Josh Hadro

These images the MTA put up at many system stations a decade ago were all sourced from the huge NYPL public domain release I worked on in 2015 and 2016. The posters have seen better days but every single one is lasting evidence of the impact of a project of which I’m still quite proud.

My new blog post at @librarycongress.bsky.social explores converting our 20 million record catalog into vector embeddings and what applications that enables. blogs.loc.gov/thesignal/20...

Fascinating and beautiful indeed! It looks like there's some overlap for the bigger structures. The city changes slowly! bsky.app/profile/floo...

Ha, this reminds me in a lovely way of the beautiful viz posters @mgiraldo.mstdn.social.ap.brid.gy used to make back in the NYPL Labs days in the 2010s (these are the outlines of the largest buildings in NYC from 1854)

Many people are asking

This is fabulous local history about, IMO, the best train in Brooklyn

5400 articles published between 1835 and 1920 from over British 20 newspaper titles with metadata, crowdsourced labels (annotations) and fresh OCR doi.org/10.23636/xrk... Described at livingwithmachines.ac.uk/the-how-did-...

Shout out to a guy named Julio in Bushwick who helped my wife jumpstart a dead battery just now. Grateful for all the kind strangers like Julio in NYC!

I've been thinking about this guide for months, and I'm so happy it's now published. NYC parents: you probably already know that school admission season started this week! myself and the AMAZING @chalkbeat.org team create this guide for you! [🧵]

Today in OCR/HTR fun: I'm working with entity extraction over an HTR corpus, and the entity extraction differentiated between versions of the term with and without a superscript "t" -- so "War Dept" and "War Dep^t" are different entities. Attached image is the latter, in different hands.

I broke this story in 2011, about the first major publisher-imposed loan restrictions on ebooks (the only truly big story of my trade journalism career). It’s been 15 years, and the only major difference is that public library budgets are lower and what publishers charge libraries has gone up.

Uploaded a dataset of 98,877 historical newspaper pages (1700s–1940s) to the Hub, each with its original OCR text, word boxes and confidence scores. huggingface.co/datasets/big...

You: "OCR is a solved problem" Me:

You can make flipbook GIFs 🤩 Play with it here: hadro.github.io/flipbook/#m=...

Happy to report that the 5yo has no problem answering the question, “what is it like to be a bat?” Checkmate, philosophy 101, next problem please 🦇

If you believe that libraries should be allowed to buy books and lend them out, then you also believe in the First Sale doctrine, for better or for worse

Fun fact, I am working on a fun work project that has an output sorta like this. Don’t worry, it’s a dataviz, not a national library wet lab

Counterpart not being on the top 100 shows list is an outrage. All 12 of us Counterpart fans are outraged.

#JusticeForBluey

Library friends (📚📜): I'm seeking examples of digitized books where the "cover image"/first image of the work poorly represents the contents — e.g. maybe there's boring cloth binding that shows as the first image, hiding something like Audubon illustrations just a bit deeper inside. Any examples?

😻🗺️📜📚

The Europeana Newspapers dataset on @hf.co now has an `alto` config: the raw ALTO XML for all 5.9M pages. The coordinates for every word, line and block, per-word OCR confidence and font info that the flattened text dropped are back. huggingface.co/datasets/biglam/europeana_newspapers

What’s your location-specific delusion? I know I could not win a fight against a chimpanzee, and I doubt I could land a plane, but based on no evidence or experience whatsoever I still feel positive I could pilot an East River ferry safely to port if circumstances called.

A true galaxy brain marketing maneuver: Brooklyn Public Library adding a free ebook PSA onto the end of the Jury Duty orientation video Well played @bklynlibrary.bsky.social! Love the suggestion to visit the nearby Center for Brooklyn History as well.

We've made a lot of progress on OCR + HTR, but nothing breaks a layout parser like Cattle Text Recognition (CTR)

If I had a time machine, after I finished all the moral imperative work etc etc., I would go back to 1930s-1950s and figure out how to be a fly on the wall for all the conversations about microfilming library collections to try and grok how it all went down

😻 bestill my local news loving heart, this is a hyperlocal dream!

Spent the last few days at #ff2026 talking about neat advancements in OCR technology, among other things. Here’s a pic of 2016 me showing the time I used GNU parallels to chain together 10 computers in NYPL Labs over LAN to do mass distributed tesseract jobs. I basically created an OCR botnet.

I look like I’m delivering a sermon but really I’m talking about the nerdiest possible uses of things like travel guides and city directories