The !!Con talk from this weekend about the project: <a href="https://youtube.com/watch?v=KqIOR-hofm4&t=23271s" rel="nofollow">https://youtube.com/watch?v=KqIOR-hofm4&t=23271s</a>
It appears to succeed my litmus test: <a href="https://brooklyn.alltexts.nyc/?search=moshiach&page=1" rel="nofollow">https://brooklyn.alltexts.nyc/?search=moshiach&page=1</a><p>For those who don't know, Chabad, an orthodox Jewish organization, has a large promotional presence in Brooklyn
This is really cool, and hints at a near-future possibility of building a search engine on top of just about anything. It's clear we've moved past the ability to just search for website url's and webpage content. Anything that can be indexed - regardless of type of data or dimension (space, time, etc.) will be searchable.
There is much that isn't indexed here. "718" yields less than 4,000 hits, which seems very low. There are many old buildings in Brooklyn that still sport "fallout shelter" signs, but ATiB returns only 13 results for that term, with many duplicate images. Here's an example of a missing result, probably one of thousands: <a href="https://maps.app.goo.gl/AdsYFzf7MAZoo4Vc8" rel="nofollow">https://maps.app.goo.gl/AdsYFzf7MAZoo4Vc8</a>
Fun app, I'm having a good time!<p>The OCR has a lot of false positives, though. "Truck" definitely was not what I was looking for, but it makes up a significant amount of the search. "culo" fuzzy/exact results were also surprisingly disappointing :) .<p>If there's a way to change the text-matching accuracy and add this filter to the front-end, I'd be lost here forever. Switching locations would also be a fun way to scale this up. Throw an Adsense add on there and you're looking at a decent passive income!
I was a bit sad that "kest gak" yielded nothing, but then tried individual "kest" and "gak" and found the familiar tags. Fun site
This is such a neat site, thanks for sharing.<p>I wanted to use a word that I figured would be rarely seen in Brooklyn, so I tried: “Gripe”<p>The correct identifications center around “Vacuna de la gripe” (flu vaccine).<p>The remainder are all mismatches, such as “Grape” at a very sharp angle. Funnily, the majority of the mislabeled samples are all due to the “Good Grips” brand logo. This logo has a small underlined “s” at the end that looks like a “E” when you squint at the JPEG. I’ll give the OCR model a pass on this one!
I've been either stunned, or disappointed depending on the word.<p>"hello" gives four images of the same building with "hello" clearly written, as well as a few images of "hello" grafiti. Impressed<p>"table" gives six results- four of which are clearly pictures of either leaves or the sky. Two are blurry buildings, but I cant seem to find the text "table"... it could be there though? Not impressed<p>"car" gives Six unique results, some of which "car" is the prefix of a word. Impressed<p>Either way, really cool project.
<a href="https://i.imgur.com/xvHD8fK.jpeg" rel="nofollow">https://i.imgur.com/xvHD8fK.jpeg</a><p>This would be really useful in GeoLocation GeoGuessing games - and in order to ID a location based on any limited text you can discern. Wonder how hard it would be to apply it to other locations.