Dynamic Range of 23 Stops

[John Sprung]

Let's see if I can upload these pictures:

[John Sprung]

OK, so here's a test image and a blowup of a small portion of it. The 500 Watt halogen broad is the only light in the room. This is from the Thompson GV Infinity, the claim is that it has a dynamic range of 23 stops. You can see detail in the coils of the filament and in the shadows lit only by bounce off the wall behind the camera. This was shown at HPA last week.

 

 

 

 

-- J.S.

[Adrian Sierkowski]

That certainly is damned impressive... if a bit lifeless of an image....

Who ever chooses what to put into these test frame anyway?

Still... 23 stops....

Though looking over the specs I wonder what recording time is like on the machine and I'd also love to see what, say, A Resolve could do with that....

[Keith Walters]

OK, so here's a test image and a blowup of a small portion of it. The 500 Watt halogen broad is the only light in the room. This is from the Thompson GV Infinity, the claim is that it has a dynamic range of 23 stops. You can see detail in the coils of the filament and in the shadows lit only by bounce off the wall behind the camera. This was shown at HPA last week.

-- J.S.

Something doesn't smell right here.

 

Up until now, the very best video cameras you can get have a dynamic range of something less than 14 stops, and "range" is very much determined by how much lowlight noise you are prepared to put up with, (or put up with having artificially removed).

 

So, we now suddenly have a camera with another 9 stops of dynamic range?!

I could believe 2 or 3, but nine?!

 

The sensors have overnight gotten 512 times better in performance?

What's the catch...?

[Walter Graff]

Something doesn't smell right here.

 

If you lit the chart with more light than you'd ever light a chart with or could bear, you could easily do that trick. This shot is clearly not what it seems to be. You can clearly see the reflection of a light coming from the right of the camera that is lighting the chart. It creates a defined shadow and creates light through the bottle of wine that is represented on the wall behind the bottle. Bounced light and not do that. Or if you had the broad on a dimmer along with the additional light that is lighting the chart you could do that same trick. And that same light that is lighting he chart is seen as speckled reflections in the bottles to the left and just above the open face. And if you try to quickly blow up the open face fixture, you can see that it is not the same shot as the second shot claiming to be a blow up. Detail is just not there in the wide shot. And notice the hard shadow to the left of the open face (it's barn door creating a shadow). If that open face was lighting itself and the light was reflecting of the wall behind the camera the shadow wold not be nearly 45 degrees camera right as it is nor below the fixture as would be created by a source light which is higher than the open face as it is. . Incident reflection off the wall would make it more head on. Clearly this is a parlor trick. It would easily fail a forensic light test based on all the flaws in the shot based on the claim. I could easily recreate this shot and demonstrate that it is a fake using just a bit of math based on the shadows to not only tell you where the source of light in the room is but at what height, and even what distance based on the sharpness of the shadow.

[Keith Walters]

If you lit the chart with more light than you'd ever light a chart with or could bear, you could easily do that trick. .

Or you could just go and shoot it out in your back yard at midday :D

[Chris Keth]

Is that much dynamic range a good thing? Can you decrease the dynamic range in camera? Think of lighting an interior for that thing. Say you needed a baby for a kicker on an F900, on this thing you may need a 10K to do the same thing without a huge increase in contrast in post. It seems to me that lighting packages would have to go up to be able to light with any decent contrast since, I assume, large increases in contrast in post would come with a large amount of noise.

[Karel Bata]

It could be 2 exposures blended together. Shouldn't that light source create more lens flare? And why is it so green? The test engineers at Thompson don't know how to white balance...? And it's a tungsten source.

[Stephen Williams]

It could be 2 exposures blended together. Shouldn't that light source create more lens flare? And why is it so green? The test engineers at Thompson don't know how to white balance...? And it's a tungsten source.

 

Hi,

 

White balancing in camera would loose loads of dynamic range, it's the same with the Viper.

 

Stephen

[John Sprung]

And if you try to quickly blow up the open face fixture, you can see that it is not the same shot as the second shot claiming to be a blow up. Detail is just not there in the wide shot.

Before we even get that far, of course out here in the real world today, we don't have any means of storing or displaying 23 stops of range. If they're really working with that kind of image data, they must be inventing a way to store it, and showing us what they extract from that in these JPEG's.

 

The other light might possibly be bounce off of something (the operator's shirt?) much closer than the back wall of the room. How they squeeze the big range into the JPEG could have something to do with that.

 

These are the Thomson/Grass Valley folks. I've known some of them for years, and they've been in the biz for like half a century now. The guys I've known from those companies have always been honest and competent.

 

 

 

 

-- J.S.

[John Sprung]

It could be 2 exposures blended together.

 

Yes, that's likely how it works. The camera has a CMOS chip (or chips). Perhaps it has photosites of different sizes for two or more sensitivity ranges, a sort of simultaneous bracketing. The other theory is that the chip somehow polls for saturation during the exposure, and goes by how long it takes to get there. If so, it might have some strange motion issues.

 

 

 

 

 

-- J.S.

[Jonathan Bowerbank]

The scene's lit, and if you stop down & ND enough you can get detail in the filament and rear reflector of any light. So what's the big deal? Do we have any details about how the shot was done or manipulated in post?

 

Ideally, shouldn't the original image be correctly exposed for the chart, with the light itself 11 stops over, and perhaps another chart in frame 11 stop under? Then two corrected images, 1 printed up and one printed down to show us images with reasonable detail?

[Will Earl]

Before we even get that far, of course out here in the real world today, we don't have any means of storing or displaying 23 stops of range. If they're really working with that kind of image data, they must be inventing a way to store it, and showing us what they extract from that in these JPEG's.

 

John, I loath to question such a definitive statement from you, but I would dare say there are plenty of image formats quite capable of storing that much range. To display that entire range all at once is another matter and would require some sort of tone-mapping operation to compress it all into a 0..1 space, of which most tone-mappers tend to produce rather horrible looking images.

 

I'm going to remand sceptical for now, but if it is what they say it is - then it has the potential to be very interesting.

Edited February 24, 2009 by Will Earl

[Karel Bata]

However it was done, I don't get how the gray scale can be gray, while the rest of the scene, like the paper cups, comes out green...

 

Rather negates the point of a gray scale! :lol:

[Phil Rhodes]

Drivel.

 

The bottle at left is casting a sharp shadow on the wall behind it, from a source considerably above its own level. There is clearly additional light involved.

 

It could be green if they've taken it directly off the chips, I suppose.

 

P

[A. Whitehouse]

Who has that much cereal? 23 flavors of Cruesli more likely...

[Adrian Sierkowski]

It's the doll that worries me... those skin-tones don't look natural ;)

[Aasulv Wolf Austad FNF]

Thanks for posting the pics. Interesting that Thomson GV is now showing these High Dynamic Range pics taken with the Infinity camera. Is this the same demo they've been showing before where they control exposure time in each CMOS pixel individually and hence can capture very high dynamic range?

 

Next thing to see is how this affects motion....

[David Williams]

John Galt is working on a HDR sensor, with up to 6 exposures per pixel in one pass.

 

Info is nearer the end.

 

http://magazine.creativecow.net/article/th...uture-of-pixels

[Thomas James]

Yes I believe that John Galt's 36 megapixel sensor is called the Dynamax and it produces 2K at 4:4:4

[Thom Stitt]

John Galt is working on a HDR sensor, with up to 6 exposures per pixel in one pass.

 

Info is nearer the end.

 

I would love to see test footage of this. I'm assuming it would take a fairly intensive post process in order to get the information you want in the frames throughout your shots - Similar to Photoshop compositing for HDR photos? Would there be an entirely new process, like a DI, that involves choosing which exposures should be used in which parts of the frame?

[John Sprung]

What we'd need is a similar process, but with a bigger data structure. AutoDesk is doing high dynamic range now, using 16 bit floating point. But really we need at least a bit per stop, so for 23 stops, 24 bits would be barely adequate. Maybe we should start at 32 bits.

 

 

 

 

-- J.S.

[David Williams]

I'm imagining something analogous to Panalog LUTs. Each frame records the maximum data within your set stop range, and you have a series of generic LUTs you can apply in post. Along with a tool to modify the generics and create custom LUTs, with keyframed moving zones, etc. Done right, it would be very flexible, and not to difficult a process.

[John Sprung]

I'm thinking of some way to record in files all 23+ stops just as they come from the sensor, sort of like Red on steroids. ;-)

 

 

 

 

 

-- J.S.

[Keith Walters]

John Galt is working on a HDR sensor, with up to 6 exposures per pixel in one pass.

John Galt will be no more "working" on that that Jim Jannard works on Mysterium sensors.

He may well have hired someone to design a chip on Panavision's behalf.

 

But this is one of those cherished notions that refuses to die. You are never going to get any more than about 12 stops of useful dynamic range out of any sort of digital video camera, at least one based on silicon. It's a technological brick wall where a whole lot of intractible problems come together to make further progress either impossible or ludicrously expensive.

 

And please do not invoke Moore's Law and other psuedo-technical Haikus. ALL electronic image sensors are Analog devices, with digital post processing. Garbage in - Garbage out. Moore's Law is not a real physical law, and in any case it only applies to digital devices, which have not yet reached the theoretical limits of their development.

 

If they want to try to re-write the laws of Physics, well that will provide more years of free entertainment.