Any advances in hand-writing recognition for scanned/photos of docs?

Homebrew, Python, oh boy


create an issue in the GitHub repo, and ask for install instructions, as I don’t see any.

Ahh I’ve never successfully installed any package for use in PyCharm except from within the PyCharm interface. You also have to be sure you are installing it the version of Python you expect to run. (I have 3 versions of Pythin installed on my development machine.) Normal command line installs put it in some weird system place that PycCharm can’t access. At least that’s what I found out.

Also, I’ve never ever gotten anything installed via homebrew to run at all on any Mac I tried it on. So I never even use that as an option now.

1 Like

I’ve updated the github with INSTALL instructions.

My apologies.


Yeah, I think the PyCharm stuff doesn’t work from the command line. I could be wrong.

Homebrew has been my go-to package manager for years, and haven’t had any trouble. I abandoned MacPorts because it just made a mess of things, require root access, etc.


I had to pip install google-cloud-vision after setting up the environment, then authorize the API, and the test code works!

Thanks @hirenhindocha for all the help!


Okay, didn’t get the whole workflow working, but tried a single image, and it looks like google’s OCR doesn’t like my handwriting.

Thu 17, Dec zozo
6:08 am Early stort this morning, on new paper. This paper is
flat and
nice. It has a little more toota then
the clair lautariae /Rhodia, and is taking ink nicely.
is nice ar well, stiff cover,

So, I gave it a shot.


Is translation might stort to be guite a bit more toota than you want, ar perhaps your handwriting is a bit too tightly spaced vertically.

(The translation might start to be be quite a bit more trouble than you want, as …)


1 Like

I don’t use PyCharm. I write a lot of small standalone scripts. PyCharm had a big project feel to it that I don’t need. That said I believe it creates a Python virtual environment for each project. That is probably the root of your issues with it. You would have to activate the virtual environment from the command line to use your code there and to pip install (inside the virtual environment).

1 Like

How does it compare to DevonThink or Scansnap OCR?

I’m pretty sure neither of those do handwriting.
Although I haven’t actually tried it.

Correct and that is why I think I’ve never been able to get anything working outside the PyCharm envronment unless I installed it there.


If the Google Vision API used is the same for the Drive → Docs —which I assume it is, the quality is much better than DEVONthink’s ABBYY OCR.

See my previous post here in this topic:

Abbyy doesn’t support handwriting recognition.

Edit: realized their answer is somewhat ambiguous, as they don’t discern between hand printed and machine printed.

1 Like

Yep, you made that clear whether the OCR service officially supports hand writing or not. I just want to let them know about my experience on using ABBYY & Google’s for handwritten text.

In other news, now is my turn to be stuck in the 10th Circle of Dante’s hell :rofl:


Joining into this conversation in order to see how things develop in 2021. After years of waffling, I finally decided just to commit to using DevonThink. So far, it’s worked, and that’s been a relief. For Christmas I got a RocketBook, which I have so far treated mostly as a kind of portable whiteboard, keeping task lists, doing some brainstorming. But meetings lie ahead in the new year, and I’m hoping to use it in meetings and then to digitize those handwritten notes into DevonThink. So I’m hoping for both good OCR possibilities and automatically adding documents to DT.

I’ll take a look at the Python automation possibilities. I use MacPorts to set up Python and PIP. For those who have difficulties in getting your Mac to recognize Python, do make sure that you’ve told your shell – I still use Bash – where to the software you’ve installed using export PATH.


I have spent forever (it seems) trying to crack this nut. My preference at this point is to write on paper, then run an action in Drafts that opens up Prizmo Go (their paid handwriting recognition is really good), then dumps the recognized text in Drafts. Then I can run actions from Drafts to particular Devonthink folders or Obsidian (I have a great Zettelaksten template in Drafts, and so a different Drafts action runs the same sequence above, but places the Prizmo text into a formatted Draft page).


The baseline Drafts action can be found here: Insert from Prizmo Go | Drafts Action Directory. And just to give you an idea of the OCR capabilities of Prizmo Go - here is my terrible handwriting with Prizmo, followed by an attempt of using the example that Google failed with (from @ybbond above.


Thanks! That’s impressive!

Yes! And, with the Drafts action, it is super fast to get it to text, then off to Devonthink or wherever. The only friction point is that you often have to toggle the handwriting OCR in Prizmo (it seems to default to the quicker, but less precise on-device recognition rather than the cloud handwriting recognition).