M8-why 10MB-vs-DMR 20MB

[Andrej Kolev]

Advertisement (gone after registration)

Jamie, I've recieved a mail and downloading a file now.

 

John, now I'm understand your view. My thoughts: Bayer array was specially designed in next form: elements that are sensitive to green part of spectrum, and thus capturing most important part of information, are place in checkboard order, less valuable red and blue channles are placed at remaining position in simple non-checkboard order. Checkboard placement choosen by Bayer allows to make quite good estimation of pixel values (mostly because we can make estimation of edge presence and direction). Using correlation between color channels we can make more precise interpolation of green and especially red and blue channel. As result their detail level becomes more close to a green channel.

 

Yes it is not a perfect restoration, but with advanced algorithms it is quite good. Precision of all calculations of course is limited to precision of input data.

 

Do you think that 8-bit coding will make interpolation less precise? Maybe, but I suppose difference will be non-significant. Anyway I'll try to make direct comparison of both methods.

 

Camera is making interpolation with full precision (altough inrterpolation algorithm may be simplier than you can use on desctop). The main limitations of jpeg is non-controlled transformations from original linear data performed by camera processor - contrast modification by using nonlinear transform, additional sharpness/smoothing filters, so than you wont to make corrections you need to use already heavilly processed and lossy data. For my camera difference between RAW and JPEGs is huge, you even dont need to do some postprocessing to see it. In JPEG mode It applies sharpening, boosting noises and than JPEG compression creates from all that garbage some kind of mess. I dont know why but RAW mode even with grater level of sharpnening and writing to JPEG gives noticeable cleaner image.

 

Anyway we need to be sure in result, thats why i will make comparison using 16 bits RAW files from DMR (from my calculations it's dynamic range is about 12 bits, but very close to 11 bits), but it is the best source of images that we have.

[Advertisement]

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

[bernardo]

Hi folks,

 

tell me, what is the m8 like. Is it worth bying?

 

Show me some samples please!

 

Thanks!

 

.: Bernardmarksfotografie :.

[gogopix]

Hi folks,

 

tell me, what is the m8 like. Is it worth bying?

 

Show me some samples please!

 

Thanks!

 

.: Bernardmarksfotografie :.

 

Dear Bernard

 

Really thoughtful image making on you site. I particularly like you Sri Laka series.Whatever you use I think it will be fine.

 

BTW, been travling a lot? Out of touch?

 

Doesn't matter. There's a rumor Leica produced a RF camera, called of all things the M8. As you see from this thread some of the best minds in the world can't figure out if it is 8 or 16 bit.

Some don't believe it even exists. There was a rumor of one in Paris, but the logo was taped over. Likely a hoax.

So lots of time to catch up. In the meantime prices of used M lenses going thru the roof.

 

There won't be much of a rush since it seems 95% or RF users are either sticking with film,

or they are just waiting for full frame or both.

 

I have one on pre-order, either 5 or 50 in line, difficult to confirm. I ordered when it was called LeicaM Digital, soit maynot even count.

 

If they do exist, and are shipped, US customs may not clear them, if they were shipped with lithium batteries, since this is a possible fire danger.

 

Anyway, read some of the threads, and we will see if this thing ever materializes.

 

Well, I've got to go and re-charge my R9/DMR batteries... excuse me

 

Regards

Victor

[johnll]

Andrej,

Looks like we're really on exact same wavelength after all!

Best regards, John

[Barry Pearson]

Is the lookup table different in different files, or always the same?

 

This is what I have seen in M8 DNGs from software/firmware versions 1.04 & 1.06. The look-up table appears to be derived from a simple algorithm:

 

Resultant value = square of the 8-bit value divided by 4

 

In other words, the 8-bit values are as though the M8 has a 16-bit value internally, and simply takes the square root to store in the DNG. Then the raw converter uses the look-up table to generate a value in the range 0 to 16383, for which 14-bits is sufficient. There are 255 different values, unequally spaced, used within this range.

 

I'm pretty confident about the above. I've been using the tool dng_validate.exe from the DNG SDK in verbose mode. These tags are not just "comments" on the contents of the DNG file - they are a fundamental part of the decoding of it, and they are not errors in the camera. I am reasonably familiar with the DNG specification, and I have probably written about DNG as much as anyone independent of Adobe:

DNG articles and links

 

Obviously, this says nothing about what later versions of the firmware may do. Possibilities could be:

 

- Leave it as it is because it gives good pictures! (Picture quality surely trumps FUD about number of bits?)

 

- Add standard DNG compression (JPEG lossless compression) to the above, giving files typically about 7 MB or less, perhaps as an option. For interest, this is the same type of compression now used in Canon cameras.

 

- Store the 16-bit (or whatever) values in the DNG instead of deriving the 8-bit value. Perhaps as an option. Perhaps combined with (optional) DNG compression.

 

For information: any ACR from 2.3 onwards can read and process the M8's DNGs, which are version 1.0.0.0. (There is a later version 1.1.0.0 which was published in February 2005). But ACR 3.6, just released, can not only read Leica's embedded profile, but also has Adobe's own profile for the M8, presumably resulting from Adobe's own tests. (The choice can be made via the drop-down menu in the Calibrate tab). That is unusual for a camera using DNG in-camera.

[Jamie Roberts]

Barry--

 

Thanks for this; it's where the thread has been leaning anyway--an 8bit encryption of a 16bit value is fine for me; rounding errors are likely to be insignificant.

 

But I'd still like the option of lossless compression (which may not be possible with the hardware involved) or writing out the full 16bit data (which should be possible, right?).

[Andrej Kolev]

Advertisement (gone after registration)

Jamie, I will get your file soon (I have a low-bandwidth). But I've made a preliminary comparison using 14 bit output from my camera, altough that pictures did not utilize all its dynamic range. I've used 8-bit output with Rec 709 gamma (with standart value 2.22 and 3.2) Difference is very small, it is visible on my LCD display with PVA-matrix only after taking a good look, very slight changes in tonality with one image and did not noticed it at all with another. Estimation of difference is next: mean=0.85, std.deviation=0.5; Will make a full report with all data and images after experiments with your images.

My preliminary resume -the difference is insignificant, altough as I've said before that images was not suitable for right comparison. Also my interpolation algorithm is very complex and with another algorithms results can be another. But i think it is better to compare images using it, because my processing pipeline is very precise, fully consists of floating point operations.

[carstenw]

This is interesting and a little worrisome in terms of image quality.

 

But, reading between the lines of what Sean has said elsewhere, I think he is very impressed with the overall image quality, and that it is at least as good, but probably better than the DMR.

 

From what I see on Sean's site, I suspect he doesn't do a lot of post-processing on the pictures he takes.

 

I am certain that the pictures will look great as taken, but what happens when someone wants to make a non-trivial exposure adjustment in post? The bits to hold the smooth gradations may not be there, and the result might be banding. I hope for the best but fear the worst on this issue. My M8 purchase is on hold until I am satisfied that this will not be an issue when editing photos with more than trivial adjustments.

[carstenw]

Honestly, this is starting to get ridiculous. There is simply NO WAY that Leica, (or any other camera company) would even consider releasing a pro-level body, that would capture only 256 levels of color in each channel.

Any idiot would be able to see the banding in their first shot.

 

Well... that depends. If they simply took the colours and distributed them evenly throughout an 8-bit space, yes. If, however, they used some perceptually-based distribution, as was rumoured, you would not see it. They would position the samples closer together in the shadows, where banding first shows up, and reduce the number of bits in the highlights, where we can't discern them.

 

My main concern is what happens under heavy editing, where the values end up far from where they started. There are all sorts of valid reasons for such editing.

 

Sean, if you read this, could you take two shots, over- and under-exposed by, say, three stops, and then bring them back in post? Do you see banding?

[carstenw]

Mark, I dont know. We only know that 2^14 is the maximum value of back-transformation LUT. But here is some new info - such type of coding was used in imaging modules of several space missions. Some reasearchers stated that such mapping helps to match the quantization level to the photon noise (I've wrote about photon noise before) and helps to distribute quantization errors uniformly across the dynamic range.

Also there is some statements that such transform provided a dynamic range of 14 bits in the 8 bits.

 

This is very interesting in another way too. Does this not mean that with M8 raw files, there is no longer any reason to "expose right"? Maybe Leica has brought us full-circle and we are now supposed to "expose correctly" again. I for one would welcome such a change. Previews would be much improved. It just remains for me to see what effect heavy editing has on such a DNG file, and if banding is ever an issue.

[Otto_Fosch]

Hi:

I have talked to a Leica specialist this morning. They plan a statement on this 8 bit topic!

He told me that they use the orginal14 bit from the kodak sensor for processing the 8bit DNG file. He confirm that the processing is not loosless. But their test show no visible differences to the original data. They always decide for quality. The advantage of this approach is the faster storage time. It is theoretical no problem to get the original data, with a slower processing. If they offer this option in the future is not clear.

They develop the software exclusively in cooperation with the Jenoptik. JENOPTIK AG

 

Hope this helps.

 

Frank

[carstenw]

Andrej ... yes, I understand the gamma issue (I was just pointing out that one could have variations on that theme), but I still feel uncomfortable with the apparent bit depth of 8 in the DNG files. Why bother to capture at 16 bits (even more than most DSLRs, which mostly capture 12 or so) if you are immediately going to throw away half of them {which represents a huge fraction of the available levels 1 - (2^8)/(2^16) }? ... and if the files are only 8-bit, why are they so large (10MB i.e. uncompressed)? Lossless compression is well known; there are numerous algorithms in common use. And whatever magic you apply, you cannot recover the lost data. Maybe you can minimize its impact, but you can never get it back. We can certainly get "any mapping from the original data" but I'm not convinced the errors are going to be "minimal" under all circumstances. It just does not make sense not to use most, if not all of the information captured by the camera. And Leica have expressly stated somewhere in the specs that the DNG files have 16-bit "resolution" (and if that does not mean 16-bits of data I have no idea what it does mean). I guess we'll find out eventually, but it would be nice to have some clarification from Solms.

 

Interestingly, perhaps we only needed the 16 bits of depth so far exactly because they are so poorly distributed. In that case, it is at least conceivable that with proper distribution, 8 bits suffice, even with editing.

[scott kirkpatrick]

This is very interesting in another way too. Does this not mean that with M8 raw files, there is no longer any reason to "expose right"? Maybe Leica has brought us full-circle and we are now supposed to "expose correctly" again. I for one would welcome such a change. Previews would be much improved. It just remains for me to see what effect heavy editing has on such a DNG file, and if banding is ever an issue.

 

I think that is an important question, and deserves more than a short answer. We have a generation of cameras with magic ISO dials (as if the ISO/ASA dial on the back of the old M series had developed the ability to change film frame by frame). I think users of the higher ISO settings have already stopped trying to "expose to the right," since that means giving up some of the extra sensitivity that they just dialed in.

 

In the film days, I could shoot short rolls of a white wall bracketing over ten stops and develop them differently to see which "zones" would remain usable and what negative densities they would receive when the film was pushed to higher ASA ratings. (Remember Ansel Adams' "The negative," the H&D curve, and the Watson bulk film loader?) In effect, Sean Reid's reviews do that, but jump to the image to analyze the results. It is probably time to develop this level of understanding for digital photography.

 

I use an Olympus E-1, which is a good DR performer, but not great at high ISO, so I haven't been motivated to do this. But with an M8 in another month, I'll probably try it -- if the serious reviews haven't teased these issues apart by then by exploring the ISO-DR tradeoffs, and not merely the ISO-noise/grain issues.

 

scott

[Jamie Roberts]

Hi:

I have talked to a Leica specialist this morning. They plan a statement on this 8 bit topic!

He told me that they use the orginal14 bit from the kodak sensor for processing the 8bit DNG file. He confirm that the processing is not loosless. But their test show no visible differences to the original data. They always decide for quality. The advantage of this approach is the faster storage time. It is theoretical no problem to get the original data, with a slower processing. If they offer this option in the future is not clear.

They develop the software exclusively in cooperation with the Jenoptik. JENOPTIK AG

 

Hope this helps.

 

Frank

 

Frank, thanks for this. I look forward to their statement, and I like the speed of the new DNGs compared with working on my DMR. No doubt about it.

 

But I would like a lossless option. My concern is more about colour shift than actual banding, though I have to say that my DMR doesn't behave the same way my 1ds2 does in extreme conditions. I had thought that was down to bit depth...but maybe it's just CCD vs CMOS sensors handling edge conditions.

[Jamie Roberts]

{snipped}I think users of the higher ISO settings have already stopped trying to "expose to the right," since that means giving up some of the extra sensitivity that they just dialed in.

{snipped}

 

Hey Scott--just the opposite in practice, actually. You just get way less noise exposing to the right. Not particularly with high ISO images (in fact, low ISO images are closer to the noise floor), but in general, you need to do this to prevent noise and quantization errors.

 

Some cameras are much more forgiving than others with this.

 

People get this a little backwards until they're doing it a lot: you don't dial in higher ISO to get more sensitivity; you dial in higher ISOs to keep the exposure "to the right" and get more information to distribute.

 

In this sense, it's better to shoot higher than lower for noise. A good (to the right) exposure always trumps a bad one (well, except for the limits of the sensor; my DMR at 1600 "push" is pretty bad, I must say. Though underexposed is still much worse than over-exposed ).

 

I'd love to get away from exposing this way too, but I don't see it happening because Leica is using a LUT / lossy compression for encoding 14 bits to 8...