More HDRx...

[Álex Montoya]

Unless The Foundry invented a new method of deriving motion blur that's never shown up in a research paper out of the blue just in time for RED and never mentioned it in relation to their other VFX research projects, yes we can safely conclude they're using oFlow. An Optical Flow algorithm is what The Foundry won its Academy Award for and it's the basis of almost all of their plugins.

I would even go so far as to say it's a safe bet that it's not only based on The Foundry's oFlow, it's based specifically on their new GPU accellerated Blink code which is in the new Kronos.
http://features.cgsociety.org/story_custom.php?story_id=5776


Both streams at the same time is possible and it does appear to be what RED is doing. But it doesn't solve the intrinsic problems with post production blur that aren't unique to RED or Optical Flow or even motion blur. The only system that would avoid these pitfalls is a camera which shot highspeed (which also might be possible for RED in the future). But it would take more than 2x the datarate to achieve (think 10-100x the datarate of a single frame) but might be the basis of their stills HDR solution. (Shoot a burst of motion blur free frames and then recombine them like multi-mass motion blur).

Sorry for my ignorance, but what do you need OF for if both streams are taken at the same time? Then you just have to merge them together, right?

One stream with correct exposure and shutter speed and, above that, the faster, underexposed stream...

 

[Sanjin Jukic]

I do have some fresh examples from using Hydra HDR plug-in at Apple Aperture:

Shot on 7D with no HDR... Location: Golfclub Schloß Schönborn, Lower Austria.

Shot on 7D with Hydra HDR plug-in processing at Apple Aperture....Location: Golfclub Schloß Schönborn, Lower Austria.

 

[Gavin Greenwalt]

Sorry for my ignorance, but what do you need OF for if both streams are taken at the same time? Then you just have to merge them together, right?

One stream with correct exposure and shutter speed and, above that, the faster, underexposed stream...

The faster underexposed stream has no motion blur on it. So unless you want your sky razor sharp and your mountains motion blurred you have to blur the sky to match.

In order to blur the sky you have to figure out which direction to blur the sky in. That's where oFlow comes in. It tracks the sky. Determines what direction and how fast the sky is moving and then applies motion blur to the sky pixels.

In that same case if your sky is all white you only have one image to track: the HDRx underexposed image. So the Foundry can't assume that there is any useful data in the normal exposure. Not only is the motion blurred normal exposure less ideal for optical flow, it's also potentially clipped to white which would be as far as the tracker is concerned unmoving.

In the exposure ranges where both images are visible the optical flow algorithms could theoretically track both to get a better optical flow track than either single image sequence but if you already are within the normal exposure range of both images there is no reason to use the HDRx image: Just use the good exposure. The HDRx is most useful in the parts of the normal image which are overexposed. The parts of the normal image which are overexposed are useless for tracking so the tracking will almost certainly be almost entirely if not entirely using the HDRx image (which is a better source for the track anyway).

And again. Even if the optical flow track were somehow perfect (which nobody has ever done nor probably ever will do for a long long time) the limitations on post production blurring wouldn't go away.

There are theoretically some interesting/cool things that The Foundry could do with both a motion blurred and a normal stream when they can compare pixels at the same point. But the whole point of HDRx is to extend the dynamic range beyond one of the image streams. So everywhere that HDRx is useful is the same place where you only have one stream of data to analyze.

 

[Álex Montoya]

But MagicMotion doesn't seem to blur the underexposed stream, right? Look at the Vegas vid: the highlights are very sharp even when panning.

 

[Gavin Greenwalt]

But MagicMotion doesn't seem to blur the underexposed stream, right? Look at the Vegas vid: the highlights are very sharp even when panning.

Magic Motion isn't The Foundry solution. It's the EasyHDR™ solution.

Personally I find the Magic Motion name kind of confusing. Maybe it's called Magic Motion because it magically looks better than you would expect sharp highlights and half blurry mid-tones to look.

One of the fun things with HDRx™ is that anybody can make their own blending solution. Since everyone has access to the two streams, which is the same data as The Foundry and RED have access to you can roll your own solution. Anybody could write a plugin for REDCineX to blend the two streams to suit their tastes. Personally I hope RE:Vision develops their own HDRx™ implementation. Personally I find their optical flow motion blur and re-timing superior to The Foundry's.

 

[Jannard]

Magic Motion isn't The Foundry solution. It's the EasyHDR™™ solution.

Personally I find the Magic Motion name kind of confusing. Maybe it's called Magic Motion because it magically looks better than you would expect sharp highlights and half blurry mid-tones to look.

Gavin... that is exactly what Adam Wilt said when he saw it. Same with Michael Cioni and Simon Robinson (Chief Scientist for The Foundry)... along with everyone else who has seen it properly on the 40' screen. You seem to be trying to compare this with a standard alternating over/under motion capture. It isn't.

We believe one of the reasons this looks so good is this is how the eye sees. Swipe your arm in a fast circular arc. You don't just see motion blur.

Jim

 

[Jannard]

BTW... there is no way to emulate this motion with any other camera.

Jim

 

[Gavin Greenwalt]

Gavin... that is exactly what Adam Wilt said when he saw it. Same with Michael Cioni and Simon Robinson (Chief Scientist for The Foundry)... along with everyone else who has seen it properly on the 40' screen. You seem to be trying to compare this with a standard alternating over/under motion capture. It isn't.

We believe one of the reasons this looks so good is this is how the eye sees. Swipe your arm in a fast circular arc. You don't just see motion blur.

Jim

Maybe my brain is broken but I just see a blurry hand unless I follow it with my eyes.

And maybe it looks great. I'm not passing judgement. It very well may look better than 'normal' footage. I haven't seen a back to back presentation, although what I've seen looks good. You don't notice any artifacts in the shown examples at all. But it's undeniably 'different'.

It's just to me "Magic" isn't "Wow that actually works, I would have expected that to look terrible." It's "Wow, you can't even tell they did anything to the image." But that's just a personal opinion on naming. For me an HDR image which perfectly matched a normal exposure's motion blur would be "Magical". But that's just based on expectations. It's also fair to say that an image with artifacts that looks better and is more desirable than an image without artifacts is also "Magic". I personally wouldn't call that "Magic" I would call that "Suprising". Suprisingly Better Motion™ probably wouldn't be a great trademark.

 

[David Quinn Carder]

Maybe my brain is broken but I just see a blurry hand unless I follow it with my eyes.

Just my 2¢… I think you're overthinking it. When I first saw the example, I wasn't sure what made it better or different. The difference—which I can see now—is that in real life, yeah, it's moving so you don't get a clear image of a (say) a waving hand, but your brain can see it's a hand. You brain gathers more information about the moving object that you actually "see".

So imagine a car or a motorcycle or something zooming across the screen. Normally, it really would just be a blur, but with Magic Motion (Jim, please correct me if I'm wrong!), you can see the car/motorcycle distinctly—or at least, more distinctly—than you would be able to if shot on a different camera. But it still appears to move smoothly.

I know exactly how HDR will affect my footage, and am very excited about it. I can't say I know how Magic Motion will affect my footage, but I'm looking forward to checking it out on tests I do myself. I don't think it will be useless, even if I choose not to use it.

I know it's early days so I don't want to beg for specific tests, but I'm wondering what sort of effect it will have on high-frame-rate footage (I intend to shoot and output at 72+ fps if possible) and how exactly it will affect other aspects of the image. If I can only use it on some shots, but not others, I imagine I will elect not to use it for consistency's sake. Because I'm a weirdo who pays way more attention to little things like that than is healthy!

 

[Gunleik Groven]

I am all in for the "surprisingly better motion" trademark.

First thing i noticed.
And only in a good way.

G

 

[Gavin Greenwalt]

You seem to be trying to compare this with a standard alternating over/under motion capture. It isn't.

Jim

No I'm comparing it to a currently non-existent 'normal' camera which could capture 18 stops natively. Aka RED Krakatoa or whatever your fourth generation chip is going to be called.

I get it. Exposure begins. Depending on the HDRx +Stops exposure value it reads out a copy of the exposure early as the x Channel and then finishes up the normal exposure. That is a better solution than alternating I agree. No weird gaps. But it's certainly different from a normal exposure. Perhaps better. Perhaps sometimes better and sometimes worse. Perhaps always worse. Either way we don't have 18 stop cameras available yet except with unwieldy 3D rigs so wherever it fails we'll be no worse off than we are currently. Where it does work, we'll have something better. Win Win (as long as you didn't assume that HDRx was going to work without any testing and overexposed your normal channel).

 

[Alex Mack]

Where it does work, we'll have something better.

I get the sense that its working in all situations...
MM is quite like burning through layers of film crystals.. the motion seems a cut above anything digital I've ever seen.

 

[Matthew Cernohorsky]

I get it. Exposure begins. Depending on the HDRx™ +Stops exposure value it reads out a copy of the exposure early as the x Channel and then finishes up the normal exposure. That is a better solution than alternating I agree. No weird gaps.

That was my guess on how they did it originally. Has this been confirmed or is it just speculation? I think it's a pretty clever approach.

 

[joshua csehak]

That was my guess on how they did it originally. Has this been confirmed or is it just speculation? I think it's a pretty clever approach.

The article (in which the guy interviewed Jim) says it takes an exposure normally (say, 1/48), and then without any gap, takes a shorter exposure. It's pretty clear that they're not taken at the same time, unless I read it wrong.

Gavin: you clearly have read the wiki article more closely than me. You win

 

[Matthew Cernohorsky]

The article (in which the guy interviewed Jim) says it takes an exposure normally (say, 1/48), and then without any gap, takes a shorter exposure. It's pretty clear that they're not taken at the same time, unless I read it wrong.

Hmmm...but Jim has confirmed that there is no temporal difference between the exposures. I guess we'll figured it out in due time when epic is released or possibly sooner....

 

[Gavin Greenwalt]

The article (in which the guy interviewed Jim) says it takes an exposure normally (say, 1/48), and then without any gap, takes a shorter exposure. It's pretty clear that they're not taken at the same time, unless I read it wrong.

That's incorrect. The HDRx exposure is a subsection of the main exposure and it's at the beginning not the end. As to whether the normal exposure includes the HDRx timeframe or the two are split and then averaged back together to form the "normal" exposure even without Magic Motion is unknown and really ultimately irrelevant since they would look exactly the same if it was in fact possible to do that without a gap. I doubt it's possible without a gap. That's why I'm betting the trick they've pioneered is probably the early readout without interrupting the normal exposure and it sounds like that's a cool hardware trick they had to build into the chips to achieve.

Jim also said it's only possible to have the HDRx read at the beginning not the end. So that also discourages the idea that they are two separate frames stitched together and not an early read of the same exposure. If it was just recording another frame without a gap they could do it arbitrarily anywhere in the frame (even centered). The fact that it's only the very first part is a strong hint that it's two reads of one exposure at different times through the exposure.

 

[Uli Plank]

@ Sanjin: We did extensive testing when we had our RED (#1462) for half a year and didn't find any tone-mapping algorithm that didn't create strange fluctuations in motion. They seem to be all optimized for stills. I'd be glad to be proven wrong.

@Gavin: I agree that Optical Flow algorithms from RE:Vision are better then The Foundry's, but nothing would stop us from experimenting with them as soon as we have those two streams from HDRx™. There are so many possibilities to apply them and blend/mix/combine the results in any good compositing program.

Regarding "Magic Motion", I think RED discovered something very interesting about human vision accidentially AFTER developing the most intelligent solution to HDR in motion – even if it's RED I don't believe this was planned. But it definitely deserves further investigation. I got to apply for some funding from our research institutions…

 

[Sanjin Jukic]

@ Sanjin: We did extensive testing when we had our RED (#1462) for half a year and didn't find any tone-mapping algorithm that didn't create strange fluctuations in motion. They seem to be all optimized for stills. I'd be glad to be proven wrong.

Regarding "Magic Motion", I think RED discovered something very interesting about human vision accidentially AFTER developing the most intelligent solution to HDR in motion – even if it's RED I don't believe this was planned. But it definitely deserves further investigation. I got to apply for some funding from our research institutions…

Uli,

I just finished rendering 1 (one) second of my experimental R1-MX HDR enhanced footage and didn't discover any motion related issues.

DOWNLOAD R1-MX HDR test footage 1.4 MB >>>>

A snapshot from HDR generated part of the test footage in Photomatix Pro.

A snapshot that used for HDR processing @ ISO 100.

A snapshot that used for HDR processing @ ISO 1600.

Curve Settings at RCX for the second part of the test footage for those skeptics who think is better solution than using HDR, but really not..

 

[William Young]

Nice!

 

[Shane Betts]

No I'm comparing it to a currently non-existent 'normal' camera which could capture 18 stops natively. Aka RED Krakatoa or whatever your fourth generation chip is going to be called.

Gavin, I'm curious as to what you're labelling a 'normal' camera. Is it a film camera? If so, at what shutter angle? Is it a video camera? If so, interlace or progressive and what shutter speed? Is it a digital camera? If so, once again, what shutter angle?

I don't believe I'm the only one who has found progressive motion capture from electronic cameras to be somewhat less pleasing than the images that come from film cameras. I believe this is partly to do with the motion blur and partly to do with the 48Hz cadence of 24fps footage as delivered by film projectors.

For me, and quite a few others it would seem, the effect Magic Motion™ delivers, with its combination of sharp and blurry motion is the most pleasant representation of motion seen - from electronic or sprocketed cameras yet and delivers what appears to be a leapfrog over the look of film. Only time will tell.

Of course, there could be pitfalls ahead. Of course SFX guys might find the whole thing a pain in the arse. But, ultimately, story wins out and, if this motion makes the movie easier and more seamless - more believable - to watch, then post people had better get used to it. From what I've seen, I'll need some pretty strong convincing not to shoot everything this way. I love the look!