I'll zap you 10K sats if you can combine these photos into a quality scannable QR code
Discussion
Está difícil pero voy a intentarlo
The only important thing really is the round exterior and the lightning bolt in the middle lol
i made one with maximum error correction but the reality is with these codes they are not designed to be mangled pictures those little marker squares are alignment points and the bits are redundantly encoded with reed solomon encoding and the particular way they used to do this precludes wiping out more than the central 1/9th of the image
it would be possible to devise other ways to make it look prettier, for example you can blur and then round the edges of the shapes so it's a bit more pretty but that is the limitations of the encoding
definitely you can't make the edges round, and you can't leave out the margin around the square... that is how the algorithm finds the shape it needs to straighten into a square and aligns where all the bits are placed
you can probably do it relatively easily if you massively increase the error correction rate so the overlaid image can mangle much of the data but still scan
you probably have seen the ones with the icon that takes up the centre 1/9th, well, if you increase the error correction rate you can probably have it take up the central HALF of the image
it likely will also need to have 4 or even 9 marker squares sectioned up as well to achieve this
this is with maximum error correction generated with qtqr, i'd say you can probably do this to it
if that scans, and you like it, of course use it!
this works
the first version destroys too many bits i guess
you are onto something here, my friend
idk if you know how to use blur and levels/contrast on it, if you take the QR code, blur it out to about the diameter (using gaussian) of half one of those pixels, then alter the curve of the black to white so you sharpen up the edges and make the squares blend into each other a bit it looks nicer
you can also even overlay a grid through all the pixels, wiping out a few bits either side of everything, except for those marker squares, they must remain intact, you may have seen this dotty style, it also works, and is more pretty than your basic style
the contrast ratio has to be fairly high, and the dark/light parts must remain dark and light, you can't invert the image, it won't decode in many decoders
yes i am a bit of a graphic artist and i also wanted to make pretty codes like this once that's how i learned all these things
Definite pros and cons with the AI stuff. It is such a black box at the moment that a tiny few have figure out.
I was trying to build something where an AI where read a pressure gauge and tell me the pressure and it was all kinds of shitty. After multiple days of trying to convince the AI to do my bidding, I just figured out OpenCV instead to do it. I can even make the AI write the openCV code, but at least it isnt a black box, and I can make small iterative progress.
Ill read this you post again later and see if I can apply some of your knowledge.
very often with control systems you want a PID controller for pressure and regulation, doesn't require AI, just simple statistical analysis
Yeah i have the PID but i needed the pressure as an input variable to the PID
well, the temperature and pressure are more or less the same given an excess of water, a la boyle's law, you can just use a temperature sensor and calibrate off it, the only time the pressure is gonna go down is if you run out of water and if you run out of water the temperature will probably drop quite dramatically
yes, a little laggy but it would work, you can tune the PID to compensate for that lag with the P and I factors
