|
|
Post by Alex Dering on Feb 21, 2014 21:57:04 GMT
... and Borges library.
A 25 minute episode consists of a finite (very long, but finite) series of 1s and 0s. I'll simplify it as 1010101010... [skip many digits] 1010101010. However, although there's only one perfect recreation, there are an enormous number of single-bit-off reconstructions (that is, instead of 1010101010 ..., the first 10 digits are 1110101010. The single bit of information is wrong, but the overall effect is so trivial, it doesn't wreck the reconstruction). With the extant material (stills, video fragments, sound tracks, etc.), the universe of possibles shrinks considerably. As it's black and white, that also helps. As the soundtrack already survives, that cuts down too.
So, in theory, there are probably a staggeringly huge number of reconstructions that would be 95% of the way there.
How complicated a project would it be to reconstruct via computer algorithm like they would on a Star Trek holodeck? Is it just completely insanely impossible, or is it a "you know, it 10 years, they could probably do a pretty passable fake."
|
|
|
|
Post by garysrothwellx on Feb 21, 2014 22:19:41 GMT
Interesting. Put it like that, all that is required is the computing power - which may not today be available, but which may in the future. lets face it, if 30 years ago someone suggested you could have 30,000 songs on something the size of a deck of cards you'd have been laughed at.... so you never know!
|
|
|
|
Post by Brad Phipps on Feb 21, 2014 22:24:33 GMT
Impossible! Calculations like that would take hundreds of years...
|
|
|
|
Post by garysrothwellx on Feb 21, 2014 22:32:24 GMT
Impossible! Calculations like that would take hundreds of years... With today's technology... who knows what will be possible!! having said that, probably not in our lifetime, and after that who will care, but the exponential rate of computer power will certainly continue. |
|
|
|
Post by Brad Phipps on Feb 21, 2014 22:41:36 GMT
Impossible! Calculations like that would take hundreds of years... With today's technology... who knows what will be possible!! having said that, probably not in our lifetime, and after that who will care, but the exponential rate of computer power will certainly continue. ...I was quoting a certain episode... |
|
|
|
Post by Marty Schultz on Feb 21, 2014 22:44:40 GMT
How would you check? It would be easier to use computer technology to reconstruct the episode than to use an undeveloped quantum processor in order to randomly generate the correct sequence. Monkeys typing Shakespeare.
|
|
|
|
Post by Marty Schultz on Feb 21, 2014 22:47:02 GMT
In short insanely impossible.
|
|
|
|
Post by garysrothwellx on Feb 21, 2014 23:03:30 GMT
With today's technology... who knows what will be possible!! having said that, probably not in our lifetime, and after that who will care, but the exponential rate of computer power will certainly continue. ...I was quoting a certain episode... Oh jeez... i am embarressed that I didnt get that, given i have watched and read it dozens of times..... i need to focus!! |
|
|
|
Post by simonashby on Feb 21, 2014 23:11:17 GMT
There isn't enough extant visual material to be that accurate without a lot of manual input, which renders the whole concept useless. The idea might be able to produce convincing results when (not if) computers can be programmed to be smart enough, but when compared to the real thing (if it ever turns up) will almost undoubtedly be different. But I do believe that CGI will become realistic enough, quick enough and cheap enough to reconstruct episodes that could pass for the real thing. And I do believe at one point some shots may be convincingly recreated automatically. When that will be is a different question. The same ideas could be applied much sooner to colourisation and resolution upscaling. Full colour HD Power of the Daleks here we come! Estimated launch date c. 2066! iTunes only... naturally...  |
|
|
|
Post by Paul McDermott on Feb 21, 2014 23:53:25 GMT
All we need to really do is work out just how improbable this reconstruction is, feed it into that machine over there, and give a cup of really hot tea... Even if some day such a thing is possible - and there's many a slip, ask the dinosaurs - there's no accounting for taste. The Trek connection made me think of Picard's disdain for replicated caviar, being "not quite right". Won't be surprised if for some, this will remain true. Personally, I'd be as interested in this as I would any other first rate recon - provided there's no way to see the original, of course!  And who knows, maybe some day Big Finish will have the technology to make their audios into actual videos - using licensed performance recreations of the cast no longer with us.  Will they be able to make stories in keeping with their respective eras, or will they be too removed to get the tone right? I reckon they'd walk both sides of the street, as now, for wider appeal. But until it's in the shop, even if at the interesting prices BBC Video originally charged for their initial - and edited - offerings, I think we'll have to continue to be happy with what we've got. Which admittedly, is quite a lot, compared to many other lost film and TV enthusiasts, who'd surely pull rank on us for access to that magic ep remaker machine!  And there's always hope for new additions to the ranks of the returned, for all fans of missing material. So, fingers crossed!  |
|
Simon Collis
Member
I have started to dream of lost things
Posts: 536
|
Post by Simon Collis on Feb 22, 2014 0:40:21 GMT
I don't really think you would get anything automatically able to do it. I think you'll probably end up with something doing it manually - maybe more like the recent Who vs Sherlock on YouTube. (it's here and the how-it-was-done is here) |
|
|
|
Post by martinjwills on Feb 22, 2014 12:23:43 GMT
... and Borges library. A 25 minute episode consists of a finite (very long, but finite) series of 1s and 0s. I'll simplify it as 1010101010... [skip many digits] 1010101010. I think you will find it is stored in Hex rather than binary, the Mpeg2 Streams ive looked at are in Hex. so its a string of double bytes from 00 to FF in groups. |
|
Simon Collis
Member
I have started to dream of lost things
Posts: 536
|
Post by Simon Collis on Feb 22, 2014 13:13:03 GMT
Martin, without getting too technical, it really isn't. At the lowest level, all modern processors use on/off switches at the bottom - transistors, that correspond to ones and zeroes.
Hex is used by programmers (like me) to group 8-bit bytes into two digits to make life easier for we humans. But it's really bits all the way down. If MPEG2 streams work anything like the compression streams I've looked at in the past, it's likely that each part of the stream is variable length, meaning you can't then tell where each part of the next block begins.
The simplest encoding scheme - and what I think Alex was really driving at conceptually - would be one where the image was stored simply as greyscale levels - let's say each byte (00 to ff, or 00000000 to 11111111) corresponded to the intensity of a single pixel.
If we assume a greyscale video, packed out with all 377 visible lines, each line at the modern 720 pixels wide, that gives you a total of 271,440 bytes (combinations of 8 bits) per frame. Of which there are 25 frames a second, for 25 minutes (each of 60 seconds), and that gives us 8,143,200,000 combinations in total.
Assuming that we can have errors in the least significant bit (i.e. the least possible noticable error) and that each one gives a separate unique combination, hits the point where my experience as an assembly language programmer gives way to my lack of ability in higher mathematics...
|
|
|
|
Post by felixdembinski on Feb 23, 2014 18:31:07 GMT
Martin, without getting too technical, it really isn't. At the lowest level, all modern processors use on/off switches at the bottom - transistors, that correspond to ones and zeroes. Hex is used by programmers (like me) to group 8-bit bytes into two digits to make life easier for we humans. But it's really bits all the way down. If MPEG2 streams work anything like the compression streams I've looked at in the past, it's likely that each part of the stream is variable length, meaning you can't then tell where each part of the next block begins. The simplest encoding scheme - and what I think Alex was really driving at conceptually - would be one where the image was stored simply as greyscale levels - let's say each byte (00 to ff, or 00000000 to 11111111) corresponded to the intensity of a single pixel. If we assume a greyscale video, packed out with all 377 visible lines, each line at the modern 720 pixels wide, that gives you a total of 271,440 bytes (combinations of 8 bits) per frame. Of which there are 25 frames a second, for 25 minutes (each of 60 seconds), and that gives us 8,143,200,000 combinations in total. Assuming that we can have errors in the least significant bit (i.e. the least possible noticable error) and that each one gives a separate unique combination, hits the point where my experience as an assembly language programmer gives way to my lack of ability in higher mathematics... If you had it in the original 4:3 ratio it'd be 377 by 503, (377 X 4/3) so it'd be 189,631 bytes per frame. So it'd be 189,631 X 25 X 60 X 25 for a full episode's worth of bytes, which would be about 7.1 billion bytes. Divide that by 97 if your just concentrating on dr. who, (although there are obviously far more lost programmes, but most don't have complete soundtracks) and you'd get about 73.3 million. Given that the chances of winning the typical 6 numbers from 49 numbers lottery are about 1 in 14 million, a 1 in 73.3 million chance for randomly generating a complete lost dr who episode seem to be very low to me, am I missing something? |
|
Simon Collis
Member
I have started to dream of lost things
Posts: 536
|
Post by Simon Collis on Feb 23, 2014 18:47:02 GMT
That's an interesting point, but I'm not sure that it's statistically the whole answer. Is a negative of one version an entirely different permutation, or just the same thing with all the bits flipped? Are two that are identical except for one frame different permutations? And how would you sort through the millions that were similar?
What if we took the existing telesnaps, digitised them, and then filtered the results so that they had to contain those telesnaps, exactly as we digitised them, at roughly the same point we'd expect to see them in the episode, in order?
Generating this with a random algorithm would be monstrously difficult and time consuming, it seems to me. But I have begun to wonder if a sufficiently large neural network, given sight of the existing episodes, and then given the existing telesnaps, might not be able to do some weird kind of "reconstruction". Sure, it would be a weird project to do and require far more computing power than I suspect any of us here have access to, but a fairly efficient C (or even FORTRAN) program, nicely optimised, might have a good shot at it. (I only mention FORTRAN incidentally because the FORTRAN gurus I've met are very proud of the ridiculous efficiency of their compilers compared to more modern languages).
|
|
