by liberal japonicus
An openish thread that starts off with me observing that the Japanese national broadcaster NHK played Civil War on July 4th. If you are unaware of the movie, the trailer is below the fold. I’m sure it was just a coincidence. Observations of all flavors welcome!
nous, I can sympathize. Sprained right wrist (the result of lifting and twisting while loading stuff for the (yes! Legal!) fireworks show) is reminding me that being ambidextrous isn’t a complete solution in these cases.
Heal fast!
Best fireworks display by a law enforcement agency.
nous, don’t they have miracle adhesives now instead of pins? If they can glue together an F-35, a metacarpal ought to be easy.
I went to my annual wellness visit with my GP yesterday. No charge, courtesy of Medicare. Mostly we took old things off the computer records. Kaiser’s computers never voluntarily classify anything as “you don’t need to ask the patient about this any more.” The only thing that surprised me was she didn’t ask if I wanted the shingles shots. I wonder if I’ve crossed another age line?
I did a few more miles on the bicycle this morning than I’ve been doing. This was the first time in a few weeks where we didn’t have an air quality alert for ozone and fine particulates. F*cking wildfires. The biggest of the current lot is known human caused, but it will take some months for them to prove the person. They’ll be poor forever when the courts fine them for the fire-fighting costs and destroyed houses.
From April 2025 I spent odd moments for 12 months digitizing an old fiction paperback that I couldn’t find an e-book version of. I took lots of notes about the open-source software I was using. From April 2026 I have digitized the first book in that series using my own replacements for the image preprocessing and OCR post-processing. At least for this application, the amount of manual fixes is greatly reduced, so it goes much faster.
An F-35 doesn’t self-heal. gotta let those edges knit.
I seem to recall reading about some new adhesives that act as scaffolding for the new bone growth, then disappear. I’ll have to add that to my list of things to read about…
I was thinking about how AI is such a hot topic that many people think of plain old software as AI. Your document-dewarping software came to mind, Michael Cain. It seems like something a lot of people would assume is AI.
Is my thermostat AI? I mean, I just have to set it for a certain temperature and it will “know” what the temperature in my house is and “know” when to turn the AC on. Seems pretty intelligent!
i had someone arguing with me that finite state machines are AI. not that they are used in AI, but that they are types of AI.
this would mean the circuits that control traffic lights, microwave ovens and even light switches are AI.
Your document-dewarping software came to mind, Michael Cain. It seems like something a lot of people would assume is AI.
That’s no longer a bad assumption. If you read the technical literature from the last few years, almost all the new work is about using machine learning and trained neural networks. Or as the general population prefers to label it, AI. And if you have access to a sufficiently large training set, and the right sorts of hardware, (and perhaps minions to do the grunt work), that’s a reasonable choice. Certainly if you are thinking in terms of a commercial product.
My OCR tool chain is built around the open-source tesseract program. Its roots go back to some code Hewlett-Packard put in the public domain in the 1990s. Some years ago someone donated some neural network code to the project. The NN code is so much better than any of the previous algorithms that the default configuration deployed doesn’t include support for the old algorithms.
Michael Cain – I seem to recall reading about some new adhesives that act as scaffolding for the new bone growth, then disappear.
Im holding out for good old-fashioned lug brazing, with decorative fillets – a randonneuring thumb.
If you prefer the modern SotA, there’s Atherton bikes 3D printed Ti lugged Carbon machines: https://www.athertonbikes.com/technology/build-arange.html
I googled “ocr software.” The first hit featured the following text below the link:
Replace brittle OCR with AI powered document processing built for enterprise teams.
It’s a brave new world.
Replace brittle OCR with AI powered document processing built for enterprise teams… It’s a brave new world.
It is. Some of them are very, very good — within their limits. I spent an afternoon submitting test images to several of them. Most limit themselves to straight single- and double-quotes, don’t recognize em- versus en-dashes; for e-book replacements for paperback fiction, which is what I’m doing with OCR, I consider those must-haves. (Yes, there are open-source programs that are pretty good at converting quotes.) Many suppress leading white space on a line, making it hard(er) to recognize paragraph breaks. Few of them provide any bounding box information, or accuracy estimates.
And of course, I’m an old. I had a tech career. I was burned more than once by vendor-provided software that was suddenly broken, or no longer available. I can build and run tesseract on my local hardware. I can tuck away a copy of the trained model for English. I can build all the software I use for my pre- and post-processing around tesseract locally.