFium library - which saves money / disk space and gives us the ability to make whatever arbitrary changes we like for it - and cleaned up PDF viewing a bit (scroll positioning is nicer and it lets you use volume buttons to scroll just like other readers). And we fixed a bunch of bugs too.
Also, for iOS people following this: the new handwriting algorithm is available in the current iOS build as an off-by-default instead of an on-by-default option in settings. We're making a similar break with our old OCR algorithms in the next 4.0 beta (and we already do offer the option to use a different OCR algorithm for still images) but debating whether to also bring a new algorithm to the old app.
The big change here is that we're finally doing our long-awaited roll-out of new algorithms for OCR and handwriting; "New OCR" is now the default OCR interface, as is a new modern machine-learning-based OCR algorithm. A downside of newer OCR algorithms is that while they're generally more accurate, and better at pulling text out of weird places and distinguishing it from other random stuff, they aren't great at matching up detected characters to precise pixel locations in an image, so with the new algorithm, live OCR shows the detected characters above the box instead of overlaying it, and still/screen OCR covers characters with an opaque white rectangle rather than a transparent green overlay.
We also added a new 'snap OCR' screen, which is basically just 'pause video and then tap on characters to look them up.' But honestly, now that the text detection is more accurate we feel like this will be better than 'live OCR' for most of the applications for which people were using live OCR, and might even make 'live OCR' a hidden / off-by-default thing at some point depending on how people respond to 'snap OCR.'
Experimental handwriting has been available in various builds before, the only difference is that now it's the default.
You can go back to the old OCR interface and algorithm and the old handwriting algorithm via Settings / OCR and Settings / Input Methods / Handwriting. We're planning to phase out the old algorithms for new users over the next few months, but to continue allowing people who already paid for OCR or 'enhanced handwriting' (or a bundle, as all of them include enhanced handwriting) the option to use the old algorithms if they prefer them.
We also updated our PDF decoder to use the open-source PDFium library - which saves money / disk space and gives us the ability to make whatever arbitrary changes we like for it - and cleaned up PDF viewing a bit (scroll positioning is nicer and it lets you use volume buttons to scroll just like other readers). And we fixed a bunch of bugs too.
Also, for iOS people following this: the new handwriting algorithm is available in the current iOS build as an off-by-default instead of an on-by-default option in settings. We're making a similar break with our old OCR algorithms in the next 4.0 beta (and we already do offer the option to use a different OCR algorithm for still images) but debating whether to also bring a new algorithm to the old app.
