Leica news: LFI #1/2007

[carstenw]

Advertisement (gone after registration)

The output of (nearly) gamma-encoded 8 bits data is a pretty ugly crutch and will kill headroom for exposure corrections or significant changes to the white balance. And this headroom is already small enough at 12 bits compared to film.

Also, Nikon's compressed NEFs use the same method so it's nothing new. You can disable it on the higher end models, and it should certainly be possible to disable it on a Leica.

 

I've been taking part in this discussion in the German forum:

http://www.leica-camera-user.com/digitalforum/11644-eine-frage-der-kommunikation-4.html

 

On page 8, I posted some samples, linear and gamma/color-corrected versions of the same image, to demonstrate the large processing step that gets taken out of your hands by the M8's current firmware.

 

There are only 6 pages... Could you post a direct link?

[sdai]

http://www.leica-camera-user.com/digitalforum/11644-eine-frage-der-kommunikation-8.html#post122319

[carstenw]

I think I see what is going on. The number of pages is dependent on your user setting for how many posts per page Mine is 30. I don't see any pictures with your link either, Simon. Are we talking about the quoted Queen Mary pictures here, or the link to the dpreview discussion?

[carstenw]

I just read the first few pages of the thread in the German forum, and the same happened as usual: a number of self-proclaimed experts ("I am a computer scientists...", "I am a photography professional...", blah blah) posted incredibly insulting and opinionated comments with little fact, and very quickly it devolved into a (not very) subtle insult-contest and nothing really got decided at all in the pages I read, except that people grouped themselves into three groups: the self-proclaimed experts who think that Leica has just committed suicide; the self-proclaimed experts who claim to understand where Leica is going with the algorithm (I cautiously belong to this group); and the people who simply post around the war at hand.

 

I find that forum incredibly tiring. I find it much more constructive here in the international forum.

 

One guy in particular (leicageek or somesuch), as well as his group of fans, found it appropriate to post over and over again how storing data in 8 bits is so incredibly dumb that even their dog wouldn't have done that (my words). Not a single one of them, however, has actually worked with the data to see what happens.

 

I think we can all agree that Leica's choice to store their raws in 8 bits makes us nervous. However, I think we can also agree that Leica has over 100 years of making excellent technical choices, and putting out the best lenses and cameras to an impressive degree, and that for that single reason alone, they deserve that we take a very careful look at the results of their choice, and don't simply argue based on whatever (bullshit) "facts" we have on hand.

 

From my previous posts and analysis, I see that there is some loss, but that it is less than expected. What remains to be determined is if it is visible in common scenarios. What are important scenarios? Two come readily to mind: printing large (poster-size, even), and web/computer viewing. Are there others? For the purposes of web viewing, we can probably assume a maximum size of somewhere around 1000 pixels on the long side. Viewing larger images than that is too hard for too many people. Probably the printing test will end up being harder. Nonetheless, I want to see if we can provoke artifacts in full-size images viewed on the computer. While it might not be relevant to any final-product scenario, it would still be worrisome.

 

Here is what needs to happen: we need to test a series of worst-case scenarios for 8-bit files, and see if we can spot the differences. Unfortunately, the LFI test, while well-intentioned, is not the worst-case scenario. They simply photographed, analysed and printed images with a lot of subtle variation in the bright regions.

 

The worst-case scenario, however, is not that, but images which need heavy editing. Perfectly exposed 8-bit images, even JPGs, have never offended anyone. The problem is editing imperfect images. I wanted to do this myself, and still will, but I think that an acceptable worst-case scenario might be to take the same photos with a 5D, 1Ds2 or D2x(s) and an M8, over-exposed and under-exposed by 2-3 stops, and then to correct the exposure in the raw converter and seeing if banding appears. Over- or under-exposing by more than 2-3 stops would probably not yield acceptable results anyway, so the results of such a test are less relevant. The subject matter should both be high-key subjects and low-key subjects, to give a full idea of the results. Is anyone up for doing such a test, and posting final images, as well as 100% crops here? I will do so when my M8 comes back, but perhaps someone else feels like taking up the gauntlet meanwhile?

 

To be complete, the final results should be printed or down-scaled, if banding does occur, because this process step may hide it again.

[waterlenz]

You should see the Dutch manual for Canon DSLR's. Sometimes utterly incomprehensible as it was from Japanese to some kind of English to Dutch by translators who had never seen a camera. "B setting" came out as "lamp setting."

{B=originally blower bulb (to operate the air-driven ancient shutters)=bulb=lightbulb=lamp. }

 

Not to mention the fact that most folks from tulip-land should know what a bulb is :->>>

Tom

[wide.angle]

Anyone happen to know where I can buy a copy of the current LFI in Chicago off a newstand, etc?

[Peter41951]

Advertisement (gone after registration)

Try your local Leica dealer. My UK one keeps stock.

[jlancasterd]

You should see the Dutch manual for Canon DSLR's. Sometimes utterly incomprehensible as it was from Japanese to some kind of English to Dutch by translators who had never seen a camera. "B setting" came out as "lamp setting."

{B=originally blower bulb (to operate the air-driven ancient shutters)=bulb=lightbulb=lamp. }

 

Camera manuals have always been a source of unintentional humour. IIRC a 1960s Pentax manual once translated the instruction 'do not touch the instant-return mirror' as 'do not touch the flipping mirror'... Graphically precise and direct - but probably not what was intended!

[johnll]

IIRC, the manual for the computer I am writing this on has the following in the early pages of the manual "Do not throw the computer in the fire".

On a more serious note, I just aquired a Canon G7, which only produces JPG (therefore 8-bit) files. Earlier G-series digicams produced 12-bit RAW files, so I was nervious about this. I have not done the extreme testing suggested above, but both with my old G5 and new G7, I found that if I converted the JPGs to 16-bit TIFFs, then edit them, then convert to 8-bit TIFs for printing, then so far I have not had any "banding" / posterizatiojn issues. On the JPGs, of course, there is no additional latitude - if you blow the highlights, then they cannot be recovered, so you have to be careful not to blow any that you don't intend to. I find the G7 to be a sort of poor-man's M8. The one thing about it that I find a nuisance is the optical viewfinder - only about 80% coverage. I'll still get an M8 when the "issues" are resolved. For now, I'm "making do" with an M7, G7 and 5D.

[carstenw]

John, that is encouraging, although to be sure, we of course need to test the actual M8, due to its funky bit storage.

[ennjott]

One guy in particular (leicageek or somesuch), as well as his group of fans, found it appropriate to post over and over again how storing data in 8 bits is so incredibly dumb that even their dog wouldn't have done that (my words). Not a single one of them, however, has actually worked with the data to see what happens.

While he isn't particularly polite, he does have a point. Also if you read my posts you will have noticed that I have actually worked with a DNG and tried to get back some detail from blown highlights, unsuccesfully of course. I also linked to dpreview's test of the 5D's dynamic range in raw. It has good examples with texture coming back into what was previously perfectly white areas.

This latitude, well known from film, is what's missing with 8 bits per channel.

 

However, I think we can also agree that Leica has over 100 years of making excellent technical choices, and putting out the best lenses and cameras to an impressive degree, and that for that single reason alone, they deserve that we take a very careful look at the results of their choice

Apparently, digital technology is where their excellent choices end.

I just saw Nikon uses a more complex method using an average of 9.4 bits per pixel for their compressed NEFs: Jeffrey Friedl’s Blog » A Qualitative Analysis of NEF Compression

This method does retain a good bit of headroom.

 

Of course the M8's reduction is not visible on a monitor or a print under normal circumstances, probably not even in a worst case scenario. This doesn't need to be tested. But it also doesn't need to be tested that 8 bits has no latitude, and this has been bad enough already with 12 bits per pixel compared to film. These files simply don't have much headroom for editing, and no amount of discussion will change that.

[carstenw]

While he isn't particularly polite, he does have a point. Also if you read my posts you will have noticed that I have actually worked with a DNG and tried to get back some detail from blown highlights, unsuccesfully of course. I also linked to dpreview's test of the 5D's dynamic range in raw. It has good examples with texture coming back into what was previously perfectly white areas.

This latitude, well known from film, is what's missing with 8 bits per channel.

 

N., this is exactly what is so lacking in that whole thread: the constant reminder that 8 bits is no good. This ignores the fact that the 8 bits are not used in the same way that 8 bits are usually used.

 

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

 

Anyway, that has more to do with dynamic range and less with potential gaps in the histogram. We need to shift the values a whole bunch, and then look at the gaps in what remains, not the bits at the end.

 

Of course the M8's reduction is not visible on a monitor or a print under normal circumstances, probably not even in a worst case scenario. This doesn't need to be tested. But it also doesn't need to be tested that 8 bits has no latitude, and this has been bad enough already with 12 bits per pixel compared to film. These files simply don't have much headroom for editing, and no amount of discussion will change that.

 

You are yet again treating the 8 bits in a Leica M8 DNG as if they are the same as the 8 bits in a JPG or something. This is very much *not* the case. The encoding scheme is darn clever, and there is not much missing compared to a full 14-bit file (or 12 or whatever). The question is if it is enough.

 

This is where we need to stop arguing, because we will never get anywhere. We need tests. I hope to get my M8 back today.

[sdai]

Carsten ... here's one clever test although it was done with Nikon compressed and uncompressed NEFs. Leica's 8-bit mapping into the DNG works pretty much the same way as how Nikon stripped the "redundant" bits ... of course, the conclusion is always drawn by yourselves.

 

Jeffrey Friedl’s Blog » A Qualitative Analysis of NEF Compression

 

Dare I say Leica's 8-bit mapping is stupid? definitely no ... but it surely makes no sense (to me) if it was done in favor of writing speed and saving some disk space ... memory cards are dirt cheap these days.

[eronald]

In the end why not let the customer decide ? If Leica thinks selling me a camera that shoots decent Raw files with no artefacts and acceptable color is something I do not "deserve" because I'm only a customer and not a photographer (fashion photography apparently doesn't count as photography to people like you) I will sell this camera. I have had enough of this snobbery.

 

Edmund

 

 

The worst-case scenario, however, is not that, but images which need heavy editing. Perfectly exposed 8-bit images, even JPGs, have never offended anyone. The problem is editing imperfect images. I wanted to do this myself, and still will, but I think that an acceptable worst-case scenario might be to take the same photos with a 5D, 1Ds2 or D2x(s) and an M8, over-exposed and under-exposed by 2-3 stops, and then to correct the exposure in the raw converter and seeing if banding appears. Over- or under-exposing by more than 2-3 stops would probably not yield acceptable results anyway, so the results of such a test are less relevant.

[carstenw]

Edmund, I certainly do consider fashion photographers to be photographers. I don't see anything in the post you quoted which says anything related to your comment.

 

I do hope that Leica puts a switch in the menus for the 8-bit saving.

[ho_co]

The Friedl blog page is interesting, but it's qualitative. He has done a lot of analysis of the files and has found a very good way to display them and prove that there is a difference between 8- and 16-bit files; but his intention is to simply demonstrate the accuracy of Fazal Majid's calculations at Is the Nikon D70 NEF (RAW) format truly lossless?.

 

In other words, Friedl's basic approach is to take two files that show no visibly obvious difference and demonstrate that there really is a meaningful difference. The LFI article is reported to say (I haven't received my copy yet) that Leica looked at their two images and saw no difference, so went with the compressed DNG.

 

My paltry understanding of what was beaten to death in the 8- vs 16-bit thread on this forum was that although there was virtually no visible difference in the image as output by the M8, it was mathematically provable that data were lost.

 

So it seems to me that it would be nice if someone like Friedl could compare the 16- and 8-bit versions of M8 files (not likely since no consumer will see the 16-bit versions) as he did the Nikon files.

 

Or, failing that, it would be nice if someone who understands the math involved in the M8 compression algorithm could compare it with the Nikon compressed NEF calculations and tell the rest of us how similar they are.

 

Or perhaps, the best thing is that you just say to me, "Forget it. Look at the finely nuanced files the M8 writes and quit worrying about it. Oder, wie Faust zum fliegenden Augenblick hätte sagen können, 'Verweile doch, Du bist so schön'."

 

--HC

[jaapv]

Camera manuals have always been a source of unintentional humour. IIRC a 1960s Pentax manual once translated the instruction 'do not touch the instant-return mirror' as 'do not touch the flipping mirror'... Graphically precise and direct - but probably not what was intended!

 

Exactly how I, as an old rangefinder dog, feel about SLR's :D

[Lode]

David Blattner & Bruce Fraser "Real World Photshop CS" page 230:

 

"The key thing to bear in mind is that when you use Photoshop's tonal controls, you're stretching and squeezing various parts of the tonal range, and doing so, you inevitably lose some information. A key point to understand is that you lose a great deal more information in 8-bit-per channel files than you do in high-bit ones".

 

There's also an example that demonstrate this lose of information.

 

Lode

[carstenw]

On a purely theoretical basis, it is of course true that we lose something. This is trivially the case. The interesting part is if it is visible with a normal workflow.

 

The Bruce Fraser statement is also true, but *still* doesn't state that it is visible. There is loss, this is known, but has Leica been clever enough to avoid significant losses, detectable with the bare eye? This is the crux. Secondly, Bruce Fraser is talking about files with 8 bits used in a linear fashion, where the highlights are quite well covered and the shadows very poorly covered. This is not the case with the Leica M8 DNG files, so we need to re-check all our assumptions, not blindly refer to them as if nothing has changed.

 

To re-iterate: this is a new approach. This has not been done before.

 

We need tests, not talk.

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[blakley]

"The key thing to bear in mind is that when you use Photoshop's tonal controls, you're stretching and squeezing various parts of the tonal range, and doing so, you inevitably lose some information. A key point to understand is that you lose a great deal more information in 8-bit-per channel files than you do in high-bit ones".

 

There's also an example that demonstrate this lose of information.

 

Yes, but that's not the situation we're in here. The M8's files are compressed to 8 bits, but (if you shoot RAW, which I'm presuming you will if you care about this issue at all), they are expanded to 16-bit TIFFs by Capture One (and, I presume, other RAW processors, though I've not used another one yet). So when you're using Photoshop's tonal controls, you're not working on an 8-bit file, you're working on a 16-bit file. It's a 16-bit file which has been created with less than 16 bits of data per pixel, but all that means is that the RAW processor had to do some interpolation to create data at some of the pixel sites - it does not mean that the file "behaves like" an 8-bit file in Photoshop - it does not!