a crazy idea for a libre frendly camera (warning: long winded and bloggy)

Nessuna risposta
Iscritto: 01/10/2013

I've toyed with the idea of using analogue camcorders, a couple pal region Hi8 camcorders modded with RGB out and have them digitally captured via RGB capture cards and have them through a beam splitter at 4 stops apart and have the middle exposure "guessed" by an AI. So one camera would be at +2 and one would be at -2 and 0 would have been guessed by looking at +2 and try to remove blown highlights and even if there were blown highlights at -2, the same AI could guess missing detail from there and from there, I've thought of merging with enfuse and turn those image sequences at different exposure layers into one sequence of .TIFs at a 16-bit colour depth and on top of that, after everything's all graded and edited, export to image sequence and upscale image sequence using AI and if it's horror, it would look really good on the grunge of VHS, so it might be a good idea to downscale back to VHS and it would push the limits of shooting with VHS Camcorders and if you would transfer back to VHS, it would almost as good as a horror film shot on film and then transferred back to VHS, it would just give more granularity and have more control of what the viewer can see, can't see and what they think they see.

But that sounds like a logistical nightmare because well, working analogue camcorders are kinda of a collector's item and those capacitors could go at anytime not to mention you would have to mod it to use more contemporary batteries and RGB capture sounds like a nightmare as well not to mention it would be very heavy and would have gotten very expensive for the novelty. But, the same methodology could probably apply to the C920 in something like the "kurokesu" lens mod and since they're cheap, you could probably get four of them and not bother with AI imagination for exposure tweening (especially if you can't do it with free software) or you could scale that AI dynamic range hackery even more and also since the C920 has stereo mics, there could be a microphone mod soldered in where say if you were using four c920s in unison, you would have 8 synced audio channels to record from. I figure it would at least look better than Betacam as far as dynamic range. But even without hoping for AI to "save the day", there are still things that can be done with the resources that are currently available. Especially if I can get uncompressed 1280x960 from the c920s, that would make a good overhead for SD downscale and a lot of the best shows are only available in SD.

Here are some things that inspired this tangential train of thought.

RGB modding: https://invidio.us/watch?v=DLz6pgvsZ_I

The Kurokesu mod: https://invidio.us/watch?v=7xqwVVYvWEY

Enfuse: https://invidio.us/watch?v=G4Jl3Zzxuwk
(yes, there's a libre backend)

Beam Splitter HDR: https://invidio.us/watch?v=pnQpvun8nzI
Yeah, it looks surreal, but that's the way they graded it. I can imagine you could be more subtle.

AI Dynamic Range reconstruction: (I hope something like this is implemented in free software)

A demo of what current non-free image upscaling can do: (it will probably happen in free software one day)

A half an hour video essay on the magic VHS gave us in case you think I'm crazy talking about that: https://invidio.us/watch?v=xbZMqS-fW-8

People say "the best camera is the one you have with you" but I never plan on having a phone with me and this might work if not from being tethered to a T400, but maybe to a studio using those asus boards. I'm also curious if Kurokesu would expand their niche by offing source code for the camera drivers they use. (or at least good documentation or having it be trivial to a seasoned USB driver engineer)