Hey Sarangiman, welcome back. Way to dig up an OOOLD thread, man!
This was years ago...
Before I dive in...which "original post" of yours are you referring to? This is such a long thread, and it went through so many phases...
The point stands: +4 EV shifts are not unreasonable. It's the difference btwn ISO 100 and ISO 1600. If such pushes were unreasonable, why would Magic Lantern have implemented a dual ISO feature that people use to shoot ISO 100 and ISO 1600 simultaneously?
Well, first...I wasn't so much arguing against the use of 4 EV shifts "period". It was more an argument against Mikael's persistent use of particularly extreme "examples" to push is consistently extreme ideas about how uncompetitively bad Canon cameras and sensors are and (at least in his mind) would always be. Now that Mikael is gone, I think we can have a far more reasonable discussion on this subject. I fully agree that more DR is not a bad thing, and in many circumstances (sunsets being an ideal one) potentially extremely useful.
I'd offer that more accurate terminology could be used here, though. The difference between ISO 100 and ISO 1600 is four stops of LOST dynamic range. When you set ISO 1600, you lose
four stops of DR, or lose 16x the fineness of levels between your black point and white point, and end up with 16x more noise. I understand what your getting at, though, your thinking about what ML is doing with Dual ISO, however I still wouldn't say that's the difference between ISO 100 and ISO 1600. To be most accurate, it would really be the theoretical difference between ISO 100 and ISO 6.
(If you catch my drift).
Sony Exmor sensors still use analog amplification, BTW. When you set a higher ISO, the same fundamental rules apply to Exmor pixels as apply to Canon pixels. You have some charge level in your pixels, you amplify them BEFORE readout, then read (which at that point introduces the read noise). If you don't amplify before reading, even with the D800, noise levels when lifting an ISO 100 shot by four stops are higher than if you just shot at ISO 1600...the difference just isn't as extreme as when you do the same thing with the 5D III.
As for people who don't run into these problems much - YMMV based on the DR of the scenes you shoot. Posting an example of a scene you think has high DR, but where you didn't see offensive noise in shadows, is not a way to prove a camera has just as much DR as another. It's just a way of showing that it had enough DR for that particular scene.
Totally agree. Anecdotal claims are no substitute for proper controlled testing.
Which is why we have controlled tests, like the ones DXO perform.
There ARE problems with DXO's results, though. Their direct measures are fine, they are pretty much the only thing of value that DXO produces. The problem with DXO is that not all their "measures" are actually measured, many are derived and extrapolated from actual measures, and they are NOT entirely clear about their exact methodology or formulas. They HAVE been caught on a couple occasions changing their results after the fact when flaws in their methodology were found, without any explanation (at all, not even a blog post or article) of what was changed, how, or why, leaving those who identified or knew about the flaws left to wonder what kind of biased re-weighting or whatever was performed in the black box behind the scenes to "fix" the problems (assuming anything was actually fixed, and that data wasn't simply massaged). DXO has shown considerable bias with their lens reviews, on a very consietent basis, producing very arbitrary results and scores that show certain lenses to be equivalent when they are nothing of the sort, and conversely show other lenses to be inversely related to their actual status (i.e. the EF 50mm f/1.4 lens outscores the EF 600mm f/4 L II lens for the sole reason that it has a wider maximum aperture...results that COMPLETELY undermine the 600mm's vastly superior traits...that clearly indicates poorly balanced lens score weighting that heavily leans towards T-stops as being the most fundamentally important trait as far as DXO is concerned.)
The biggest single issue I think most of the people who complain about them (which does indeed include myself) have with DXO is their often outright and obvious bias, which is exhibited in their scores
. If DXO would completely eliminate, drop entirely, the whole notion of linearly scoring
the complexity of cameras and camera traits via a single scalar number, I think that would go a very LONG way to improving their reputation among those who hold them in fault. I think that scoring should be eliminated entirely, as I think it is a generally fallacious approach to take when the goal is to produce "scientific" results. When you throw in the fact that many of DXO's scores are weighted, and that some of those same scores are based on "measures" that are not
actually measured, they are extrapolated via simple and pure mathematical formulas (and therefor do not actually match reality, when compared with other REAL measures from other testers), DXO is quite simply not as scientific as they claim to be.
That's my core problem with DXO. I like some of their literal measures, ones taken directly from original data. An example would be their Screen DR measure...it's based directly on samples taken from RAW image files. That's a true, honest measure
. Print DR, on the other hand, is a purely mathematical EXTRAPOLATION, derived
from the Screen DR measure. They call Print DR a "measure", but it is nothing of the sort. Their simplistic and pure mathematic formulas are absolutely no substitute for actually measuring the change in noise levels after actually downsampling a full-sized RAW image to their standardized 8x12" "print" output. The only reason the D800 could possibly score some 14.4 stops of Print DR when downsampled is because the Print DR "measure" is a mathematically computed number, a computation that assumes 100% perfectly ideal circumstances. Reality cannot be ideal, so their formulas are missing aspects of real-world downsampling. If they did actually measure downsampled images, they would very likely produce a less significant difference between the 13.2 stops Screen DR and whatever Print DR ended up being if properly measured, and the differences would shrink as pixel size shrunk (fill factor in FSI sensor designs shrinks with an increase in pixel density, as a greater and greater percentage of sensor die area becomes dedicated to readout wiring and the like...even with reductions in fabrication process, fill factor still shrinks, and when you throw in extra logic onto the sensor die, which is usually the case with smaller processes, fill factor shrinks further).
Sorry to dive into the diatribe on DXO again, but the reasons why there are complaints about them are well-founded, and I think it's important to be clear that they aren't just "because DXO is biased against Canon". I don't really care what the brands are, that honestly doesn't matter to me. There are definable problems with the way DXO does things, because the way they do things allows bias, and bias is quite obvious in many of their scores. The day Canon comes out with a 16-bit ADC and scores 17.5 stops of Print DR on DXO, I guarantee you I'll be making the EXACT same argument AGAINST their Canon scores...because it will be the exact same deal as the D800 and it's 14.4 stops of DR with a 14-bit ADC.DXO's scores suck.
The way they score is very unscientific, regardless of whether it is self-consistent. They should just stick to the raw measurements, they should actually literally measure
downsampled versions of each image instead of just running a simple mathematical extrapolation, publish clear and concise facts about ALL of their testing methodologies for EVERYTHING (i.e. be completely open, get rid of the black box), and ditch scoring entirely.
They should rebuild their comparison engine to allow people to pick cameras, and let an openly defined algorithm compute comparative differences and highlight them in a meaningful manner. That would eliminate any option for bias, eliminate this whole inane notion of reducing complex products to single numbers, and potentially educate DXO viewers about the cameras they are comparing, without any DXO opinions sitting in the middle. Just share the raw facts.