5dMKIII.....taking the picture....Post Processing.....PRINT

[jrw]

Silly questions for the day.

1) When comparing camera noise levels why are both not shown at native resolution? Upsizing and downsizing is known to affect apparent noise. Even though it is nice to bring both to same size for easier direct detail capture comparison, why can't we also get the originals for both to assess whether resizing software has also changed the noise levels?

2) Much is made of the Nikon blacks and how they can be pushed up to such high levels. It is also known that Nikon clips the blacks in the camera. Is it possible to determine the levels in the blacks to see at what level they have been clipped in the Nikon shots and compare that to the Canon shots to see if that is one possible source of the noise? So many people go on about the need for the same treatment of the files, but if it is to be done should not the differences of in-camera processing be adjusted for as well?

Not trying to fan any flames here, just hoping to increase my understanding.

[Daniel Browning]

1) When comparing camera noise levels why are both not shown at native resolution?

Because that's not how the images are used in real life. If I was trying to decide between two cameras and I was comparing noise, I would print out images at the same size. Increasing magnification on just one of the cameras wouldn't provide an equal footing.

Upsizing and downsizing is known to affect apparent noise.

Quite right. While you should always apply resizing in order to get realistic results, different programs do resizing very differently. That's why I prefer to get the raw files and do the conversion and resizing myself, so I can see how they perform with my particular set of software and methods.

Even though it is nice to bring both to same size for easier direct detail capture comparison, why can't we also get the originals for both to assess whether resizing software has also changed the noise levels?

I don't know why. Usually when I post comparisons, I also post the raw files.

Is it possible to determine the levels in the blacks to see at what level they have been clipped in the Nikon shots and compare that to the Canon shots to see if that is one possible source of the noise?

Yes, it's possible to determine the black levels. It's always at or very near to the read noise floor. It's about the place where most Canon raw converters will clip blacks -- and in fact most of them do it before any other processing, so you get the exact same result as if it was done in camera. (A few converters, such as RawTherapee, don't clip it right away though, which is smart.)

And no, that is definitely not the source of the noise difference. Nikon clips the blacks way, way below the level at which Canon becomes unusable.

So many people go on about the need for the same treatment of the files, but if it is to be done should not the differences of in-camera processing be adjusted for as well?

Yes. That is always a good point to remember.

[Mark Elberson]

Wow! Sorry Dr. Cooper.

I never claimed exact science just tried to give some real world example to the folks that actually operate in the real world. You can tear it apart anyway you'd like. The 5DIII's images have shown to have more detail. I'm sorry it isn't the 20 stops of DR you are looking for.

I often wondered why this board is so slow these days.

I appreciate the real world sample as I'd be willing to wager that Aperture + ACR (Lightroom) makes up the vast majority of RAW processing.

[Daniel Browning]

I appreciate the real world sample as I'd be willing to wager that Aperture + ACR (Lightroom) makes up the vast majority of RAW processing.

Do you think the example I posted (http://www.fredmiranda.com/5DIII-D80...led-tests.html) is "real world"? It uses Lightroom. Here it is again:

5D3:


D800:


To me, that illustrates a pretty significant difference in dynamic range.

[HDNitehawk]

@Daniel; I went and read the review you linked, very informative. I think these samples demonstrate the D800's superior resolution more so than a superior dynamic range. Only because how they were taken and how they were cropped. I don't doubt that a test designed to demonstrate a wider dynamic range would show the D800 to be superior.

That said, I have this observation about the test. The tests were performed with a Zeiss 21mm Distagon, arguably one of the best landscape lenses made. You have to have the whole package to make the system superior, no doubt with the Zeiss attached you can do this. But I could only afford so many Zeiss lenses, and they are all manual focus. How would IQ comparisons hold up when you start attaching the best Nikor on the D800 compared to the best L lens on the 5D III?

For me, how I shoot I could swap the 5D II for the D800 and go with the Zeiss lenses. Most of what I do with the 5D II is landscape type work. I don't think I would want to give up my supertele's or my 70-200mm F/2.8, or my 1D IV for the Nikon equivalents though. The lenses are a huge part of this equation as well.

When I first started out, I choose Canon over Nikon because of the glass not the camera bodies.

[Daniel Browning]

You have to have the whole package to make the system superior, no doubt with the Zeiss attached you can do this.

Not really. Almost *any* lens will yield superior resolution on the D800. Exceptions are only going to be lenses that are exceptionally soft (we're talking Lens Baby type softness) and shot wide open and looking at corners. Stop down or look at the middle, and even the cheap-o lenses turn in better resolution than on the 5D2. You only need the best lenses if you want to get the maximum possible increase in contrast/resolution. For example, I shot the same $90 macro on both the 5D2 and D800 and noticed a huge jump in detail.

How would IQ comparisons hold up when you start attaching the best Nikor on the D800 compared to the best L lens on the 5D III?

Good question. Personally, I'd also like to see more comparisons of Nikon lenses on Canon bodies.

The lenses are a huge part of this equation as well.

Agreed.

[HDNitehawk]

Not really. Almost *any* lens will yield superior resolution on the D800. Exceptions are only going to be lenses that are exceptionally soft (we're talking Lens Baby type softness) and shot wide open and looking at corners. Stop down or look at the middle, and even the cheap-o lenses turn in better resolution than on the 5D2. You only need the best lenses if you want to get the maximum possible increase in contrast/resolution. For example, I shot the same $90 macro on both the 5D2 and D800 and noticed a huge jump in detail.
.

No doubt almost *any* lens would yield superior resolution, provided you could mount it to the Nikon. The problem is that Canon lenses are not going to mount up to the Nikon. In that case you have to look at what Nikon would have as an equivalent. In the three part review you linked, the author didn't seem to impressed with the Nikon lenses. You know the saying "glass first", should we be considering the glass over the body when picking a system?

Here is a possible solution to the camera body and sensor wars; buy a full set of Zeiss prime lenses with a Nikon mount. Buy an adapter for a Canon camera. That way in the future which ever company pulls ahead you don't have to change out all your lenses.

[Mikeinminn]

Daniel, you mentioned that the sensor in the D800 is a breakthrough in digital technology. Do you know what Nikon and Sony did to achieve this? I'm new to photography so bear with the silly questions.

About 5 years ago, I read an article that stated that resolution worked inversely to Dynamic range, moire, and noise. So with anti-aliasing filters there was no workaround because they were up against physics (without just making the sensor physically larger). They even made the claim that 15mp on an APS-C size sensor (50D) was seeing diminishing returns because of these artifacts.

Now we see the new D800 and perhaps even the Foveon technology in the Sigma allowing much higher resolution without the apparent defects. Some are even claiming results that are pushing MF results a bit.

Since I thought that these defects were a given, I am confused about what they did to achieve such outstanding results. Is the Nikon just doing internally what you mentioned as a workaround (overexposing the shot, then normalizing it later). Or is that impossible? Thanks.

[Daniel Browning]

Daniel, you mentioned that the sensor in the D800 is a breakthrough in digital technology.

I don't recall saying the D800 is a breakthrough, but if I did I was mistaken. The breakthrough in dynamic range started with the Sony A900 back in 2008. Nikon used it in the D3X, D7000, D5100, and now the D800; each time improving the performance.

Do you know what Nikon and Sony did to achieve this?

Basically, what happened is that Sony quit hanging on to CCD and decided to jump into CMOS with both feet. They were the kings of CCD, but never did very well with CMOS sensors. A few years ago, they decided they were going to become the best of the best in CMOS as well (you can see this in the shareholder presentations they did afterwords).

What's really amazing is that they actually did it. Now, top image sensor designers (e.g. Eric Fossum, who invented CMOS AP image sensors) have said that that Sony is pretty much the best right now.

To answer your question more specifically, I think the biggest big part of the improvement in dynamic range comes from the massively parallel read out. Sony uses 7,360 Analog-Digital Converters (ADC), one for every column of pixels. Canon only uses 8 ADC on the 5D Mark III. Therefore the Canon ADC has to run at an extremely high frequency, where it's more difficult to achieve a low-noise result. That's not the only factor, though, because Nikon-designed sensors like the D4 don't have column-parallel readout either, and are middling in performance between the D800 and 5D3.

About 5 years ago, I read an article that stated that resolution worked inversely to Dynamic range, moire, and noise.

Yeah, there's still a lot of that going around. It's mostly untrue.

They even made the claim that 15mp on an APS-C size sensor (50D) was seeing diminishing returns because of these artifacts.

It's true that returns are diminishing in some cases. Even a 6MP->8MP jump can see diminishing returns from motion blur. But even now that we are up to 24 MP APS-C, there are still plenty of people getting the maximum return.

Since I thought that these defects were a given, I am confused about what they did to achieve such outstanding results.

The defects are not a given. There is plenty of room for improvement, even from the D800.

Is the Nikon just doing internally what you mentioned as a workaround (overexposing the shot, then normalizing it later). Or is that impossible? Thanks.

Actually, that wasn't me that mentioned it. But no, they are not doing that internally. And it's not a workaround either. It's just throwing away highlights to improve shadows.

[Mikeinminn]

Sorry about the confusion about who wrote the article you linked. Very interesting stuff. Looks like Canon has some catch-up to do....except for in lenses where they appear to be doing pretty well.