@AWhitTwit I doubt it would make much difference, since a PDF of a non-OCR'd scan is just one or more image files with a PDF wrapper around them. Though if the images are compressed during PDF-ification, that could maybe lead to slightly worse OCR results vs just using the original image.
@HimarshaJ The blogs went away during a very recent redo of the library site. I wasn't involved, but prior to the cutover, there was a concerted effort to preserve the old site, including the blogs, on our local web archiving portal as well:
https://t.co/GzNSMWNfOW
@migueljogja I may have seen the paper earlier, but the code is new to me, thanks! I'm glad that pose armature/kinematic data is finally getting the full self-attention treatment. Hopefully getting from this to compelling downstream applications won't be too much of a slog.
@quinnanya @gworthey @tillgrallert @DH2022_Tokyo I'm imagining attendees deciding to blow off the last day of the conference to wander the streets of Tokyo on Google Maps...
"Comparative K-Pop Choreography Analysis through Deep-Learning Pose Estimation across a Large Video Corpus", a paper @tango63 and I wrote, just dropped at @DHQuarterly. Shake a leg on over there: https://t.co/rQRBNSWowx #AVinDH
@TheKpopProf@javiercha We cobbled together some "historical" data from discogs, Namu, https://t.co/zaalnINIIH, Generasia, Wikipedia, and even https://t.co/6qKCt8JuyZ, but found that the more computation-friendly sources all tend to have a strong presentist bias.
@latimes bringing the snarky and entirely irrelevant details. Though when I heard that my former apartment in Irvine is in the mandatory evacuation zone, I'll admit my first thought was to picture a parade of gleaming fancy cars fleeing through a blizzard of ash.
@Elijah_Meeks@jsundram Interesting talks at DH2019 on Big Data Jazz Studies from https://t.co/fNJSrttO3J. Like other efforts (RISM, Josquin Project), main focus seems to be on building and searching DBs. Auto-transcription and OMR are priorities as well, but hard.
@migueljogja@migueljogja Really interesting work! No doubt there would be some complex accuracy vs. cost/hassle tradeoffs to consider between this kind of instrumented, highly accurate data gathering/analysis and using pose detection, but the latter is probably worth trying, at least.