Leica news: LFI #1/2007

[ennjott]

Advertisement (gone after registration)

This ignores the fact that the 8 bits are not used in the same way that 8 bits are usually used.

You can use 8 bits in any way you want and it will still stay 8 bits, especially with such a primitive kind of reduction. It helps that demosaicing is done afterwards in 16 bit colorspace, but latitude especially in the highlights is lost, period. And yes I have tested this with an M8 DNG and the result was the only logical one (not even a small amount of blown highlights recoverable), but you conveniently overlooked that bit.

 

Have you successfully retrieved the same blown bits with a 5D? If not, you can conclude nothing.

The facts are simple enough that I could actually draw conclusions without ever having used a digital camera. But I often retrieved blown highlights from my high-end 12 bit camera called 350D.

 

The encoding scheme is darn clever, and there is not much missing compared to a full 14-bit file (or 12 or whatever).

Nikon's NEF compression is clever (hence the "9.4 bits") but not a simple LUT or whatever they are using to map each single value to another one. The latter is actually very primitive and not all that different from using an 8 bits JPEG in 16 bits color space.

 

Here's the author of Lightzone on the issue:

Thanks Leica for DNG: Request for 16 BIT files now. - Open Photography Forums

 

Leica's peculiarity (and the one of its customers) used to be likeable but when it reaches into technical incompetence, my patience is limited.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[carstenw]

You can use 8 bits in any way you want and it will still stay 8 bits, especially with such a primitive kind of reduction. It helps that demosaicing is done afterwards in 16 bit colorspace, but latitude especially in the highlights is lost, period. And yes I have tested this with an M8 DNG and the result was the only logical one (not even a small amount of blown highlights recoverable), but you conveniently overlooked that bit.

 

NJ, I certainly did not overlook that! You are trying to do something with 8 bits (recover blown highlights) which isn't even possible with 16 bits (or any number of bits). If they are blown, they are gone. If they are not gone, they weren't blown. We are dealing with digital here, not film.

 

You need to underexpose an image, and then bring it up and see if there is banding in the shadows, not if you can get definition out of black. You then need to overexpose an image, and then bring the exposure down to see if you get banding in the non-blown-out-but-very-bright regions, not see if it can make detail out of white. Just because a camera says Leica on it does not mean it can perform the impossible. I am not expecting that. I want to see if it can retain sufficient definition in stretched regions, nothing more. What else is there which shows the limitations of 8 bits?

 

One important point you do not mention: the end-points of the 8-bit and the 16-bit scales are identical. It is what happens between there which is important, not what happens beyond the ends.

 

The facts are simple enough that I could actually draw conclusions without ever having used a digital camera.

 

Lots of people make lots of conclusions all the time, without the slightest thread of evidence. What we need here is not more opinions. We need proof. If no one else is willing to do these tests, then I will, once I get my M8 back! I had just hoped that out there would be a person who doesn't think that they are more clever than the Leica engineers, who does not think that they can determine without evidence that Leica's idea is bad, but who actually wants to see if they can *prove* anything. Proof is useful, opinions are a dime a dozen.

 

Leica's peculiarity (and the one of its customers) used to be likeable but when it reaches into technical incompetence, my patience is limited.

 

Prove it, instead of talking. Do you have an M8?

 

To give you some justification for the need to *prove* it, all reviews centering around image quality (ignoring the magenta and banding issues since they are not related to the 8-bit issue), have come back with fantastic reviews. Anyone who wants to say that the M8 produces useless pictures has to go up against many great photographers, including at least two Magnum photographers.

[carstenw]

Here's the author of Lightzone on the issue:

Thanks Leica for DNG: Request for 16 BIT files now. - Open Photography Forums

 

You will notice that there is nothing in that thread to back up such a statement. Just the statement itself. I want to see proof. I want to have demonstrated (or to demonstrate myself, or fail while trying), that there is actually a short-coming somewhere.

 

NJ, do you have any idea how many shades of almost-white you can differentiate? Do you have any idea how many shades of almost-white lie between two which you can differentiate, in an 8-bit JPG image?

 

Do you know how the screen on the back of the 350D works, when it shows you blown highlights? If not, why are you using this as an example of how you "recover" highlights?

 

Answer: you are not recovering highlights. The screens on the backs of high-end cameras show "blown" highlights when your almost-whites exceed a certain almost-white value, not when they are actually blown. When whites are actually blows, the value in the DNG,. 8-bit/16-bit, will be 255/16383, respectively, and there is nothing anyone can do to convert that into detail. If you don't believe me, go ask someone who knows whose answer you will respect.

 

It seems to me that you are choosing to believe certain people, without any evidence at all, for no apparent reason other than that they are intelligent. Intelligent people make mistakes all the time. I don't know whether it will end up being me and those who think like I do, who is right, or you and those who think like you do, but I want to *know*, not guess.

 

So far, no test has shown any problems, although some have tried. We need to push further, not argue more.

[sdai]

All tests and comparisons need a target, a reference.

 

Let's say this, Carsten ... without a 16-bit enabled M8, no one can prove the 8-bit DNG files are equally good, right?

[carstenw]

Right, but I would claim that it doesn't really matter. What really matters is whether we can see banding, i.e. sub-optimal results, after heavy-handed adjustments with M8 DNGs. Of course, a 16-bit M8 would be ideal, but a 5D or something would be fine too. We just have to ignore colour differences and such, and focus on the artifacts that editing might introduce on an M8 DNG but not on a 5D CR2.

[blakley]

Well it's clear that an uncompressed DNG from a (currently non-existent) M8 which produced such a thing would create files with more information than the current M8 DNG files - because the compression algorithm used to create the 8-bit files is not lossless.

 

It's also probable that the biggest impact of this "extra information" would be in the highlights - but it's not clear that retaining the additional information would provide extra latitude in the highlights. It might reduce "processing fragility" in the highlights, or it might not, very much - I think we'd need to see an experiment to be sure.

[mjh]

Advertisement (gone after registration)

Nikon's NEF compression is clever (hence the "9.4 bits") but not a simple LUT or whatever they are using to map each single value to another one.

It is just a simple LUT, nothing fancy. Nikon is doing an LZW compression on the result, though. The LUT has 683 entries and you need 9.4 bits for differentiating 683 values (since log2(683) = 9.4). While you cannot split bits, LZW is well suited for dealing with variable length words, and Nikon is using variable length words with up to 10 bits here.

[carstenw]

That's exactly it, Bob. This is not about "latitude", whatever that means. This is about "fragility", as you put it, or visible gaps in the histogram, like Bruce Fraser's book shows for 8-bit JPGs. Can we provoke such poor results with an 8-bit Leica M8-stored DNG? We need to find out.

 

Hello Michael. Nice to see you here. I was going to back you up in the German forum, but there were so many people arguing that I gave up before posting a single time. Do you have an M8 in hand? If so, do you have time to try some pushing of M8 DNGs, and see what happens with respect to banding and other potential artifacts?

[mjh]

Do you have an M8 in hand? If so, do you have time to try some pushing of M8 DNGs, and see what happens with respect to banding and other potential artifacts?

The M8 I have access to is in Solms at the moment, but I will continue to explore this issue as soon as possible.

[carstenw]

Ach, meine auch Morgen bekomme ich meine wieder, vielleicht... bitte bitte bitte!

[rvaubel]

Ach, meine auch Morgen bekomme ich meine wieder, vielleicht... bitte bitte bitte!

 

carstenw

Do you know there is a bonifide M8 16bit file in existance? It seems that the infamous Photokina girl (AKA HP girl) photo that was purlioned out of the show, was shot with an M8 that had an early 16 bit DNG file storage setup.

 

Guess you has a copy of the file?

 

Rex

[rvaubel]

carstenw

Do you know there is a bonifide M8 16bit file in existance? It seems that the infamous Photokina girl (AKA HP girl) photo that was purlioned out of the show, was shot with an M8 that had an early 16 bit DNG file storage setup.

 

Guess you has a copy of the file?

 

Rex

 

Or I should have spelled "quess who has a copy of the file "

 

Rex

[sdai]

Rex, I've read the thread in German ... Michael has played with a M8 prototype with 16-bit DNG enabled and he didn't see any notable difference between the compressed and uncompressed files. Let's wish he could get his hands on it again ...

[carstenw]

Rex, I have the file, but for some reason it is only 4.7MB on my disk? I don't know what is going on with that file. None of the test pictures I have collected is larger than 10.1MB. Even searching for the filename on Google, the picture I come up with is also 10.1MB:

 

http://www.lightmediation.com/blog/podcast/kina/L9994925.DNG

 

It may be that Lightroom re-saves the DNG in compressed format to get the 4.7MB.

[rvaubel]

Rex, I have the file, but for some reason it is only 4.7MB on my disk? I don't know what is going on with that file. None of the test pictures I have collected is larger than 10.1MB. Even searching for the filename on Google, the picture I come up with is also 10.1MB:

 

http://www.lightmediation.com/blog/podcast/kina/L9994925.DNG

 

It may be that Lightroom re-saves the DNG in compressed format to get the 4.7MB.

 

carsten

 

Gee, my file is 59.17MB ! Took about 5 minutes to download. It looks like a low gamma RAW file. But it has the file extention .ppn

It opens in CS2 fine. Do you want it? I'm a little computer illiterate to figure out what to do with it.

 

Rex

[carstenw]

59.17MB??? I don't think I want it What is it exactly, and where does it come from?

[scott kirkpatrick]

It seems that the infamous Photokina girl (AKA HP girl) photo that was purlioned out of the show, was shot with an M8 that had an early 16 bit DNG file storage setup.

 

 

I have that one, and it is a standard 8bit, 10.1MB dng file. The camera had firmware level 1.04.

 

scott

[Englander]

According to the metadata in CS3 Bridge the RAW DNG file posted for Uwe's contest is 16 bit or isn't the metadata how to tell if the camera is using 16 bit?

 

http://www.jirvana.com/contest/lz_04/0611M8_0539.DNG

 

Also, I think the NYC Central Park shots on Asher's Open Photography Forum contest were 16-bit, too. But again, that's from the metadata and if that is not the way tell if it is 16 bit, then perhaps none of these are good samples for you.

 

Hope that helps

 

Joe

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[mjh]

Do you know there is a bonifide M8 16bit file in existance?

Sure, I have got a few (not particularly interesting) 16 bit files from an early M8. These are 20 MB files bearing a “.RAW” suffix – definitely not DNG format. As only Leica can convert those files, they are not much use, except as a proof-of-concept that an M8 can create 16 bit files. Obviously, had this feature made it into production models, DNG would have been used instead. This kind of 16 bit raw file is probably as raw as it gets.

[scott kirkpatrick]

Also, I think the NYC Central Park shots on Asher's Open Photography Forum contest were 16-bit, too. But again, that's from the metadata ....

I have those as well. 10.1 MPx files, 10.3 MB, so they all started as 8 bit representations. Perhaps what you are seeing in Bridge is a transformed form for editing.

 

scott