LFI exposed?

[jaapv]

Advertisement (gone after registration)

I've just been rereading the 1/2007 issue and my eye fell on the following:

 

1.Photographs: the rear end of an old model Mercedes

 

2. Text:"The upper picture was converted from 16 bits colour accuracy per channel using special firmware and software; the one below was taken with the serial model at 8 bits colour depth."[...........]

"We borrowed an M8 that was still running one of the very early firmware versions to [......]Unlike serial models, the pre-serial model was still capable of storing uncompressed RAW files"

 

Look at the photo's: they are hand-held shots and the shadow of the photographer is identical, pixel for pixel in both. So it is just the same photograph, taken once by one of both cameras printed twice.

 

In view of the suggestive text: Translation error? Mistake at the printer? Or just misleading?....

[ChrisC]

.....Look at the photo's.......

 

Jaap - I would like to but do not have access to them. Please describe what you see.

 

................ Chris

[Advertisement]

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

[ho_co]

Interesting question. You're right--it's definitely the same shot.

 

However, the German caption doesn't say that two images were used. It reads

Das obere Bild wurde mit einer speziellen Firmware und Software direkt mit 16 Bit Farbgenauigkeit pro Kanal umgewandelt, das untere entstand wie bei den Serienmodellen mit 8 Bit.

That is, "The top image was converted with 16 bits per channel directly with special firmware and software; the bottom one was generated as in the production models with 8 bits." ["Entstand" could also be translated as 'arose' or 'resulted' instead of 'was generated.']

 

So LFI apparently made a single image, then converted it first with the earlier 16-bit firmware you mention, and then again by applying the same algorithm as used in the production models. If they've written the software to extract the 16-bit data, it should be easy enough to add the 8-bit conversion feature.

 

In other words, one shot but converted two different ways.

 

Seems LFI should have explained that if it's the case; but if they had used two separate cameras and two different shots, some of us would likely be complaining that the comparison was unfair simply due to that fact.

 

I'm glad you called attention to the matter. The paternalistic "well, we tried it and decided what's best for you" still rankles, though the M8 does make good images. We complain at the lack of choice, not at the lack of image quality.

 

--HC

[jaapv]

So you think it is translation error. But why call the second shot "taken by the serial model"after explaining carefully that a pre-serial one had to be borrowed for the first shot? And if "wie bei den...." that is obviously impossible, as the pre-serial one was not configured like a serial one. The only way that could have been done is by using a simulation in raw conversion, which invalidates the test imo.

[jaapv]

Jaap - I would like to but do not have access to them. Please describe what you see.

 

................ Chris

 

Two 100% identical hand-held shots.

[sdai]

I agree with Howard ... it's just one shot differently saved and processed.

 

That can be done pretty simply actually, just flash the camera with the special 16 bit firmware and the serial production firmware twice ... there you go. Also, they can flash the camera with a special firmware ONLY saving the "RAW" data dumped straight from the sensor bypassing all downstream electronics and process it with different firmwares on the lab test bench.

 

It's just hardware bottleneck dragging the camera down to halt when it was writing 16 bit raw files. I believe Leica had told us the truth (or at least what they believed to be truth) while they obviously made the wrong bets.

 

Let's not give any hard time to them anymore, budget and projected MSRP are the only issues. As I've pointed out many times, if the M8 was originally designed as a 6000 or 7000 dollar camera, all issues will become non-issues.

 

People will still buy it ... and now many complain about the price hike coming in October.

 

Very bad ... strategic mistake.

[jaapv]

Advertisement (gone after registration)

I'm not complaing about the camera - I am inquiring about the information supplied by a magazine.

[sdai]

I'm not complaing about the camera - I am inquiring about the information supplied by a magazine.

 

I don't blame you, Jaap. I'm just feeling bad for not having spotted this in the first place and then made a big fuss with it. LOL

 

All being said, Leica is pretty straight forward on this matter IMO. Much better than a very lame business who has kept telling their customers that DX is all they'll ever need and 6 years later, brought out something called FX.

[ho_co]

So you think it is translation error. But why call the second shot "taken by the serial model"after explaining carefully that a pre-serial one had to be borrowed for the first shot? And if "wie bei den...." that is obviously impossible, as the pre-serial one was not configured like a serial one. The only way that could have been done is by using a simulation in raw conversion, which invalidates the test imo.

>Jaap, you and Simon sneaked in your exchange while I was writing this. Simon and I are both upset that you caught this and we didn't! LFI could have been more informative about the method used. Why did the English language version of LFI say 'taken by the serial model'? I don't know. The German text clearly says 'as in the serial model,' which implies but doesn't say that a serial model was used.<

 

I don't want to be argumentative, but here are a couple points:

 

Yes, the images are identical so far as I can tell; the angle doesn't vary at all. That would have been possible by leaving the M8 baseplate attached to a tripod and changing out bodies on it, of course, though again by not being specific about how the images were made, LFI has left itself open to criticism. We can't really tell that there wasn't a tripod involved, because the photographer's shadow blocks the area where the tripod would have been. In fact, he seems to have a hunched position such as one might take if using a tripod. But that's all supposition.

 

I've only (re-)read the image caption, not the article text. The caption isn't self-contradictory, but if the text refers to use of a second, production model, then that is. >I hope to re-read the article with an eye to the possible discrepancy you've uncovered, but haven't time at the moment.<

 

As I see it: LFI got hold of a camera that still had earlier 16-bit software. Then they took a file from that camera and processed it in two ways, just as we might process a file in two raw converters.

 

The algorithm used in the M8 is very simple: Take the raw output value, multiply by four, and take the square root.

 

So once you've got the 16-bit file created by the camera, there's no problem in applying external software to emulate what the series production camera does internally.

 

It doesn't matter whether the camera does it, or whether it's done in post-processing; after all, in our use, we still have to run the file through a raw converter. Here the 16-bit and the 8-bit processing may even have replaced the raw converter because the camera may not have written DNGs. In other words, perhaps the software actually did convert directly from the raw output, not from a DNG-reformatted output of the camera.

 

Your question is valid: How were these two images actually made?

 

I would go further and ask how many sets of images were made before the magazine chose to print this pair. The caption implies that the choice was made in order to show that the white of the car doesn't produce banding as one might expect from the compression algorithm.

 

--HC

[simon_hughes]

I have this issue and in looking at the pictures they do appear to be completely identical in registration. I wonder, however, if the shadow that we are assuming to be the photographer is actually that, or just the shadow of another static element in the image. This might explain why the two images appear to be so exactly the same. Is it also possible that the images were cropped to show the exact same area?

 

It seems to me that I can see a difference in the gradation where the roof pillar meets the main body (above and to the lft of the gas door), and also on the lower left fender area. The (16 bit) top picture shows (to me) more gradation than the lower (8 bit) one does. I see a similar gradation difference in the hubcap too, where the 16 bit version appears more subtle.

 

Cheers,

 

Simon

[Guest tummydoc]

I allowed my subscription to LFI to lapse in 2000. One reason was the text was worded as if it'd been done by someone with six weeks of English-for-the-foreign-born. I could only suspect, not being a German speaker, that if the translator had that poor a command of English, the translation itself was probably rife with errors. So I never knew if the author of an article was an incompetent, or just the translator.

[miami91]

I allowed my subscription to LFI to lapse in 2000. One reason was the text was worded as if it'd been done by someone with six weeks of English-for-the-foreign-born. I could only suspect, not being a German speaker, that if the translator had that poor a command of English, the translation itself was probably rife with errors. So I never knew if the author of an article was an incompetent, or just the translator.

 

Quite true. I can't speak to the accuracy of the translations, but at least with regard to English grammar and sentence structure, I think LFI has improved quite a bit in the last couple of years. I used to be quite bothered (or amused, depending on mood) several times an issue a few years ago, but find the awkwardness of their English to be less of an issue now.

 

Jeff.

[jaapv]

I have this issue and in looking at the pictures they do appear to be completely identical in registration. I wonder, however, if the shadow that we are assuming to be the photographer is actually that, or just the shadow of another static element in the image. This might explain why the two images appear to be so exactly the same. Is it also possible that the images were cropped to show the exact same area?

 

It seems to me that I can see a difference in the gradation where the roof pillar meets the main body (above and to the lft of the gas door), and also on the lower left fender area. The (16 bit) top picture shows (to me) more gradation than the lower (8 bit) one does. I see a similar gradation difference in the hubcap too, where the 16 bit version appears more subtle.

 

Cheers,

 

Simon

 

Well, in that case it is an extremely photographer-like static element. Anyway I think it is justified to call for more rgiour - both in testing method and journalistic accuracy.

[sean_reid]

Shifting the topic a little...one of the problems I see with conclusions drawn about 16-bit vs. 8-bit RAW files is that people sometimes assume they're equivalent if the final output (from a given set of samples) looks the same. But what seems to be missed very often is that the 16-bit file has the potential for being a much richer source to work with as it is being converted (including, potentially, the "reach" a given camera has in the highlights as well the abruptness which which highlights blow out). So, a good comparison, IMO, would come from photographing a high-contrast subject and then manipulating each file in conversion to see how flexible it really was.

 

Cheers,

 

Sean

[ho_co]

Sean's point is the important one. Jeffrey Friedl’s Blog » A Qualitative Analysis of NEF Compression explores the 'virtually lossless' Nikon compressed NEF file.

 

I see no difference between the standard and the compressed NEF versions. But the site demonstrates that there is a mathematically obvious loss of data in the compressed version.

 

Any training class in working with RAW files makes the point that one should stay in 16-bit mode as long as possible simply because doing so leaves open the ability to make large modifications with much less loss of data. That is, with only 256 buckets, applying 'curves,' say, will necessarily empty some of those buckets. Working with even merely a 12-bit file, you've got 4096 buckets to catch the values you're moving. Eight-bit files allow only 256 values per channel; and that's all M8 files allow as well.

 

--HC

[sean_reid]

Sean's point is the important one. Jeffrey Friedl’s Blog » A Qualitative Analysis of NEF Compression explores the 'virtually lossless' Nikon compressed NEF file.

 

I see no difference between the standard and the compressed NEF versions. But the site demonstrates that there is a mathematically obvious loss of data in the compressed version.

 

Any training class in working with RAW files makes the point that one should stay in 16-bit mode as long as possible simply because doing so leaves open the ability to make large modifications with much less loss of data. That is, with only 256 buckets, applying 'curves,' say, will necessarily empty some of those buckets. Working with even merely a 12-bit file, you've got 4096 buckets to catch the values you're moving. Eight-bit files allow only 256 values per channel; and that's all M8 files allow as well.

 

--HC

 

Hi Howard,

 

Right, in the end, it comes down to how hard one is pushing the file during conversion. I miss having 16-bit for the highlights especially.

 

Cheers,

 

Sean

[mjh]

Look at the photo's: they are hand-held shots and the shadow of the photographer is identical, pixel for pixel in both. So it is just the same photograph, taken once by one of both cameras printed twice.

“Pixel for pixel” is an exaggeration, as you would need to compare actual pixels to substantiate that claim, but having said that, maybe it is – the original German text doesn’t suggest that it isn’t based on the same raw file. That is, it may be the result of two different raw conversions, one using the full 16 bits, the other based on the lossy 8 bit format as used in production samples of the M8. I know that LFI at that time had both a standard M8 and a pre-production version creating 16 bit files, because I have seen and handled both, but I don’t know about the genesis of these two images – I would have to ask Holger Sparr, the author of the article in question.

 

Now be that as it may – I have seen people voicing complaints about how the compressed 8 bit format might cause problems that an uncompressed 16 or 14 bit format wouldn’t, but only in theory and never in practice.

[mjh]

Any training class in working with RAW files makes the point that one should stay in 16-bit mode as long as possible simply because doing so leaves open the ability to make large modifications with much less loss of data. That is, with only 256 buckets, applying 'curves,' say, will necessarily empty some of those buckets.

Quite true, but then, all modifications within the raw converter are happening within a 16 bit space, so it is quite unlikely that any manipulations would cause a loss of tonal values. The M8 uses 8 bits for storage, but the 8 bit values are just indices into a 14 bit space of tonal values.

[sdai]

Mystery solved. They're two different images ...

 

Ok, here's the result ... basically, I scanned the images from the LFI magazine and compared them in layers, if they're one shot differently processed, the images would complete collapse with each other and all you'll see is pure darkness.

 

Now you can see everything pretty clearly.

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[Advertisement]

Hallo,

Look here: Leica M

[jaapv]

Mystery solved. They're two different images ...

 

Ok, here's the result ... basically, I scanned the images from the LFI magazine and compared them in layers, if they're one shot differently processed, the images would complete collapse with each other and all you'll see is pure darkness.

 

Now you can see everything pretty clearly.

 

If the typographical process did not enlarge one file marginally more the other. But I admit, the evidence is on the side of different images now.