HDRx motion interpolation

[Shane Betts]

Hey Ruairi.

I've seen this line of questions before on HDRx and it puzzles me a little. Maybe it's me who has the wrong end of the stick here (sure wouldn't be the first time) but, Storm's interpolation aside (no idea how that works), it seems obvious to me that there isn't any computer generated pixels here to louse up the image. It's just two exposures, using the same pixels, from the same sensor, at the same (start) time - only one is a longer exposure. Blend in the longer exposure more (say 70/30%) and you get more motion blur (wider shutter angle), pull the highlights back more (say a 30/70% mix) and you get less blur and more stutter (because you're using more of the image with the more narrow shutter angle). The computer is doing nothing more than blending two images - it doesn't "make" any new pixels.

Try it yourself. Shoot something that you can be relatively sure of getting accurate motion repeated (a machine for instance) twice, locked off. Once with a narrow angle and again with a wider angle shutter (or using a 3D rig). Comp them up and see how it looks.

Like I say, I could be way off here but that's how I believe it works. There will, no doubt, be limitations but I don't think we need to fear the "automatic button" effect with HDRx. Just my 2c.

 

[Aaron Newsome]

It's just two exposures, using the same pixels, from the same sensor, at the same (start) time - only one is a longer exposure.

Shane, this is the first I've heard that the two images have the same start time. I thought each exposure was taken one after the other so they would have different start times. Is this spec posted anywhere that explains authoritatively how it all work?

 

[Gunleik Groven]

Shane, this is the first I've heard that the two images have the same start time. I thought each exposure was taken one after the other so they would have different start times. Is this spec posted anywhere that explains authoritatively how it all work?

It has been suggested a number of times...

 

[Aaron Newsome]

It has been suggested a number of times...

Shane's post didn't sound like a suggestion. It sounded like something he knows to be fact. I'm just curious how it really works.

 

[Floris Liesker]

In the Las Vegas Firetruck HDRx™ footage the two exposures have the same end time.
Or rather, the short exposure is taken back to back after the long one.

 

[I Bloom]

My point is that you are screaming 18 stops 18 stops, and the fact that it may come at a cost to motion quality is going to get lost in the hype. Because "most people won't notice the difference". And maybe they won't but this is something I look very carefully at. And it's not a tradeoff I find thrilling. Great dynamic range or accurate motion? Pick any one! you make this choice sound much easier than it may end up being. Who wants to give up one over the other?

You can't really blame Red for marketing their products as best they can. I'm glad they are excited.

I look at it this way. It seems the camera has great DR already, enough to make the cost benefits analysis move many productions toward Red. I also have questions about HDRx but I think of it like this, which looks worse:

--some different motion blur in parts of the image, that may or may not be noticeable or even undesirable.

or...

--blown out highlights.

I think in most cases one will do everything in there power to control highlights within the normal track. But in situations where there is a huge uncontrolled loss of detail its nice to know that repair is possible and maybe kind of easy. I think I even saw some of this loss in a movie that's probably winning best picture this year, so I guess it happens to the best of us.

Regarding monitoring paths for HDRx:
Seems like the best and probably obvious thing is just to make a user programmable button that switches the view between the normal and highlight track, just as we now often switch from Redcolor to Raw. I think I would prefer that to an in camera combination of some kind. I guess I'm hoping the design of the camera reflects the fact that many of us would like to keep track of exactly what parts of the image are augmented by this technology.

IBloom

 

[Jannard]

... you are screaming 18 stops 18 stops.

R.

We think 18 stops is pretty significant news regardless of whether or not it is perfect for every situation.

Jim

 

[Shane Betts]

Shane, this is the first I've heard that the two images have the same start time. I thought each exposure was taken one after the other so they would have different start times. Is this spec posted anywhere that explains authoritatively how it all work?

It's my understanding Aaron. Otherwise there would be major difficulties in combining the two images together - it simply would not work. Jim has already stated that you can use the A image to lock to motion tracking. If the two images were taken at different times, this would not be possible.

Many suggestions have been made on these forums, by people way smarter than me, as to how this (HDRx) could be done and the only way multi-image HDR can possibly work in motion is if the two frames are taken at the same time. Now, whether they start or finish at the same time, I'm not sure.

Despite my prayers I don't have an Epic yet :sad: so I can't be talking actual known fact. Only a very select number of people can know anything for sure. But it has been alluded to a number of times and it, frankly, can't make sense any other way. You think Michael Cioni or Mark Pederson or John Schwartzman would be raving over the tech if it simply didn't work for motion? So, no, it's my assumption. Draw your own conclusions.

 

[Shane Betts]

Shane's post didn't sound like a suggestion. It sounded like something he knows to be fact. I'm just curious how it really works.

WTF? My statements such as "Maybe it's me who has the wrong end of the stick here (sure wouldn't be the first time)" ... or "it seems obvious to me" ... or "I could be way off here but that's how I believe it works" were read by you as a statement of something I know to be fact?

No wonder Jim gets annoyed by people not reading posts.

 

[Tim Nolte]

In the Las Vegas Firetruck HDRx™™™ footage the two exposures have the same end time.
Or rather, the short exposure is taken back to back after the long one.

The light streaks precede the pinpoints from the underexposure portion of the HDRx™. If you look at the headlights of the fire truck, the streaks shoot out ahead of them, rather than trailing behind. This indicates that the two exposures start at the same time, with the underexposure ending first. It's like a first curtain sync when using strobes on a still camera.

A while ago, a lot of people were speculating about how HDRx™ works. I don't think anyone official ever said how it does works, but a lot of suggestions were officially ruled out. If I recall correctly, Jim specifically said that there are not two different exposures from two non-overlapping spans of time, as this would create double image problems. My apologies if I'm putting words in your mouth, Jim, and I hope you'll correct me!

(Edit: Jim has corrected me on my recollection of his statement. We still have no confirmed information on how HDRx works, just some educated guesses.)

 

[Jannard]

The light streaks precede the pinpoints from the underexposure portion of the HDRx™. If you look at the headlights of the fire truck, the streaks shoot out ahead of them, rather than trailing behind. This indicates that the two exposures start at the same time, with the underexposure ending first. It's like a first curtain sync when using strobes on a still camera.

A while ago, a lot of people were speculating about how HDRx™ works. I don't think anyone official ever said how it does works, but a lot of suggestions were officially ruled out. If I recall correctly, Jim specifically said that there are not two different exposures from two non-overlapping spans of time, as this would create double image problems. My apologies if I'm putting words in your mouth, Jim, and I hope you're correct me!

That is not correct.

Jim

 

[Gunleik Groven]

That is not correct.

Jim

Which part?

 

[Adam C Lubkin]

Regarding monitoring paths for HDRx™:
Seems like the best and probably obvious thing is just to make a user programmable button that switches the view between the normal and highlight track, just as we now often switch from Redcolor to Raw. I think I would prefer that to an in camera combination of some kind. I guess I'm hoping the design of the camera reflects the fact that many of us would like to keep track of exactly what parts of the image are augmented by this technology.

IBloom

That would be great, especially if the histogram switched as well to reflect the exposure of the track being displayed. I don't know if that's possible but it would be awesome if it was.

 

[Tim Nolte]

That is not correct.

Jim

Presumably you mean the part about denying two separate exposures. My apologies. I've edited my original post so no one gets the wrong idea if they stumble across it in search or something.

I looked at some past posts, and I think I must have been extrapolating from posts like these:
(I know, I know, you hate when people read into things! I'm really trying to quit! :blush

In keeping with RED philosophy... a new way.

Jim

Nope... "just two exposures" gives disjointed (unconnected and stuttering) motion.

Jim

 

[Jannard]

Presumably you mean the part about denying two separate exposures. My apologies. I've edited my original post so no one gets the wrong idea if they stumble across it in search or something.

I looked at some past posts, and I think I must have been extrapolating from posts like these:
(I know, I know, you hate when people read into things! I'm really trying to quit! :blush

With all due respect, I wish people would quit trying to interpret what I am saying. No one seems to get it right. You can't take part of a sentence and build the whole story. At some point we will make it clear exactly what is going on.

Jim

 

[chrislancaster]

I dont think its our job to figure out how it works.. thats RED's


all we should care about is that it does work and how to make best use of it


I mean why do you care ? are you guys building cameras or are you creating images??
:confused1::confused1:

Jim I cant wait to purchase an Epic...!!

 

[Patrick Ready]

Whether the use of HDRx™™ is deliberate or not, and given that it is intended to be a tool in post, is there an argument why it wouldn't be enabled on every frame shot? In every situation?

 

[Tim Nolte]

I dont think its our job to figure out how it works.. thats RED's


all we should care about is that it does work and how to make best use of it


I mean why do you care ? are you guys building cameras or are you creating images??
:confused1::confused1:

Jim I cant wait to purchase an Epic...!!

I strongly disagree. Having at least some understanding of how the tool works means I better understand the potential uses and limitations. As such, I appreciate the more detailed explanation Jim just posted.

 

[Tim Nolte]

With all due respect, I wish people would quit trying to interpret what I am saying. No one seems to get it right. You can't take part of a sentence and build the whole story. At some point we will make it clear exactly what is going on.

Jim

I think we got crossed wires here or something. I was trying to apologize, and concede the point, but you seem to have taken it in some other way. Again, I'm sorry for any offense.

Funny enough, but the longer explanation you've since posted in a separate thread more or less exactly matches my prior understanding. "Motion-conjoined," as opposed to any gap in time between exposures, is what I was trying to describe. Apparently I didn't do a very good job.

 

[Jeff Coatney]

They're stored in tracks in a singular R3D.
In RCX you'll be able to view the each track or the hdr blend.
Exporting to EXR will give you a multiview EXR with both exposures (like a stereo EXR).

But essentially we're frame interleaving here, right? There's no surplus bandwidth to write the frames at "the same time". And the X track exposure correction is a function of time between read-reset cycles. This would account for the unique motion blur that be achieved. Also the efficiency of making this part of a single R3D stream so the media can handle it. But in theory, the X track could be the "first" frame in the sequence giving the motion streak of the A track as a trailing, comet-tail effect, right?