More HDRx...

[Michael Hastings]

Quick question: Will HDRx™ work with 3D? If I shoot with two Epic Xs in 3D config, can I use HDRx™?

I should defer to the experts, but from what I understand it (the sharper hdrx stream) should help greatly in doing better optical flow analysis so better software rerconvergence and such.

 

[JD Holloway]

Well, I'm thinking specifically of using two Epic FF35s in 3D config for IMAX 3D HDRx™™™ if I can get this China IMAX film off the ground. :drool:

Sounds fantastic! Somehow Red is appropriate for China...

On another note,
HEY GUYS is switchable sensor hardware sensitivity up and running? I posted earlier.

"Earlier in the Epic development you guys talked about a selectable hardware sensitivity, 250 or 800 ISO if memory serves.
I was wondering if this as been implemented/tested yet? If so, does it have some cost to the total dynamic range (thinking along the lines of Adam Wilts PVC Alexa test)?
But if HDRx™ is active I would guess much of what is lost (non HDRx™) could be regained with HDRx™ and then some.
I'm wondering about improved blacks in scenes that don't require an 800 ISO sensitivity or max DR, but would still benefit from the extra headroom and Magic Motion.
While headroom is now awesome (tilt-steering), beautiful blacks are the other side of the equation."

Just hoping to hear, I know you guys are under the gun 'n all.

Happy Canadian Thanksgiving weekend !

 

[Gavin Greenwalt]

I'm not sure that's entirely accurate. HDR technically should mean High Dynamic Range, or lots of stops. But if that was the case, the MX sensor as it stands would be HDR. HDR really means HDRBTMTPT (High Dynamic Range By Tone Mapping Two Pictures Together). But there are different ways of tone mapping.

You're making it more difficult than it needs to be. HDR is just "Lots of stops" unfortunately nobody ever bothered to define what "lots" is.

Personally I define any image with 16+ stops of information "HDR".

You don't have to have multiple shots or exposures combined together to meet the definition of HDR. That's just the easiest way we've managed to get it so far. Some of the first color movies were 3 rolls of film. But that doesn't mean a movie isn't "color" unless it's shot on 3 separate rolls and then recombined. It's just an awkward, inconvenient workflow.

If RED released a single exposure, single stream image that was 16 or more stops I would say that's HDR even though there is no stitching or exposure stacking. And personally I think even with 16 stops of native range HDRx™ would still be useful if only for the motion blur free tracking channel.

To date most true HDR images and HDR image formats have no tone-mapping. They're linear data. (well, admittedly, only kind of linear-- technically speaking the file formats themselves like EXR store the data logarithmically, but it's intended to be read out as a linear float or half float value).

Lastly to all those people who split up their single exposure photos in order to tone-map them. You're wasting your time. You can just run the tone mapping algorithm on the original single exposure and get the exact same result. Just hit File-> Save As -> *.hdr Accomplishes the same thing.

 

[Gunleik Groven]

Oh. Sticking my head out again... -
But here's the reason for it.

Simple = good
Extended DR = HR

For many people
RED = hard and geeky

To me:
RED = pretty simple

BUT explanations like
"You can really extend the DR of the chip by developing a lot of tiffs"
Is not simple.

When it also is not true... It creates confusion and uncertainty.

Confusion and uncertainty = bad

The result is neither wider effective latitude, nor easier workflows but more confusion
Thus = Bad.

There are things about the RED which isn't outright simple, yet.
But they are getting there.

And by the way:
Simple is NOT to avoid the "powers of RAW"
I dig RAW.
But RAW in itself = simple
Thus
RAW = good.

RAW implementation in workflows can be confusing, though.
Like the current ISO/preview relation is.

But that can be fixed.

The nice thing about the MX R1 is that ISO works the way you could only dream about as to noise/gain problems.

Thus the first half of the ISO equation now = good

The next half, controlling the actual recorded latitude is still not there, but it will get there, as the HDRx thingy won't make much sense without a better way to protect the latitude on the preview than what we currently have.

Thus I see less confusion in the near future.

Less confusion = good.

G

 

[Uli Plank]

Hi Gunleik,

neither do I like such tone-mapped images you were linking. But some of the best HDR specialists arrive at images that look more like paintings than photography to me – and things can get interesting here.

The basic problem is the big difference between human perception and our reproduction technology. Human vision 'scans' a scene in rapid eye movements and the brain is constructing the image as a whole. So we can perceive an image as if our eye is capable of around 15 stops (max, it's scene dependent). But it is not (just like there's only a narrow region where we detect color and full resolution), our aperture and gain (bio-chemicals) are adapting all the time. Think auto-iris plus auto-gain.

This could only be reproduced (theoretically) with a system where the eye of the viewer is tracked and the reproduction system adapts. Unfortunately, this would be a single viewer experience, and we'd rather like to work for cinema, wouldn't we?

The best contrast ratio we can get in a black room with black seats, walls, ceiling and black clothing for visitors plus a very good projector would be about 15 stops. But people don't like to go to a cinema dressed for a funeral, and then there are regulations about visibility of exit lights, stair steps and the like.

In real life a good cinema can achieve around 9 stops. So we can't reproduce the 15 stops humans could make use of or the 18+ stops from HDRx. We'll make aesthetic decisions in post about how to use the range we recorded, which is a great freedom RED is giving us. We may decide to squeeze some tonal ranges and stretch others in the whole frame to get to those 9 stops. Fine.

But who says that this is the only possibility? Digital technology is giving us the possibility to adapt locally, and neither technical limitations of todays tone-mapping algorithms not taking motion into consideration nor bad examples of abuse of HDR in images should stop us from exploring this new frontier. Maybe some films will look like paintings one day, what's wrong with that? Painters have always painted what their mind sees. Cultural conventions change.

We are used to crushed blacks from analog films S-curve, which is not congruent with our vision (while highlight compression is). We are used to the temporal resolution of 24 fps. We are used to grain. Does this mean that cinema will look like this forever? Do we need to add grain? Do we need to stay 24 fps forever? Do we need to crush blacks?

Somebody said that fiction is just a controlled (incepted…) dream, he was referring to literature. Of course, it's true for fictional film as well. As long as there are viewers who like our dreams, we may do what we want with what HDRx is offering!

 

[Lorenzo Straight]

The range is explosive. Go HDRx!

 

[Sanjin Jukic]

Hi Gunleik,

neither do I like such tone-mapped images you were linking. But some of the best HDR specialists arrive at images that look more like paintings than photography to me – and things can get interesting here.

The basic problem is the big difference between human perception and our reproduction technology. Human vision 'scans' a scene in rapid eye movements and the brain is constructing the image as a whole. So we can perceive an image as if our eye is capable of around 15 stops (max, it's scene dependent). But it is not (just like there's only a narrow region where we detect color and full resolution), our aperture and gain (bio-chemicals) are adapting all the time. Think auto-iris plus auto-gain.

This could only be reproduced (theoretically) with a system where the eye of the viewer is tracked and the reproduction system adapts. Unfortunately, this would be a single viewer experience, and we'd rather like to work for cinema, wouldn't we?

The best contrast ratio we can get in a black room with black seats, walls, ceiling and black clothing for visitors plus a very good projector would be about 15 stops. But people don't like to go to a cinema dressed for a funeral, and then there are regulations about visibility of exit lights, stair steps and the like.

In real life a good cinema can achieve around 9 stops. So we can't reproduce the 15 stops humans could make use of or the 18+ stops from HDRx™™. We'll make aesthetic decisions in post about how to use the range we recorded, which is a great freedom RED is giving us. We may decide to squeeze some tonal ranges and stretch others in the whole frame to get to those 9 stops. Fine.

But who says that this is the only possibility? Digital technology is giving us the possibility to adapt locally, and neither technical limitations of todays tone-mapping algorithms not taking motion into consideration nor bad examples of abuse of HDR in images should stop us from exploring this new frontier. Maybe some films will look like paintings one day, what's wrong with that? Painters have always painted what their mind sees. Cultural conventions change.

We are used to crushed blacks from analog films S-curve, which is not congruent with our vision (while highlight compression is). We are used to the temporal resolution of 24 fps. We are used to grain. Does this mean that cinema will look like this forever? Do we need to add grain? Do we need to stay 24 fps forever? Do we need to crush blacks?

Somebody said that fiction is just a controlled (incepted…) dream, he was referring to literature. Of course, it's true for fictional film as well. As long as there are viewers who like our dreams, we may do what we want with what HDRx™™ is offering!

Uli thanks again for those good points in this discussion.

Now I'm back again with my HDR tests out of RED1-MX files and getting better.

Below are the latest examples. Right now I would need to rename all of them (due to the test smaller tif files) to batch process in Photomatix Pro HDR application.

Then later I would be able to show here a short clip out of it.

Stay tuned.

Chosen two exposures in RCX for this test were ISO100 and ISO2000.

Exposure ISO 100.

Exposure ISO 2000.

HDR image tone-mapped in Photomatix Pro.

 

[Corrado Silveri]

Just want to' say thanks to Uli for his great post.

 

[Roberto Lequeux]

There is only one logical line to draw for an updated HDR definition, and that of course has to be human sight. For something to have high dynamic range it should have "more than normal" DR. Normal as in -the most- DR humans can see.

I read somewhere that actual and instant DR without bio-chemical or iris adjustments is closer to 9 stops, but we see in our minds. So perceived DR should be used, which I suppose is 15 stops?

 

[D Fuller]

OK, I'm not an expert, and this is coming from a background in stills, but as I understand it:

- An HDR image file is just an image file that has greater bit depth than the individual raw files (32-bits in at least some formats, whereas many RAW files, for example, have 12 bits).

No, HDR and bit depth are two quite different things. Bit depth is about the resolution of color you can represent; DR or HDR is about the range of light and dark you can represent. So increasing bit depth gives you more shades of blue, so that you don't see banding in a blue sky. Higher dynamic range lets you see detail in clouds and in shadows under the trees in the same image. (or inside the barn and outside in Jim's image)

- You can take that 32-bit file and make it displayable on a 8-bit display by reducing the contrast. As I understand it, this is NOT tone mapping. It's just reducing contrast.

Contrast has nothing to do with this. You display a 32-bit file on an 8-bit display by reducing color resolution. That's why an image with a beautiful blue sky may look lovely on a 10-bit Dreamcolor monitor, but show banding on a 6-bit laptop monitor. Contrast is not affected. You see the same amount of detail in shadows, though it's likely to show banding. You will loose detail in highlights because a lot of data is required for highlight detail, but they won't be any less bright. The contrast of the image is unaffected.

- You can tone map that 32-bit file and get really ugly crap that, unfortunately, has become popularly known as "HDR", but it's really just ugly tone mapping. But, tone mapping doesn't have to be ugly. It's basically just locally adjusting contrast between nearby pixels.

You've got it right here, except that, again, it's not about a 32-bit file, it's about a file with, say, 18 stops of dynamic range.

Obviously, we don't know exactly how RED is implementing this, but I vaguely remember a post that EPIC was going to be 16 bit, so, if that's the case, I'm guessing that Easy HDR consists of the light and dark images being combined in camera across the 16-bits of data available (possibly with a higher bit "intermediate" that then gets reduced in contrast). Whereas, for HDRx™™™™™™™, the combination into a single file is done on the desktop. In any case, I'm guessing that it's mostly a contrast reduction step in camera to produce a flat file, with tone mapping options on the desktop.

To separate the bit depth/DR concepts, imagine that you have two boxes of paints -- one has 8 tubes: white, black and 6 other colors; the other has 64 tubes: white, black and 62 other colors. No mixing is allowed. You can make a much more richly complex color painting from set 2, but black is still black, and white is still white. That's the difference in bit depth.

Now imagine that someone gives you a new tube of super-black and a new tube of bright-white. That's an increase in dynamic range (DR).

 

[Jeremy Torrie]

The comments about only what the human eye perceives is moot.

When Sony and Panasonic were saying 1920x1080 was the pinnacle of HD, Jim said 4K, then 5K...and similarly 35mm film can still resolve much better at what 8K or 12K and with what 15 stops?

However, what we're talking about is having better data to play with (from resolution to dynamic range), and thus have a greater opportunity creatively to produce a great picture.

To me, and probably others, that is what is getting us excited...Jim and co are taking the technology as far as they think they can go -and then push it some more, which is fantastic.

 

[Lee Fingersh]

...and similarly 35mm film can still resolve much better at what 8K or 12K and with what 15 stops?

Yes...if it's B&W film, and IMAX/medium/large-format sized.

 

[jamesbridges]

Hey Jim,
How does the HDRX translate image-wise to the viewfinder?
Do you choose one image over another or can you combine them?
Thanks

 

[Michael Olsen]

Hey Jim,
How does the HDRx™ translate image-wise to the viewfinder?
Do you choose one image over another or can you combine them?
Thanks

This is one question I have seen come up and always go unanswered:

How does one monitor HDRx?

 

[joshua csehak]

You're making it more difficult than it needs to be. HDR is just "Lots of stops" unfortunately nobody ever bothered to define what "lots" is.

Personally I define any image with 16+ stops of information "HDR".

Personally, I like your definition better, but the beauty of the internet is you don't have to argue, you can just look things up: http://en.wikipedia.org/wiki/High_dynamic_range_imaging

(That's also the beauty of an iPhone. Since getting one, my barroom arguments have gone down 87%.)

There is only one logical line to draw for an updated HDR definition, and that of course has to be human sight. For something to have high dynamic range it should have "more than normal" DR. Normal as in -the most- DR humans can see.

I read somewhere that actual and instant DR without bio-chemical or iris adjustments is closer to 9 stops, but we see in our minds. So perceived DR should be used, which I suppose is 15 stops?

Yeah, if someone wants to change the definition, I think 15 stops sounds about right. Scientific perception is moot I think, except for scientific application.

I used to think human perception was the pinnacle, but thinking about it now, I don't care that much about being able to match what we see, though that would be nice, and I'll probably have my epic/scarlet set on +3 HDR all the time to do that. But what we see is usually pretty ugly. I think the goal for me is to be able to match what we see in an art museum.

 

[Bryce O]

Correct me if I'm wrong, but I'm pretty sure the sensor size only affects resolution. So Scarlet footage should look just as good, but smaller -- unless there are other limiting factors. Have you used any of the 2/3" panasonic or sony ENG cameras? They look pretty darn nice. A little blurry at 2k, certainly not as crisp as Red footage, but those 2/3" chips knock the 5D and 7D out of the park, for my money.

I was under the impression that Scarlet would still have a higher pixel density that the Epic. I could be wrong though. I haven't done the math myself.

As far as ENG cams go, the last time I used one was for some USA Labatts hockey tourney. I also used my 7D and the producer was blown away at the DSLR footage... I think his words were something like "What the hell did you shoot that on? That doesn't look like anything we have..." He almost shit a brick when I pulled out my SLR. Note that they use two HPX-3000's (and they were not as fun to shoot with). I know Scarlet's going to look better, but in terms of all current ENG 2/3" cams... even the producers agree, it's DSLR all the way.

And yea, most of the spot ended up being deep-focus 7D footage. Since it was ice hockey out on a lake with the sun overhead and I was at f22 the whole time. I think allot of it had to do with the fact that we just couldn't get the HPX to expose for the details in the snow and the player's jerseys/equipment at the same time. Every time the players were exposed properly, the ice was just a giant white blob. That's why I don't buy the tests that claim these DSLRs only have 8 stops... there's just no way. Maybe that's what you get to work with once everything's compressed... but the initial exposure AT CAPTURE has to be much higher.

 

[Bill Anderson]

:

"The best contrast ratio we can get in a black room with black seats, walls, ceiling and black clothing for visitors plus a very good projector would be about 15 stops....

In real life a good cinema can achieve around 9 stops. So we can't reproduce the 15 stops humans could make use of or the 18+ stops from HDRx™™™. We'll make aesthetic decisions in post about how to use the range we recorded, which is a great freedom RED is giving us. We may decide to squeeze some tonal ranges and stretch others in the whole frame to get to those 9 stops. Fine."

----------------------------------------------------------------------------------------------------------------------------------------------

I hear this a lot, and find it quite confusing. Uli, the camera interprets the scene's brightness and contrast, and provides that translation to the display. The camera does not deliver the scene's twenty stop DR any more than a neg delivers it ( the neg would glow) It delivers an interpretation of the scene. A photo paper emulsion has a very very short DR, but enough to beautifully translate the 15 stop scene info contained in a neg. If your claim is accurate then we would never have been able to show the DR of color neg stock - not even the poorest quality print.

Maybe I'm completely missing why this ( the display doesn't have a 20 stop or eye quality DR therefore it won't matter if you shoot 20 stops DR ) post comes up so often.??????????????????

 

[Gunleik Groven]

Hi Uli,

I don't see a single point we are in serious disagreement about, in your post -

We represent "more" stops within "less" stops all day, most commonly on our 7-ish bit low-latitude computermonitors and iPhones and whatever, where we are impressed with the 18'ish stops Jim shows us as recorded from the Epic.

I it is hardly desireable to "naturally" project an image like the one Jim has posted at the beginning of this thread. We would need shades to have a good time looking at the highlights and need to take them off to see what was going on in the dark. And on our 7'ish-buts low-latitude LCDs/iPhones or whatever we are atching these images with, it isn't possible anyway.

I am not into HDR, as I said. I do like extended range, though.

This whole discussion started off with an image that was well within the MXs range being dubbed HDR.

If HDR = tonemapping, well then I have everything wrong, and I am not very interested in it.

If HDR means extending the range of a sensor and displaying it in a viewable range, then I am all for it. How you choose to grade your shot is an artistic thingy that has no relevance for this discussion.

Now...

What originally set me off is the amount of confusion about the RED I am sometimes met with that makes people think that they have to do "hard stuff" and semi-voodoo'ish rituals involving multiple tiffs @ 52 MB/frame merged on a frame-by frame basis in some HDR software to get the full latitude and little noise off the sensor and into post - like in Sanjins example.
Because that is just BS.

9 out of 10 times - like with Sanjins file - it is totally easy to recover full latitude with a couple of mouse-clicks.

It was more like a gut-reaction:
Please don't say that totally rudementary actions needs some kind of secret voodoo sauce and multiple programs and plug-ins to work. That's just FUD.

You can hit two buttons and gently adjust the curve and: hey you're done. Not hard. Not very impressive. Nice that it works and is that simple in many occations. I like that.
Thanks Graeme, Deanan RED and team! The voodoo is all yours!

It's like when Sanjin claims that the R1 MX has 14 steps, and then some other cam comes with a documented 14/15 stops. And REDs own test shows 13'ish stops on the un-HDRx™™™'ed Epic.

I understand his enthusiasm, but the result is that what RED shows from the Epic now seems less impressive, while it is in fact totally mindblowing.

The intentions are good. I really like him. And I love his enthusiasm! But:
The results are... in my world at least: Not that good.

My point is:
Don't make simple stuff look almost unachievable.
That's just gonna scare people off the platform.
And most things - as to photography and post use. inclusive grading, keeping full range, getting low noise and deliver beautifull images is pretty straightforeward these days.
THAT is a very important thing to get through....

In addition we have all the added bonuses of the system, but they are still pretty simple and straightforeward.
Mystisism is only good to mystify....

I see that I keep de-railing this thread. And bite myself in my chin to try to avoid it. So:

The HDR discussion was a bi-product of that, and I will now take a break from this discussion.

Making hard things simple is impressive.

Making simple things hard, not so much

Respectfully:

Gunleik

 

[Sanjin Jukic]

Gunleik,

I'm hardly getting what you wanted to say!!!???

The thing is very simple because there are several theories about HDR (see more at already quoted Wiki HDR>>>

Also one more question to you should be:

Have you ever work with HDR standalone application or plug-in to get more about that how you can HDR enhance your images?

If ever you looked deeper in it then probably you wouldn't get so dizzy about this HDR thing here because you would understand more about

tone mapping, exposure fusion, exposure bracketing and many other parameters that are involved in making one HDR image.

BTW, I'm still renaming my 600 images and footage will come soon.

No Voodoo with my things here and only such a simple idea about "knowing" MX sensor strengths and weaknesses as an average RED User.

Stay tuned.

 

[joshua csehak]

I was under the impression that Scarlet would still have a higher pixel density that the Epic. I could be wrong though. I haven't done the math myself.

As far as ENG cams go, the last time I used one was for some USA Labatts hockey tourney. I also used my 7D and the producer was blown away at the DSLR footage... I think his words were something like "What the hell did you shoot that on? That doesn't look like anything we have..." He almost shit a brick when I pulled out my SLR. Note that they use two HPX-3000's (and they were not as fun to shoot with). I know Scarlet's going to look better, but in terms of all current ENG 2/3" cams... even the producers agree, it's DSLR all the way.

And yea, most of the spot ended up being deep-focus 7D footage. Since it was ice hockey out on a lake with the sun overhead and I was at f22 the whole time. I think allot of it had to do with the fact that we just couldn't get the HPX to expose for the details in the snow and the player's jerseys/equipment at the same time. Every time the players were exposed properly, the ice was just a giant white blob. That's why I don't buy the tests that claim these DSLRs only have 8 stops... there's just no way. Maybe that's what you get to work with once everything's compressed... but the initial exposure AT CAPTURE has to be much higher.

That's really interesting. The stuff I've shot on ENG cams looks, at 2k, night-and-day better than stuff I've shot on DSLRs. You're right though, the dynamic range isn't the best, though I suspect it's gotten better. Haven't used one in a while; all my jobs are either DSLR or Red lately.

The 7D definitely has more than 8 stops of DR. I just shot a test where I spot-metered a white house outside at 32, and a shadow inside at 1.6. Placing the white house just below clipping, I was able to take a picture of the whole scene with 8 stops of DR (10, if you want to count pure white and black), and using Lightroom, was able to get 2 more stops from the bottom (down to about 1.4), and even recovered some more highlights (there was a whiter section of the house that was blown out). Taking a video of the same scene and messing with it in Color, I can push it to about a stop less than I can with the still, with no highlight recovery, if I don't mind the noise that comes with it, which at iso100 isn't bad, and 800 is reasonable but starts to get distracting. IM(Personal)O

Do you count pure white and black when you tally DR? If you do, I'd put the 7D at 11 for video, 12 for still (stops #1 and #11/12 being black and white). Boy, I wish I had a Red One on hand to do the same test on. My earlier tests have been more scientific. I'm all about the pragmatic tests now