Real-time HDR upto 18 stops from Canon DSLRs

Started by Andy600, October 01, 2013, 11:45:58 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Andy600

WTF?

This guy claims up to 18 stops of realtime HDR / 32bpc :o

http://www.newsshooter.com/2013/09/15/ibc-2013-gohdr-hdr-video-from-off-the-shelf-dslrs/

How is he pulling 720MB/s (43GB a min) from 2x 5D MkIII's (at least, I think they are 5D3's??)

Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

ItsMeLenny

He's not. It would be 43GB a minute if it wasn't compressed.

He's streaming HDMI from 2 cameras, one would be over exposed and the other underexposed, and then through some math the cameras are positioned so they both shoot the exact same picture (most probably after cropping).

A better way would be a lens that has 2 mounts on it :P so you hang two cameras off one lens, and it splits the image, it's like the opposite of 3D, 1D :P not that that makes any sense, although dynamic range could be considered one dimension.

Andy600

Yes, I get how the HDR is achieved but he states that it's recorded uncompressed at 720MB/s then compressed upto 500:1 and output in realtime. How can he get a 32bit uncompressed signal even from combing 2x HDMI streams? Sounds like nonsense to me  ???
Colorist working with Davinci Resolve, Baselight, Nuke, After Effects & Premier Pro. Occasional Sunday afternoon DOP. Developer of Cinelog-C Colorspace Management and LUTs - www.cinelogdcp.com

dmilligan

It's not nonsense. When he does the math to combine the two streams it's done in 32bit (probably floating point), so he ends up with a 32bit stream that is that large. It's not that there is actually that much information there. For example if I take two ~20MB CR2s from my camera and combine them in photoshop's HDR tool, I end up with a 32bit floating point image that is > 100MB. You can't just add 20MB + 20MB and expect that the result would be 40MB. You use a much larger intermediate bit depth when doing your maths so that you can avoid rounding error and clipping. Then you reduce it back down when you have completed your toning and curve adjuments in 32bit land.

painya

Quote from: ItsMeLenny on October 01, 2013, 11:58:04 AM
He's not. It would be 43GB a minute if it wasn't compressed.

He's streaming HDMI from 2 cameras, one would be over exposed and the other underexposed, and then through some math the cameras are positioned so they both shoot the exact same picture (most probably after cropping).

A better way would be a lens that has 2 mounts on it :P so you hang two cameras off one lens, and it splits the image, it's like the opposite of 3D, 1D :P not that that makes any sense, although dynamic range could be considered one dimension.
That is what how many 3d films are shot, except it is basically a intricate placement of mirrors to one lens. http://www.filmmagic.com.hk/images/redcamera/red3dcam01s.jpg
Good footage doesn't make a story any better.

ItsMeLenny

Quote from: painya on October 03, 2013, 04:52:43 AM
That is what how many 3d films are shot, except it is basically a intricate placement of mirrors to one lens. http://www.filmmagic.com.hk/images/redcamera/red3dcam01s.jpg

Not entirely, but yes. That example you showed is the rig James Cameron literally dreamt, and it allows, when shooting 3D, the two lenses to essentially overlap, which Isn't possible with a human eye, but it needs to be done when using a very very narrow lens. But yes if the two cameras were to overlap completely. But it doesnt look as entertaining as what it would look like if it was just one lens :P

painya

Good footage doesn't make a story any better.

ItsMeLenny

Quote from: painya on October 03, 2013, 08:19:44 AM
I agree! one lens would look spectacular ;D

In continuing though. I'm sure one of the red cameras had 2 sensors in it, one that was really underexposed in case one accidentally overexposed the other so that highlights could be rescued/recovered.

I'm waiting for the Axiom :P

painya

That double sensor Idea I think could be the future of Video. Slap a patent on that  ;D
Good footage doesn't make a story any better.

bnvm

The RED camera already has that. It essentially has 2 sensors that have different shutter speeds rather than ISO's. The problem with this technique is the motion blur is different per exposure due to the different shutter speeds. It is considered a creative tool by RED and is not recommended due to the motion blur issue.

painya

Wow, RED and all of their engineering. HDR video, and This! It's incredible.
Good footage doesn't make a story any better.

ItsMeLenny

Quote from: bnvm on October 04, 2013, 08:08:18 PM
The RED camera already has that. It essentially has 2 sensors that have different shutter speeds rather than ISO's. The problem with this technique is the motion blur is different per exposure due to the different shutter speeds. It is considered a creative tool by RED and is not recommended due to the motion blur issue.

Yeah, I recall something about the motion thing. It was in case one had really f***ed up and over exposed things, then highlights could be recovered.