Off the chart...

[Filip Orlandic]

Maybe RED improved on an old Fuji trick and put 2 photodiodes per photosite (one for highlights and one for shadows). It's possible

I'll give you some clues on HDRx©...

1. What we are doing has never been done before.

Jim

K.Rexach, I don't think so

 

[JD Holloway]

And good luck to you....

+1...

 

[Enrique Blein]

Just off the top of my head: Maybe it's to do with the 120 fps. Maybe they found a way to interpolate differently exposed frames shot at very high shutter speeds, closer in time to each other (The last milisecond of frame 1/120 dissolved onto the the first milisecond of frame 2/120 and so forth)...?

 

[Brian Harbauer]

Just off the top of my head: Maybe it's to do with the 120 fps. Maybe they found a way to interpolate differently exposed frames shot at very high shutter speeds, closer in time to each other (The last milisecond of frame 1/120 dissolved onto the the first milisecond of frame 2/120 and so forth)...?

Yeah that's kinda what I thought too.
But I'm thinking several reads, without resetting. That way the highlights are recorded, and then the shadows build up. Then it's just merged into one frame.

Question for Red: will this all fit into a 16bit frame or does the codec now support 32bit float?

 

[Enrique Blein]

At 120 fps, you could have 4 frames blended into one. You could squeeze 2 stops under and two over, bringing the DR from 13+ to 17+. The trick would be how to make them look like a still frame and not a blurry one.

 

[Paul Nordin]

At 120 fps, you could have 4 frames blended into one. You could squeeze 2 stops under and two over, bringing the DR from 13+ to 17+. The trick would be how to make them look like a still frame and not a blurry one.

Jim said what they are doing to add 4-5 stops in HDRx has never been done. Frame blending has, so that is probably not it, plus there would be no way to truly get rid of the temporal distortion without a lot of artifacting. Especially in complex moving shots

 

[Pietro Impagliazzo]

Even if Scarlet benefits way less from this, let's say like 2 stops, it would be heaven.

A sub $5K camera with 15 stops of DR. Oh my...

 

[Chase M. Wrenn]

I will be so crushed if this feature doesn't in some way make it to the Scarlet 2/3" model...but I guess that just means you'll have another Epic customer in the distant future ;-)

But on another note, I know a bunch of people on here have been trying to figure out the process. I'm just wondering if it is something that requires the higher data rates of the Epic, or if it's sensor based, etc. Just to see if it's even remotely possible or likely that this feature be inclueded on the smaller Scarlets (as this feature #21 was posted on the 2/3" info reveal...had the hopes up for months, man) O

 

[Lawrence Bansbach]

A sub $5K camera with 15 stops of DR. Oh my...

Unfortunately, we can't really be sure that either is the case, at least when Scarlet 3K is initially released. Jim said Red would try to get HDR working in Scarlet, but maybe if it's not possible at first it can be implemented via firmware update.

And, like everything else, the sub-$5K price is subject to change.

 

[Michel Hafner]

Jim said what they are doing to add 4-5 stops in HDRx has never been done. Frame blending has, so that is probably not it, plus there would be no way to truly get rid of the temporal distortion without a lot of artifacting. Especially in complex moving shots

To avoid strong artifacts one has to simulate the original intended exposure as if the buckets could not overflow. But since they can one has to estimate how much they could have filled up if they had no limit during exposure. So if the sensor is read out continuously one could read out all buckets and reset them. The ones that are full should have a time stamp when they got full during the current read out cycle so extrapolating to how much they would have collected without clipping is possible. There remains some uncertainty with clipping buckets since they can change behaviour over several cycles (for example as highlights pass them). Comparing the time stamps over several read out cycles should help finding out how to extrapolate each bucket best.

 

[Brian D. Goff]

Jim said what they are doing to add 4-5 stops in HDRx has never been done. Frame blending has, so that is probably not it, plus there would be no way to truly get rid of the temporal distortion without a lot of artifacting. Especially in complex moving shots

Wonder if it is possible to start two or even more sensor read-outs at almost the same time (maybe 1 pixel shift). If I understand correctly, since we don't have a global shutter, when one line of sensor pixels is read, the one befor could be free for a new task:confused1:

 

[Lorenzo Straight]

That's exciting to hear Jim. Keep swimming.

 

[James Brundige]

Didn't someone name their R1 "Ansel"? Maybe they were a little ahead of their time.

That was me.

 

[Shane Betts]

That was me.

:yesnod: OK, it's going to be embarrassingly easy for us, compared to the Zone System, pul and push processing and selectively printing to control contrast but HDR is what Ansel was doing way back then.