The Uncertainties of Timothy "Ulcerating" Booth

[Guest sirvine]

Advertisement (gone after registration)

It's probably sacrelige, but I boost "fill light" and "recovery" in Lightroom when trying to extract what I want from B&W on M8 DNG. It's a good way to correct for shooting too far to the left, without artificial blowing highlights. It's far from ideal, but looks pretty good. An example: solsphere

[Advertisement]

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

[devils-advocate]

Jamie,

 

I agree with your observations, but I have experienced some 'file fragility' when converting underexposed shots to B&W when I rely very heavily on one colour-channel in order to get the contrast I want. Can't blame the camera for this at all.

 

Here's an example. First shot straight from camera, second processed only in LR - no PS. That's a LOT of detail coming out of burnt-out shite skies....and out of deep underexposed shadows, imho

 

- N.

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!

[timothy]

If you can't get tonal relations you like with digital, you aren't working hard enough. I say this because I discovered it was true of my own experience. It can take numerous RAW conversions, and a lot of blending and local area adjustment in PS to get tonal spead of an image to taste. It's hard, and a lot of work (kind of like dodging and burning in the wet darkroom........)

 

Nick, I agree with this statement.

 

My last comment in the "B&W from M8" thread emphasizes how much I agree with you (skim over the parts that are redundant):

 

Let's bring this full circle:

 

First, Hammam raises a very interesting problem: "As much as I like the M8 for color, I don't like digital for straight b&w." In other words, the results that Hammam has produced and seen others produce on this forum have not been convincing so far.

 

Hammam then says, "It always ends up looking like some sort of dull chromogenic b&w C41 film," concluding that it is the medium, the digital-ness of the M8, if you will, that is the reason for this problem.

 

Later, Charles challenges and, I think, successfully negates this conclusion by showing us the work of Alex Majoli, done with point and shoots no less. Those pictures are far from appearing like "dull chromogenic b&w C41 film."

 

Charles offers his own conclusion that M8 photographers that want to do convincing black and white work will have to "work" their pictures.

 

My own contribution to this discussion is point out that black and white tends to be more about interpretation than description. Accordingly, "processing" and "converting" and making global adjustments is insufficient to produce convincing black and white. Convincing black and white has to have life breathed into it. It has to be crafted. Something has to be envisioned, and then post production tools have to be employed to achieve that vision.

 

My suggestion is that we talk about post-production approaches toward achieving vision with the view that "the computer is the darkroom" as Charles states forthrightly:

 

"I think one thing that's forgotten to a degree here is that the computer is now your darkroom. And if you were never that great in the darkroom...... To me it's all about dodging and burning and making an image come alive as one visualised it in the first place."

 

This is key:

 

"Doesn't matter if it's got the resolution of an 8x10 view camera (or the grain of tmax 3200) if it's not "printed" with some degree of sophistication and understanding it won't draw the viewer fully into the experience."

[timothy]

See--you mess with film a lot to get the great midtones and subtle highlights and depth of shadows.

 

If you expect to get this kind of quality out of (any) digicam without post processing, your just fooling yourself. You won't do it.

 

Jamie, I, personally, am not expecting great midtones and subtle highlights and depth of shadows without post processing. In other words, I'm expecting to do a hell lot of post-processing. As it is, I do a hell lot of post-processing of scanned negatives. The M8 RAW files and converted RAW files will hold up to a hell lot of post-processing, eh?

[Jamie Roberts]

With an M8, the situation is actually more extreme than the LL article would suggest. Reason is that the M8's compression to 8-bits scheme compresses lows a lot more than highlights. To use the LL article's terminology, for an M8:

 

In the first f-stop: 74 levels

second f-stop: 53

third f-stop: 37

fourth f-stop: 26

fifth f-stop: 18

sixth f-stop: 13

seventh f-stop: 9

eighth f-stop: 6

 

But this is probably not really worth more than 1/3 of a f-stop

 

Sandy

 

Sandy--no, I believe you're completely wrong about this. The M8's system is a 14bpp AD compressed--not downsampled--to 8 bits.

 

IOW, the file is stored as 8bit then uncompressed to 14bit. An order of magnitude better than 12bpp cameras.

 

There is a big difference, especially in the shadows, where you still have many thousands of levels even with underexposure. In fact, the M8 compresses highlights more than shadows, IIRC.

 

This is more dynamic range, folks.

 

As a consequence, my M8 has 2 stops easy on my 5d, and even more than the 1ds2.

[Jamie Roberts]

@ Nick--

 

Yes--look at what you pulled out of the shadows! If I may suggest, try underexposing (preserving more highlights) at low ISOs (I know--heresy!) then pulling out the resulting shadow detail.

 

I think you'll be completely pleasantly surprised. I was frankly astounded the first time I did this.

 

And Nick--I don't know if this makes sense for your workflow, but you might be able to get a better RGB file to blend or select with if you make the initial (slight) contrast adjustments in LAB first in the Luminance channel.

 

@ Tim--given you maintain highlights you can push and pull these files all over the place. It's quite amazing, and it's in the shadows where this thing really sings.

 

I'm sure the 14bpp Canon 1d3 will be the same, and then everyone will agree with me

 

As for highlights, you've got about a half stop of play with the DNG, instead of a brick wall you might be used to, and when the highlights DO go, they go without the bandy mess you get in 12bpp cameras.

 

Hope that helps. ISOs up to 1250 can be really manipulated; 160-640 is best though for noise. Because of the gain, I always try to expose to the right with ISO 2500 if shadow detail is what I want.

[timothy]

Advertisement (gone after registration)

Here's an example. First shot straight from camera, second processed only in LR - no PS. That's a LOT of detail coming out of burnt-out shite skies....and out of deep underexposed shadows, imho

 

- N.

 

N.,

 

That's awesome recovery of details in the sky! Personally, I would like to see, if you were taking the picture all over again, what kind of head room you would have given less exposure. I bet you could get even more definition from the clouds and give them a more three dimensional look without so much bland white area. But let that not detract from doing a great job on the post-processing. Congrats!

 

Timothy

 

P.S.

I couldn't resist playing with the JPEG you posted for us. It posterized, but what the hey.

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!

[timothy]

Thanks Jamie,

 

Your comments really give me a good perspective on all this.

 

I'm 90% decided on getting the M8. Now the question will be how to get one. I'm in Winnipeg, Manitoba, Canada, where there are no M8 (or any Leicas) to be found locally. I'm going to be researching whether to order from Toronto or the States.

 

Timothy

[sandymc]

Sandy--no, I believe you're completely wrong about this. The M8's system is a 14bpp AD compressed--not downsampled--to 8 bits.

 

IOW, the file is stored as 8bit then uncompressed to 14bit. An order of magnitude better than 12bpp cameras.

 

There is a big difference, especially in the shadows, where you still have many thousands of levels even with underexposure. In fact, the M8 compresses highlights more than shadows, IIRC.

 

This is more dynamic range, folks.

 

As a consequence, my M8 has 2 stops easy on my 5d, and even more than the 1ds2.

 

Jamie,

 

Sorry, but that's straight out of the look-up table in the DNG. This is not to say that there might not be more dynamic range in the M8 DNGs than in other cameras - dynamic range is different to number of levels - but that really is the number of levels.....

 

Sandy

[devils-advocate]

Thanks Jamie,

 

Your comments really give me a good perspective on all this.

 

I'm 90% decided on getting the M8. Now the question will be how to get one. I'm in Winnipeg, Manitoba, Canada, where there are no M8 (or any Leicas) to be found locally. I'm going to be researching whether to order from Toronto or the States.

 

Timothy

 

Tim,

 

Call the Camerastore in Calgary. Ask for Julian and tell him I sent you Best shop in Canada, and no PST!

 

- Nick

[Jamie Roberts]

Jamie,

 

Sorry, but that's straight out of the look-up table in the DNG. This is not to say that there might not be more dynamic range in the M8 DNGs than in other cameras - dynamic range is different to number of levels - but that really is the number of levels.....

 

Sandy

 

No, I believe you are incorrect on a number of issues here.

 

First, you're looking at a compressed DNG, not a downsampled one. The DNG has to be uncompressed according to an algorithm that preserves the extra levels in the shadows.

 

Looking at the compressed levels in the DNG as what the camera delivers, if I understand this correctly, is something like looking at a PKZIP file and concluding there are no levels of redundancy

 

This also explains why the M8 DNGs didn't just open in major RAW converters until the compression algorithm was implemented.

 

If you search the forum, you'll see very good explanations of this.

 

If I'm wrong about this, then there is something much weirder with the actual number of levels the sensor produces.

 

BTW--regardless of the above--the number of levels the sensor can produce at the AD stage is the DR, it's precisely about the gradations of tones that can be reproduced from darkest to lightest without hitting the noise floor or the highlight burn out.

 

This is the same in printing, too, but we don't think of noise so much as ink blockage with no detail. dMax is DR, and so is bit depth.

 

This is why Photoshop calls 32bpp high dynamic range images.

 

@ Timothy--I bought mine in Toronto at 8 Elm; they're a great Leica dealer too. Looks like Nick's is the one to go with, though!

[sandymc]

Jamie,

 

There is only one kind of M8 DNG, and its 8-bit level compressed. Decompression is by the look-up table that is in the DNG, and that's exactly where the number of levels data came from. Agree that that it is decompressed to 14 bits, but that 14 bits still only has 256 levels in it. Happy if you can point me at somewhere reputable that says different (or if you can point me at any M8 DNG that is different to 10.3 M in size - if there's more than 8 bits of data per pixel, they would need to be bigger) , but DNG files are encoded per the Adobe standard and M8 files are very clearly have 256 levels.....

 

Sandy

[devils-advocate]

The science of this is beyond my ken, but for what it's worth, guys like Bill Atkinson, who do understand this intimately, are ardent about exposing-to-the-right on their Phaseone P45 backs, which are true 16-bit capture devices.

 

In simple terms, every stop of decrease in exposure brings a 50% decrease in the number of discrete, differentiable levels. The actual number of levels is irrelevant to the truth of this phenomenon.

 

Because you can 'place' that data anywhere on the tonal scale in the processesed image, a "lump" of date on the right hand side of your histogram litterally spreads of further across the left side of your out-put histogram. This creates subtler and wider differentiations of tone, making a more pleasing picture. I don't think there's any escaping the physics of this one.

 

- N.

[Jamie Roberts]

Jamie,

 

There is only one kind of M8 DNG, and its 8-bit level compressed. Decompression is by the look-up table that is in the DNG, and that's exactly where the number of levels data came from. Agree that that it is decompressed to 14 bits, but that 14 bits still only has 256 levels in it. Happy if you can point me at somewhere reputable that says different (or if you can point me at any M8 DNG that is different to 10.3 M in size - if there's more than 8 bits of data per pixel, they would need to be bigger) , but DNG files are encoded per the Adobe standard and M8 files are very clearly have 256 levels.....

 

Sandy

 

I will try to find the posts here--of course, someone else could help me out here!.

 

Anyway, while the encoding was within the Adobe DNG standard (which has a lot of room for other things) I don't actually think the encoding was the norm.

 

Most DNG RAW programs would simply not open the M8 files, even if they would open the DMRs non-compressed ones.

 

Also, IIRC, the size has nothing to do with it; they are using lossy compression (the highlights take the hit, IIRC) to maintain a smaller (therefore faster to work with) file.

[Jamie Roberts]

The science of this is beyond my ken, but for what it's worth, guys like Bill Atkinson, who do understand this intimately, are ardent about exposing-to-the-right on their Phaseone P45 backs, which are true 16-bit capture devices.

 

In simple terms, every stop of decrease in exposure brings a 50% decrease in the number of discrete, differentiable levels. The actual number of levels is irrelevant to the truth of this phenomenon.

 

Because you can 'place' that data anywhere on the tonal scale in the processesed image, a "lump" of date on the right hand side of your histogram litterally spreads of further across the left side of your out-put histogram. This creates subtler and wider differentiations of tone, making a more pleasing picture. I don't think there's any escaping the physics of this one.

 

- N.

 

In absolute terms of course, this is correct--you want to maintain as much shadow detail as possible,

 

But coming from digicams that have only 12bpp sampling means you have much more latitude by comparison.

 

And this accords with my experience of the actual camera in comparison to others. That it's not as good as a Leaf exposed to the right doesn't surprise me at all

 

I'm going to ask Leica when I see them at the WPPI conference next week. I'll see if I can get a definitive statement on compression and sampling with the M8.

 

The specs as printed still say "16bpp" but I understand that it's actually closer to 14bpp (again, with lossy compression and the hit in the highlights).

 

As for the physics, James Nisly, a photographer and digital expert whose digital work I quite admire, explained the 12bpp vs 14bpp difference in levels as follows:

 

Quote:

 

"In order to measure those small differences in light, you need two things:

 

1. A very, very clean signal from the sensor so that those small differences in light arn't masked by random noise.

 

2. An A/D and digital pipeline that has the bit depth to measure those small linear signal changes comming from the sensor.

 

Now, being that the photosites are linear, and stops are not, we need to map one stop incriments to changes in value at the A/D. Each stop would be a halving of the measured value. In a 12-bit system, it would look like this.

 

111111111111b = 4095d = the limit of the sensor. Full saturation.

011111111111b = 2047d = one stop down

001111111111b = 1023d = two stops down

000111111111b = 511d = three stops down

000011111111b = 255d = four stops down

000001111111b = 127d = five stops down

000000111111b = 64d = six stops down

000000011111b = 32d = seven stops down

000000001111b = 16d = eight stops down

 

Now, we have to make a decision. Is four possible values enough to make the last stop usable? Probably not, so I'll stop there.

 

In a 14 bit system, it would look like this.

 

11111111111111b = 16382d = Full saturation

01111111111111b = 8191d = one stop down

00111111111111b = 4095d = two stops down

00011111111111b = 2047d = three stops down

00001111111111b = 1023d = four stops down

00000111111111b = 511d = five stops down

00000011111111b = 255d = six stops down

00000001111111b = 127d = seven stops down

00000000111111b = 64d = eight stops down

00000000011111b = 32d = nine stops down

00000000001111b = 16d = ten stops down

 

Now we're back to where we were with four values left for the last step, but now we are ten stops below the saturation point, instead of eight stops.

 

Also notice you have four times as many values between stops two and three in the 14-bit case as you do in the 12-bit case. This is how you get better tonal range and smoother gradations out of a high dynamic range system. Again, the two go hand in hand, and don't really need to be seperated."

 

End of Quote

 

This is precisely what I see with the M8--an extra two stops of dynamic range in the shadows at low ISOs, and exceptionally smoothy gradations to blown highlights (and an amazing amount of recovery there).

 

At high ISOs, you hit the noise floor sooner, and so even though the extra levels are there, you can't use them (same with the DMR over ISO 800)

[sandymc]

Jamie,

 

Yes, I'd agree with you, at least in theory, on the dynamic range. The LL article suggests that most DSLRs have 6 stops of dynamic range, so anything below the 6th f-stop of my table would be noise. It's possible that what you (and others) are seeing in your images is that the M8 can go down to eight f-stops, and even though the eighth f-stop only has 6 levels, that's still more f-stops than other cameras. So the M8 could have more dynamic range even though it has fewer levels than e.g., the D200. This where dynamic range and number of levels are different, as I mentioned in a post above. I would be interested to see any actual data comparing the dynamic range of say a D200 vs the dynamic range of an M8. I've not seen that published anywhere, which is why I say I agree in theory...

 

Sandy

[Jamie Roberts]

Sandy--

 

Again, I think we're arguing over definitions.

 

As I understand it, the 16bpp (practically 14bpp) sensor on the M8--by definition--has more dynamic range than a 12bpp sensor.

 

Whether it can store, or use those bits due to noise limit, is another question, and I think that's what you're saying.

 

As I said, I'll check with Leica on the official say on this.

 

Even *if* the resulting DNG is only 256 levels (something I don't think is actually accurate), given equal noise, you're still starting with an order of magnitude more information from the M8 sensor.

 

About two stops worth, IMO and experience (from the 5d, which is saying something).

 

IOW, and by analogy, we all know a better 16bpp TIFF makes a better JPEG, when all is said and done.

 

So this isn't just "in theory", and anyone thinking about this for long has to account for the remarkable M8 shadow quality and overall tonality in low ISOs. I don't think it's improved noise, actually.

 

The only other 35mm digicam I've seen anything like this is the DMR, which is a 16/14bpp uncompressed DNG.

[Guest stnami]

The other thing I have found is that LR will tend to give a grainier rendition across all bw, as compared to Raw Developer which has a much more advanced noise reduction and sharpening algorithm. I bought RD just for BW alone, there are some things you can do it it that are impossible in other developers.

 

Robert

LR is pretty nifty in thet you can import scanned film convert it into a DNG and open up in Raw Developer for some B&W conversion

[Advertisement]

Hallo,

Look here: Interesting Leica gear