I think that is entirely plausible. And so is 18>15, 15>13 and so on. linear resolution is what matters most, and 21>18 is an 8% difference. Rescaling will introduce more difference (by the very nature of "resampling methods") than the difference you're trying to equate - in differences as small as this.
Sorry Emil, I'm not ignoring you. But there's so many things to do and so many places to be.... :-) Quite a lot of LR's capabilities can be deduced "from the outside" by treating it as a "black box" problem and feed it varying DNG variables with the same raw. Some are very clearly hinted at in the DNG SDK as well.
DPP is harder, as the program structure is a lot more convoluted. You can feed synthetic CR2's into the system though, and learn a lot from it - if you're into interpolation algorithms, which I know you are.
paulfeng wrote:
Ok, so it's possible that in my self-induced sleep deprived state, I am making a silly mistake, but I don't believe this is the case. Each OOF color checker CR2 file was opened in DPP, and I just cranked the contrast and sharpness up to the max. Here are some results:
(Yes, these are viewed at 200% to make seeing the effect easier. It certainly is visible at 100% on the screen as well.)
(As noted above, I realized after taking the shot w/the second body that it had the 1.0.9 firmware, while the first body was running 1.1.0 firmware. So I did redo the second body shot w/1.1.0 firmware, but the results were essentially the same as seen here.)...Show more →
wow, that is interesting, both of my copies are worse than your first copy and your second copy actually looks like it is usable and that is the interesting thing there is a huge difference (at least in post demosaic output) from my first copy to your second copy, so large that one begins to doubt it really was entirely a design decision....
hmmm maybe it is down to QC then? at least in large part?
but wow this would be an unprecedented level of poor QC from Canon.
after all the other release disasters (1d3 af) it seems shocking they would get so sloppy so hmmm.....
there seem to be so few copies that manage to come in under the threshold where it starts wreaking too much havoc
so how many copies do I need to go through to get one like your second copy??
if everyone demands a good copy how many copies does canon need to chuck in the garbage?
EDIT: let me check it with ACR 5.5, maybe it will still show that there is quite a bit of mazing there, maybe it only just falls under the threshold for certain types of converters
EDIT EDIT:
well, with ACR 5.5 you can still see it even with just slightly aggressive but still reasonable real world sharpening although it is getting weaker now; if you sharpen it by extreme amounts it is very obvious (and for comaprison, extreme, extreme oversharpening of a 50D file still shows nothing in the way of this mazing noise)
maybe it is a mix of QC and design?
maybe they wanted to let a little more light through so they 'cheated' on one of the greens and that is why that green channel has less noise as it collect a little bit more light?
maybe if things are absolutely 100% to their design spec, a perfect body, it pretty much brings it under the cutoff for DPP and internal camera processing and smooth areas will pretty much stay smooth? and detailed areas will generally have just enough detail to swamp out the mazing from causing false detail
but since things are already a little unbalanced to begin with, perhaps even very tiny variations in QC can make it go from barely any mazing in DPP to producing a lot more artifacts? Maybe they didn't tighten the QC constraints any and left them at the same level as they had with previous bodies which doesn't work so well with this design?
I noticed that DPP tended to make images look smoother and smoother compared to ACR 5.5 the less detail there appeared to be in the area so perhaps it is using a bit of an adaptive algo as I had surmised.
Unless care is taken, it can actually mess up even detailed areas a bit, unless you WAY WAY oversharpen you may not notice it as artifacts, but if you do way way way oversharpen you can see that what appeared to be amazing and fine detail is really just traces of mazing (look at the ACR 5.6 sample and then process it with ACR 5.5 and DPP, 5.5's detail on some objects is fake and even with DPP although to a much lesser extent where with ACR 5.6 you can see where what actual detail there is to be had really goes).
since QC needs to be sooo tight and since even so, some algorithms still leave a lot more artifacts and since the gain in extra light would seem to be so relatively minor I really wonder how much sense it makes for them to have designed it this way, if indeed they did design it this way
with one of the better bodies and a slightly adaptive RAW converter I would guess you won't lose in any way compared to a 20D/40D/D300s, I wonder about compared to a 50D though at ISO100 and going for maximum reach? can you really get the same amount of TRUE detail while delivering no more noise? maybe (and hopefully) so?
Well, the simplest workaround is to do a local averaging of G1 and G2, perhaps after doing an overall rescaling based on column averages. Locally normalizing the column averages might in fact be sufficient if there are no spectral response imbalances of the greens, which there were in skibum's first copy, but don't seem to be in either of Paul Feng's.
brainiac wrote:
Marty DiBergi: Why don't you just make ten louder and make ten be the top number and make that a little louder?
Nigel Tufnel: [pause] These go to eleven.
ejmartin wrote:
Yes Keith, we heard you long ago...
Cute - my compliments to your script writer.
Two points for you to think about:
One, I didn't start this nonsense - you and people like you were making themselves heard long before I first chipped in;
Two, it apparently bears repeating, if only to stop certain people from endlessly regurgitating soundbites about mazing (to whit that it's universal and will happen regardless of converter, that only DPP can get anywhere near handling it, and that it does so by "smoothing" the files) that have been proven to be completely untrue...
You can see that there is a difference even in ACR between the two 7D copies.
With normal sharpening the difference is not as extreme between the two copies as it is with DPP (where it goes from bad to perfectly fine instead of bad to quite a bit less bad) but it is still there.
It is still of a worse character than the 50D and 5D2 but at least his second copy is within some striking distance in terms of the nature of the noise.
You can see just how much worse a poorer (typical) 7D copy is in the character of the image than a 50D though.
His second copy gives hope that you can get something reasonable if you are lucky or keep exchanging a few/many times, it might still fall a trace shy of a 50D for pure reach photography at ISO100/200 but at least it might be pretty close and there will be a lot less banding at high ISO and the AF and everything else should be a lot better, if you get a copy like his second one. Still not sure it is ideal, but maybe good enough to move on.
And here is a reduction showing that the poorer copies, even when downscaled to a low MP camera size still leave so many artifacts behind it would still probably do worse: http://skibum4.smugmug.com/photos/712933683_Zdsb8-L.jpg
while i don't think i like the design, it seem that in the end the QC differences seem to be FAR more crucial than nay design differences. If they were all like Paul's second copy the issue might not even have been raised, or at the least put aside as a modest compromise. That said even on his copy the raw RAW data are not of as nice a form as from my other Canon bodies.
my impression, which could EASILY be entirely wrong, is that a good 75% of the 7Ds out there so far look more like my copy than his second copy.
if ACR 5.6 tries to avg the columns a bit and does a trace of local adaptive averaging then with a copy like his second or better things should be pretty good i'd guess (likely better than any other APS-C DSLR at ISO800 and above and slightly worse to slightly better than a 50D at ISO100 and slightly to largely better than the rest at ISO400). I'd imagine that with a copy like mine though, even with ACR 5.6 it might still be a bit less than ideal performance.
anyway, the question is how hard is it to a copy like his second one currently? are most the latest batch like his second one? Paul, do you want to swap? I will even pay postage.
keithreeder wrote:
Cute - my compliments to your script writer.
Actually, I typed that myself
Two points for you to think about:
One, I didn't start this nonsense - you and people like you were making themselves heard long before I first chipped in;
No, you didn't start it; you do seem to be bent on dismissing what have become rather obvious QC issues, however. You're the one who should write a script --
Do[
start post fredmiranda.com
enter "there is no problem"
submit post
Until hell=frozen]
Two, it apparently bears repeating, if only to stop certain people from endlessly regurgitating soundbites about mazing (to whit that it's universal and will happen regardless of converter, that only DPP can get anywhere near handling it, and that it does so by "smoothing" the files) that have been proven to be completely untrue...
There are measurable and significant flaws in the RAW data from the 7D. No converter -- just direct analysis of the RAW files. That is undeniable (well, except for Canon apologists such as yourself) and has been verified by quite a few -- alundeb, theSuede, myself, John Sheehy, ...
As to what converters do about that, time will tell the extent to which the flaws can be worked around to generate acceptable output, and the degree to which the resolution potential of the camera will be compromised.
As I recall, the only time people were saying that DPP is the only option was when DPP was the only converter with finalized support for the camera. It does seem that DPP renders a softer image when sharpening is turned all the way down. It's hard to say whether that is a result of coping with the poor RAW data or something else, without deconstructing DPP.
skibum5 wrote:
while i don't think i like the design, it seem that in the end the QC differences seem to be FAR more crucial than nay design differences. If they were all like Paul's second copy the issue might not even have been raised, or at the least put aside as a modest compromise.
Skibum, thanks for the analysis.
It is starting to look like that this is indeed a QC issue.
I really hate playing the good-copy / bad-copy game .
something about this almost reminds me of the new quote from one of the heads of Canon R&D (regarding the new 1D4):
"We built new a AF sensor and algorithm – these are completely new, therefore we believe we have solved this issue. We don’t say our 1D Mark III had an ‘issue’."
(for the record the 1D3 is one of the canon bodies I have never touched and I have no comment about the AF good or bad as originally shipped or in its current incarnation)
For your usage I don't think you need any AF feature as long as there is no mazing I think you are pretty happy...
skibum5 wrote:
something about this almost reminds me of the new quote from one of the heads of Canon R&D (regarding the new 1D4):
"We built new a AF sensor and algorithm � these are completely new, therefore we believe we have solved this issue. We don�t say our 1D Mark III had an �issue�."
(for the record the 1D3 is one of the canon bodies I have never touched and I have no comment about the AF good or bad as originally shopped or in its current incarnation)
IMHO 1D III is probably has the best value today. The AF is overkill for normal day to day user. 7D is very good I'll pick it up one more time when the price is slightly lower.
:
.