Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: Panopeeper on March 09, 2009, 04:49:10 pm
-
I saw some surprizing results of comparisons between the Sony A900 and other cameras, like the A900's DR being a full stop greater than the D3X' and the 5DMkII's.
I set out to measure the noise and DR myself, and I soon ran into a problem: the measurement results were iconsistent, in some cases plainly contradicting each other; for example the noise measured on the raw channels separatedly is different between the channels.
It appeared that I needed better, i.e. for measurement more suitable raw files. Erik Kaffehr, a fellow poster here and a A900 owner volunteered to create the images; after some fine tuning, he delivered excellent bases for measurements.
(Btw, Erik's contribution was much greater than it is usual when I am asking someone and receiving specific raw files. In the span of several weeks he shot again and again, all together about 100 images. Those, who appretiate this report should say or think a big "thanks" to him.)
Then the surprize came: the measurement results were strange even with the best images. I had to dig deeper in the analysis (thank to Rawnalyze, this was not a problem). Some of the findings were astonishing.
I will create a much bigger essay with painfully pixel-peeping documentation of the background (I created hundreds of screen captures for this purpose), but that is a tiresome business, it will take some more time, and anyway many photographers are allergic to proofs, which are too factual; this is particularly true on these forums.
Here are the findings in condensed form:
1. Noise reduction Off, Low, Normal and High are identical from ISO 100 to 800, ON THE RAW DATA. They correspond something like NR Low @ ISO 1600.
In other words: all reviews used pre-NRed samples.
However, NR Off means almost Off with ISO 1600 (this explains the "sudden" increase of noisiness at ISO 1600).
2. The noise reduction affects mainly the red and the blue channels, much less the green. This is a very primitive noise reduction; it simply eliminates some pixel levels in the affected areas.
3. Due to the nature of this noise reduction, the very dark, noise reduced areas become darker. The consequence is, that not only the visual appearance but the noise measurement too indicates not only lower noise that it would be without NR, but that in deeper shadow, suggesting a greater dynamic range.
The magnitude of this shift is about 0.6-0.7 EV between the red/blue and the green channel. As the green channel itself is not virgin either, the shift must be even greater.
I estimate that in end effect the noise gets shifted and the DR "enhanced" by at least one full stop.
A side effect of the above is the non-linearity of the pixel values, and a WB shift in the low ranges: if the WB is correct in the "normal" ranges, it is certainly incorrect in the very dark range.
The nature of this noise reduction explaines the blotchiness as well. This will be appearant from the coming demonstrations.
Anyway, after dozens of hours squeezing these images I still can not put reliable numbers on the noise of the A900. However, I can say, that the noise is considerable higher and the DR is considerable lower than some reviews have suggested.
-
Anyway, after dozens of hours squeezing these images I still can not put reliable numbers on the noise of the A900. However, I can say, that the noise is considerable higher and the DR is considerable lower than some reviews have suggested.
It's a choice. You can buy the A900 for $3000, get the highest resolution possible in a current 24 megapixel DSLR and use post-processing noise reduction, or, you can spend $8000 for the D3x, have no choice in whether noise reduction is applied because it is in-camera and you can't turn it off. Sony cameras, since their inception have had the highest noise but also in their class, the highest resolution. That goes back to the A100.
-
Gabor,
Thanks for sharing information! Are you aware of any forum where users are more open to proof?
Erik
I saw some surprizing results of comparisons between the Sony A900 and other cameras, like the A900's DR being a full stop greater than the D3X' and the 5DMkII's.
I set out to measure the noise and DR myself, and I soon ran into a problem: the measurement results were iconsistent, in some cases plainly contradicting each other; for example the noise measured on the raw channels separatedly is different between the channels.
It appeared that I needed better, i.e. for measurement more suitable raw files. Erik Kaffehr, a fellow poster here and a A900 owner volunteered to create the images; after some fine tuning, he delivered excellent bases for measurements.
(Btw, Erik's contribution was much greater than it is usual when I am asking someone and receiving specific raw files. In the span of several weeks he shot again and again, all together about 100 images. Those, who appretiate this report should say or think a big "thanks" to him.)
Then the surprize came: the measurement results were strange even with the best images. I had to dig deeper in the analysis (thank to Rawnalyze, this was not a problem). Some of the findings were astonishing.
I will create a much bigger essay with painfully pixel-peeping documentation of the background (I created hundreds of screen captures for this purpose), but that is a tiresome business, it will take some more time, and anyway many photographers are allergic to proofs, which are too factual; this is particularly true on these forums.
Here are the findings in condensed form:
1. Noise reduction Off, Low, Normal and High are identical from ISO 100 to 800, ON THE RAW DATA. They correspond something like NR Low @ ISO 1600.
In other words: all reviews used pre-NRed samples.
However, NR Off means almost Off with ISO 1600 (this explains the "sudden" increase of noisiness at ISO 1600).
2. The noise reduction affects mainly the red and the blue channels, much less the green. This is a very primitive noise reduction; it simply eliminates some pixel levels in the affected areas.
3. Due to the nature of this noise reduction, the very dark, noise reduced areas become darker. The consequence is, that not only the visual appearance but the noise measurement too indicates not only lower noise that it would be without NR, but that in deeper shadow, suggesting a greater dynamic range.
The magnitude of this shift is about 0.6-0.7 EV between the red/blue and the green channel. As the green channel itself is not virgin either, the shift must be even greater.
I estimate that in end effect the noise gets shifted and the DR "enhanced" by at least one full stop.
A side effect of the above is the non-linearity of the pixel values, and a WB shift in the low ranges: if the WB is correct in the "normal" ranges, it is certainly incorrect in the very dark range.
The nature of this noise reduction explaines the blotchiness as well. This will be appearant from the coming demonstrations.
Anyway, after dozens of hours squeezing these images I still can not put reliable numbers on the noise of the A900. However, I can say, that the noise is considerable higher and the DR is considerable lower than some reviews have suggested.
-
you can spend $8000 for the D3x, have no choice in whether noise reduction is applied because it is in-camera and you can't turn it off
I'm afraid you are mixing up some issues. Although the raw data of Nikon cameras is not as raw as the Canon raw, and the difference contributes to a lower apparent noise, that is not comparable to the noise reduction carried out by the A900 on the raw data, and even less to the costomary noise reduction.
-
Are you aware of any forum where users are more open to proof?
I am working on that anyway. I guess there are more posters on DPReview, who are not afraid of pixel peeping and are interested on such details.
-
I am working on that anyway. I guess there are more posters on DPReview, who are not afraid of pixel peeping and are interested on such details.
May I HIGHLY suggest www.dyxum.com. It's a Sony/Minolta site, but the people there are wildly open-minded and exceedingly friendly. The founder of the site has repeatedly knocked the A900 for this kind of behavior and is a D700 shooter. Seriously, I know they'd appreciate it. (not to mention a LOT of them have extensive sensor and software knowledge).
DpReview is okay.... I guess, but there are a lot of spaz/troll posters.
-
Gabor,
Thks for the info and extended tests.
By the way, I had forgotten about that, but do you still need d3x raw files?
Cheers,
Bernard
-
May I HIGHLY suggest www.dyxum.com. It's a Sony/Minolta site, but the people there are wildly open-minded and exceedingly friendly
All right, I'll give it a try; let's see how they digest the news :-)
I am not particularly interested on the Sony issues, but I am interested on all raw processing related issues. I am certainly not interested on bashing or pushing any camera or system, as I have no allegiance of any kind. My present work is a critique, not of the camera but of the professinal evaluations.
-
I had forgotten about that, but do you still need d3x raw files?
Bernard, my collection of raw files contains about 4000 files (beside those tens of thousands of my own cameras, of course) - and this is still not enough to answer some questions.
I don't have *any* raw files, which are suitable for reliable measurements of the D3X noise characteristics. I do need black frames (this is the cheepo request), and special shots for measuring the noise.
I need to say at this point, that I do not think there is any big surprize in the published findings re the D3X, like re the A900. However, if I get suitable raw files, I can document my findings, i.e. publish the raw files and the evaluations. I see this as an "open review", others are free not only to criticize but to counter-evaluate the findings.
-
I am not particularly interested on the Sony issues, but I am interested on all raw processing related issues. I am certainly not interested on bashing or pushing any camera or system, as I have no allegiance of any kind. My present work is a critique, not of the camera but of the professinal evaluations.
I agree with you on the importance of suitable camera evaluations and double checks never hurt.
Now, the fact that Sony does raw level noise reduction is not by itself a problem, and DxOMark measuring this is also not a problem. What the problem is is the fact that the impact this raw level noise reduction has on micro detail is de facto not taken into account in DxOMark results. So Sony ends up getting to good mark on DR, but not the blame on poor micro detail (or at least poorer than it could have been).
From a user point of view cleaner shadows with a bit less detail might not be a problem in itself, but they should be aware that this choice is being made for them by Sony.
Cheers,
Bernard
-
This is very interesting to me. I for one have been flabbergasted by the DR range of the A900. I can't imagine it being "faked" too much. I totally lack the requisite knowledge to be very helpful in this thread, all I can say is that these two shots the convinced me of the astounding DR to be had:
(http://farm4.static.flickr.com/3657/3325308602_9120ba5554.jpg) (http://www.flickr.com/photos/lattiboy/3325308602/)
and
(http://farm4.static.flickr.com/3655/3339445955_8c0d726070.jpg) (http://www.flickr.com/photos/lattiboy/3339445955/)
Click for larger version.
-
From a user point of view cleaner shadows with a bit less detail might not be a problem in itself, but they should be aware that this choice is being made for them by Sony
From the point of this user, i.e. me, the best is to leave the decision to my discretion. I hate patronization, I can't stand when an asshole decides what the best is for me - even if that costs $5000 less tha if two assholes together would make the decision. The sad fact is, that *all* Japanese cameras show the lack of intellect in the design, and attitudes like this of Sony show, how bad this can get.
-
all I can say is that these two shots the convinced me of the astounding DR to be had
While I find these shots very good, I think it is important to note, that your (or mine or anyone else's) judgement on the dynamic range of a scenery is of secondary relevance compared to the actual measurement. Normally I would ask you to please post the raw files and then we talk about the facts - but the issue is, that I could not tell the facts in case of the A900!
Anyway, my experience taught me not to rely on such subjective "DR measurement" (I for myself ceased to make such judgements). But if you want to get into a contest: my dog looks much better at ISO 800 under a 40W lamp in the kitchen than your cat under your reading lamp (I mean this is often the level of proof presented on forums).
-
Good point! I think I understand your original post now. How do you explain the fact that users seem to have similar experiences to me, even coming from the A700 w/ v4 firmware (which I thought had no noise control at all)? An honest question.....
PS I've noticed the A900 data has flummoxed a few different people when the RAW files are analyzed. I believe David Kilpatrick of PhotoclubAlpha was astounded that it produces images as well as it does because, "the red channel is a mess!!!".
-
Thanks, Gabor, for all your work and very interesting information. I enjoy reading it, though I'm really more into subjective impressions (in prints) than scientific measurements. And on that subject, I've just finished processing (with a lot of winnowing!) about 3000 a900 frames from a recent trip. What struck me most was how few shots, compared to my former equipment of another famous manufacturer, needed any highlight or shadow recovery in C1.
Please keep up the good work!
Bill
-
From the point of this user, i.e. me, the best is to leave the decision to my discretion. I hate patronization, I can't stand when an asshole decides what the best is for me - even if that costs $5000 less tha if two assholes together would make the decision. The sad fact is, that *all* Japanese cameras show the lack of intellect in the design, and attitudes like this of Sony show, how bad this can get.
Hum... aren't you a bit over-reacting here? How many % of image quality are we talking about?
Many folks around here have been super happy about the A900 lower ISOs, even people having worked with both the A900 and D3x.
Doesn't it show that the compromise chosen by Sony is pretty good, even if not as close to best in class as some like to think?
Cheers,
Bernard
-
aren't you a bit over-reacting here? How many % of image quality are we talking about?
Partonization is a mild form of terror; one can not over-react to it.
As to the % of image quality: do you mean that one should rescind a percentage of the hardware capabilities for .. for... WHAT in exchange? Perhaps it is an intentional degradation in exchange for the low price? Perhaps an A910 will be released for $200 more, the same camera but without forced NR?
Many folks around here have been super happy about the A900 lower ISOs, even people having worked with both the A900 and D3x
They should read carefully what I wrote in the initial post.
Doesn't it show that the compromise chosen by Sony is pretty good, even if not as close to best in class as some like to think?
Now, this is the point. A compromise between WHAT? A compromise presumes two or more sides of an issue; something for something. What exactly is the advantage of doing a premature noise reduction without my asking, when doing that later, at my discretion is *at least* as good? What other consideration is here beside image quality? Do Sony collect the stolen pixel levels and sell them someone?
The sensor has certain capabilities, which can NOT be enhanced, only degraded the way Sony chose to go. I am not complaining about the sensor not being good enough, but about the firmware intervening (trespassing) in the process of developing the image. What is happening is cheeting with the appearance on the cost of real quality.
There is absolutely no compromise here, only sacrifice, due to some misguided decision, just like with the A700. I wonder if Sony will do this with every new model, waiting until some customer groups are yelling loud enough.
-
Well, yes, indeed...
My experience is that I see clearly these effects at high ISO and/or specific test targets. I never shoot really high ISO and if I do it's not in really bad light. My experience is that a lot of shadow detail can be retrieved from the KM 7D, the Alpha 700 and Alpha 900 but less so from the Alpha 100. I cannot compare with other cameras.
One of the issues I have that there is a Sony blotchiness which is very hard to eliminate. I have never seen it in other than test images, but I do these tests to find out the limitations and avoid trying to go beyond what the camera can do.
Best regards
Erik
Hum... aren't you a bit over-reacting here? How many % of image quality are we talking about?
Many folks around here have been super happy about the A900 lower ISOs, even people having worked with both the A900 and D3x.
Doesn't it show that the compromise chosen by Sony is pretty good, even if not as close to best in class as some like to think?
Cheers,
Bernard
-
Following captures do NOT demonstrate the unwanted NR at lower ISOs; these are NR Off, Low, Normal and High with ISO 1600. These show, how brutal evel the lowest level of NR is, performed this way, shown on the red channel on Erik's coffee bucket.
-
Hi,
This test image was intended to be brutal. The issue that Gabor (Panopeeper) illustrates so well is that the image with NR "off" can probably handled quite well by a late version DPP (a very good but quite different raw-converter) or by noise Ninja. The image with the high NR is beyond rescue.
Best regards
Erik
Following captures do NOT demonstrate the unwanted NR at lower ISOs; these are NR Off, Low, Normal and High with ISO 1600. These show, how brutal evel the lowest level of NR is, performed this way, shown on the red channel on Erik's coffee bucket.
-
Regarding the blotches: this requires a bit pixel peeping. The attached captures show the blue channel from the coffee cup, one pixel level at a time. It starts with 129, for everything lower becomes null. Note: this is ISO 800, NR Off, i.e. this is the forced NR. These groups of pixels in the same level are characteristic to the forced NR of the A700 and A900, they have nothing to do with the texture of the cup, they appear everywhere the same way.
The last capture shots the histogram on the side of the cup; notwworthy is the difference between the red/blue and the green: the red and blue are much more "shrunk" around 128, thereby "compressing" the levels, making the noise lower and the appearance darker.
The same is happening in the red channel, but the green is much different, even though it does not remain untouched. This is not restricted to the levels so close to 128, the same happens somewhat higher as well (the pixel values of the ninth and higher stops are between 128 and 143).
(Let's try to get back the attachements)
-
Partonization is a mild form of terror; one can not over-react to it.
You are over-reacting...
As to the % of image quality: do you mean that one should rescind a percentage of the hardware capabilities for .. for... WHAT in exchange? Perhaps it is an intentional degradation in exchange for the low price? Perhaps an A910 will be released for $200 more, the same camera but without forced NR?
Euh...
- ease of use through more streamlined workflows,
- better DxOMark results that make the A900 puchasers feel good about their investement
- less apparent noise in low ISO shadows that will prevent people from over-exposing, therefore blowing highlights
Now, this is the point. A compromise between WHAT? A compromise presumes two or more sides of an issue; something for something. What exactly is the advantage of doing a premature noise reduction without my asking, when doing that later, at my discretion is *at least* as good? What other consideration is here beside image quality? Do Sony collect the stolen pixel levels and sell them someone?
The sensor has certain capabilities, which can NOT be enhanced, only degraded the way Sony chose to go. I am not complaining about the sensor not being good enough, but about the firmware intervening (trespassing) in the process of developing the image. What is happening is cheeting with the appearance on the cost of real quality.
There is absolutely no compromise here, only sacrifice, due to some misguided decision, just like with the A700. I wonder if Sony will do this with every new model, waiting until some customer groups are yelling loud enough.
Everything is relative Gabor. You are only dissatisfied with the shadows of the A900 because you have found that they could have been closer to your taste (meaning noisy but more detailed). All those users who don't know this are just happy about having clean shadows... See above for some ideas on the value of clean shadows.
Don't get me wrong, I am with you here, noise reduction in camera is not something I find to be a good idea, but I am just trying to put things in perspective.
Cheers,
Bernard
-
without getting too technical, how are you getting the raw data to compare the true DR to the augmented DR for the a900 ?
also, if the a900 makes too many assumptions in their processing algorithms, why not just hack their firmware? Naturally this would be a real pain in the ass but if you connect with a good enough community of developers (using git or something of that nature), you could make something to accomplish zero change in the raw data, no?
I'm surprised that nobody hacks camera firmware so far that I've seen... It happens with so many other devices, why not cameras?
I'm not a firmware hacker for any device, but I have a sufficiently deep knowledge of CS to begin to pick it up.
EDIT: a google came with this;
http://chdk.wikia.com/wiki/CHDK (http://chdk.wikia.com/wiki/CHDK) - Canon hacking wiki
http://www.camerahacker.com/ (http://www.camerahacker.com/) - enthusiast site
-
without getting too technical, how are you getting the raw data to compare the true DR to the augmented DR for the a900 ?
I am not, for I can not. However, if Erik manages to make shots with his A900 and with a Canon at the same time, then I can relate the intensities.
if the a900 makes too many assumptions in their processing algorithms, why not just hack their firmware?
Well, why not? However, the effort is lost when the next firmware version is released and you need it.
There is a firmware hacker/enhancer community, but they are dealing only with certain processors and with certain cameras. Camera makers are not keen to publish any internals.
-
Bernard, the issue is not so small. At first sight it might appear that this affects only the very-very dark regions, but this is not so: even a medium dark area can have one or two very low component. For example one of the greenish (not pure green) patch on the color checker sheet Erik has been photographing contains greens in the sixth stop and blues in the ninth (or it appears in the ninths). The blues are blotchy; this may make the result blotchy.
Particularly, if one channel gets pushed up due to WB, it can affect much brighter areas as well. Why do you think Sony stopped the unwanted NR with the A700?
In some cases these pixel groups (the blotches) are like perfect mosaics; the pixels of two levels occupy over 90% of the area. Sometimes the groups are not in adjacent pixel levels; this makes their blotchiness particularly strong, when carrying over in the RGB value.
-
I am not, for I can not. However, if Erik manages to make shots with his A900 and with a Canon at the same time, then I can relate the intensities.
so then you are speculating that a 'true' raw file would have better DR than a NR off file, based on the knowledge of the differences between NR on and NR off and extrapolating because any onboard raw pre-processing must make it- the same or- worse than the original data, yes?
am i missing a crucial step?
-
so then you are speculating that a 'true' raw file would have better DR than a NR off file
1. The dynamic range of the camera is determined by the sensor; that can not be changed.
2. Presently the noise reduction (even with NR Off) causes dark areas appear even darker; this makes the impression, that the DR is greater than it is in fact. This is reinforced by measurements, which are based on the intensity after this noise reduction.
I too am measuring the noise in conjunction with the average pixel intensity on a selected patch; if that pixel intensity is artificially lowered, then the measured standard deviations get assigned to deeper shadows then it should be, i.e. the DR appears greater.
-
1. The dynamic range of the camera is determined by the sensor; that can not be changed.
2. Presently the noise reduction (even with NR Off) causes dark areas appear even darker; this makes the impression, that the DR is greater than it is in fact. This is reinforced by measurements, which are based on the intensity after this noise reduction.
I too am measuring the noise in conjunction with the average pixel intensity on a selected patch; if that pixel intensity is artificially lowered, then the measured standard deviations get assigned to deeper shadows then it should be, i.e. the DR appears greater.
check. thanks for the methodology clarification . I'll be interested to read your report, once compiled
-
It seems that Sony is guilty of what we called "tuning" their hardware to get a specific reaction from uncritical test sites. For example, Card manufacturers use to send their video cards to specific sites, knowing what software/hardware they used to analyze their cards. They would "tune" the card to perform extremely well on the tests. That latest about one quarter because the test sites caught on really fast to what was happening. That's why they always test synthetic and then back that up with ever revolving real time FPS analysis running each new iteration of game's built in FPS (frames per second) Demo. This is what Gabor is doing, backing up the "test" files by doing his own analysis at a level that cannot be faked.
And even though some people might think "So what, it looks really good, it looks like really high DR, and it looks like really low noise, so who cares?" And the answer is that if you need to manipulate an image that is noise reduced to a point of being unmanipulated, you lose the ability to further manipulate the image to your specific needs. And yeah Sony could put in a switch easily that allowed one to turn off noise reduction completely, but then everyone would be saying, Not that great noise control and DR but pretty good." And then all the people who ran out and bought the A900 might have bought the 5DMKII instead. That's a marketing decision, one that limits our control over our photography for sure.
-
Hi,
Hacking is not easy, you know...
Best regards
Erik
without getting too technical, how are you getting the raw data to compare the true DR to the augmented DR for the a900 ?
also, if the a900 makes too many assumptions in their processing algorithms, why not just hack their firmware? Naturally this would be a real pain in the ass but if you connect with a good enough community of developers (using git or something of that nature), you could make something to accomplish zero change in the raw data, no?
I'm surprised that nobody hacks camera firmware so far that I've seen... It happens with so many other devices, why not cameras?
I'm not a firmware hacker for any device, but I have a sufficiently deep knowledge of CS to begin to pick it up.
EDIT: a google came with this;
http://chdk.wikia.com/wiki/CHDK (http://chdk.wikia.com/wiki/CHDK) - Canon hacking wiki
http://www.camerahacker.com/ (http://www.camerahacker.com/) - enthusiast site
-
Hi,
There is certainly some tuning in order to get good reviews. Almost all reviews are done on JPEGs, however, so this issue with NR on raws is not necessarily related to tuning for reviews. The blotchiness reported by Panopeeper does also effect high ISO JPEGS, however.
DPReview had never given good marks for noise to any Sony or Minolta camera, by the way, so would Sony have a strategi to do excessive noise reduction it certainly does not work with DPReview, nor does it work with DxO-mark.
Another way to see it is that without the A900 there would not be any 5DII. Canon would have no reason to release the 5DII unless there was competition. Would Nikon has released the D3x without the A900 and the 5DII being around no one would complain about the price.
Aside from that, the Sony A900 is essentially doing what it is supposed to do, nameley taking pictures and it is quite good at it. It is not really a high ISO camera nor intended for sports. As a landscape camera it works very well.
The noise reduction we are discussing is not a software solution to my best understanding but something done on chip,not necessarily in softfware but probable having to do with preamps before the ADC. That may be related that it cannot be shut off completely. Sony has not been secretive about this in any way, it has been published in their description of the sensor chip.
If you check DxO-mark the 1DsIII and the 5DII are in the same league as the A900 (within some decimals). The Nikon D3x is significantly better. Panopeeper is somewhat sceptical about the claims for the D3x, but we cannot now for shure unless DxO makes their "raw" images public or someone having a D3x makes the images Panopeeper needs for his evaluations.
Best regards
Erik
It seems that Sony is guilty of what we called "tuning" their hardware to get a specific reaction from uncritical test sites. For example, Card manufacturers use to send their video cards to specific sites, knowing what software/hardware they used to analyze their cards. They would "tune" the card to perform extremely well on the tests. That latest about one quarter because the test sites caught on really fast to what was happening. That's why they always test synthetic and then back that up with ever revolving real time FPS analysis running each new iteration of game's built in FPS (frames per second) Demo. This is what Gabor is doing, backing up the "test" files by doing his own analysis at a level that cannot be faked.
And even though some people might think "So what, it looks really good, it looks like really high DR, and it looks like really low noise, so who cares?" And the answer is that if you need to manipulate an image that is noise reduced to a point of being unmanipulated, you lose the ability to further manipulate the image to your specific needs. And yeah Sony could put in a switch easily that allowed one to turn off noise reduction completely, but then everyone would be saying, Not that great noise control and DR but pretty good." And then all the people who ran out and bought the A900 might have bought the 5DMKII instead. That's a marketing decision, one that limits our control over our photography for sure.
-
Hi!
I don't really feeleasy about this. The problem is that I can probably borrow a Canon, but I probably need to change a lot of settings, which I don't know how to do, worse is that the owner probably wants it back in the same shape as it was before. But I may be able to engage a Canon owner in this project.
Best regards
Erik
I am not, for I can not. However, if Erik manages to make shots with his A900 and with a Canon at the same time, then I can relate the intensities.
-
I am duplicating this post from dyxum.
"Interesting thread. I've got a couple of comments/questions.
[Kiklop,] am I correct in remembering that Sony has been using non-linear RAW data as far back as the A100?
Secondly, I take DxO Mark, like all reviews, with a grain of salt. That being said, it seems strange to me that in light of their brand new A700 review, which goes into some detail about NR on the green channel prior to firmware V4, DxO mark wouldn't notice NR on the blue/red channel in the A700/A900 if it existed. Panopeeper, I have a VERY limited knowledge on your raw analyzer and how it works, so I'm wondering why there is this discrepancy?? Thanks.
Dxo Mark A700 NR article (http://www.dxomark.com/index.php/eng/DxOMark-Sensor/DxOMark-reviews/DxOMark-review-for-the-Sony-Alpha-700)
-
Hi,
What I gather is that Sony does a different kind of noise reduction than DxO is checking for. DxO is checking for spatial noise reduction, that is some kind of mixing the signal from a center pixel with surrounding pixels, the kind of noise reduction Panopeeper has detected is more on the signal itself.
"2. The noise reduction affects mainly the red and the blue channels, much less the green. This is a very primitive noise reduction; it simply eliminates some pixel levels in the affected areas.
3. Due to the nature of this noise reduction, the very dark, noise reduced areas become darker. The consequence is, that not only the visual appearance but the noise measurement too indicates not only lower noise that it would be without NR, but that in deeper shadow, suggesting a greater dynamic range."
From some of the writings I have seen from Sony may indicate that this design was chosen so the chip would cooperate well with the DRO (Dynamic Rnage Optimizer) that brightens up dark parts.
Best regards
Erik
I am duplicating this post from dyxum.
"Interesting thread. I've got a couple of comments/questions.
[Kiklop,] am I correct in remembering that Sony has been using non-linear RAW data as far back as the A100?
Secondly, I take DxO Mark, like all reviews, with a grain of salt. That being said, it seems strange to me that in light of their brand new A700 review, which goes into some detail about NR on the green channel prior to firmware V4, DxO mark wouldn't notice NR on the blue/red channel in the A700/A900 if it existed. Panopeeper, I have a VERY limited knowledge on your raw analyzer and how it works, so I'm wondering why there is this discrepancy?? Thanks.
Dxo Mark A700 NR article (http://www.dxomark.com/index.php/eng/DxOMark-Sensor/DxOMark-reviews/DxOMark-review-for-the-Sony-Alpha-700)
-
The noise reduction we are discussing is not a software solution to my best understanding but something done on chip,not necessarily in softfware but probable having to do with preamps before the ADC
The on-chip noise reduction is a different class (it should be called "noise avoidance"). However, that is single pixel oriented and never becomes apparent on the image.
The technology is described here (on the "public level"): http://usa.canon.com/uploadedimages/FCK/Im...non_CMOS_WP.pdf (http://usa.canon.com/uploadedimages/FCK/Image/White%20Papers/Canon_CMOS_WP.pdf)
-
Could the second pass of sensor NR illustrated in this graph be the possible culprit? Much has been made by Sony that they are adding on-chip NR to both the analog and post ADC digital signal, but this is all done on-chip, due to the EXMOR ADC design being on-chip. I'm curious if the D3x differs in 12bit mode.
(https://www.camerasdirect.com.au/reviews/convergence/SR-12-Exmor-diagram.jpg)
-
That being said, it seems strange to me that in light of their brand new A700 review, which goes into some detail about NR on the green channel prior to firmware V4, DxO mark wouldn't notice NR on the blue/red channel in the A700/A900 if it existed
I have a much smaller "raw base" (raw images files) with the A700 than with the A900; particularly, the Imaging Resource images are not underexposed. Anyway, I have not seen any sign of THAT kind of noise reduction, not even in the darkest regions, what I see in the A900 images. This is immediately visible already on the histograms; the A900 fine histograms proudly display the signs of manipulation (the first one is A700 ISO 100 firmware version 3, the second is A900, ISO 100):
-
Could the second pass of sensor NR illustrated in this graph be the possible culprit?
This is really interesting, but too superficial (not as if I would understand it if it dealt with hardware measures). The question is if it could cause the blobs, the tight groups of pixels with the same value.
Thzere is one consideration against that, and for software (firmware) manipulation: the effect of NR with ISO 1600, where it is selectable, is the same as what is happening in ISO 100-800, but the degree depends on the level of NR. I posted captures with the "involuntary" blobs above. The followings are from ISO 1600, NR High; the blobs have the clearest appearance here; with lower level NR it is becoming like what I posted above. The histograms of the darkest patch on the color checker too show the effect of increasing noise reduction: the levels get "pushed together" (the file names in the top white field in the capture indicate the level of NR).
-
3. Due to the nature of this noise reduction, the very dark, noise reduced areas become darker. The consequence is, that not only the visual appearance but the noise measurement too indicates not only lower noise that it would be without NR, but that in deeper shadow, suggesting a greater dynamic range."
If this is anything like typical LCD displays, a darker black level isn't necessarily a bad thing. I suspect Sony has purposely tweaked the A900 towards landscape and lower ISO uses. After all, this is what most of the people who would buy the camera would use it for. For $2500, it's a very inexpensive 2nd place to the Nikon. I'd be happy to own one.
It's obviously using a different method of NR than you usually find. I've seen the raw files from the A900, and honestly, for the price, it's not a bad way of doing NR, IMO. Whatever they are doing seems to be good enough.
-
I have found the *perceived* noise characteristics of *my* D3x to be really strange: At 1600 it's unbelievably good, more than enough for fashion/portrait use. After 1600 it's like going off a cliff.
Edmund
-
I have found the *perceived* noise characteristics of *my* D3x to be really strange: At 1600 it's unbelievably good, more than enough for fashion/portrait use
There is no healthy patient, only incompetent diagnosis.
-
There is no healthy patient, only incompetent diagnosis.
you're being elliptically cryptic
Edmund
-
I'm afraid you are mixing up some issues. Although the raw data of Nikon cameras is not as raw as the Canon raw, and the difference contributes to a lower apparent noise, that is not comparable to the noise reduction carried out by the A900 on the raw data, and even less to the costomary noise reduction.
Just to clarify, you are saying that even at low ISOs, such as 100-400 ISO, you are saying there is some kind of NR being applied in-camera?
-
you're being elliptically cryptic
That's called crypto-logiC.
(The thing with the diagnosis is an old joke among doctors.)
-
Just to clarify, you are saying that even at low ISOs, such as 100-400 ISO, you are saying there is some kind of NR being applied in-camera?
Pls read the opening message of this thread; it is about nothing else.
-
Pls read the opening message of this thread; it is about nothing else.
I'm curious whether it's being applied on-chip in the second NR portion, or in the BIONZ??
-
I'm curious whether it's being applied on-chip in the second NR portion, or in the BIONZ??
On chip is possible because it is a CMOS !
Thierry
-
Euh...
- ease of use through more streamlined workflows,
- better DxOMark results that make the A900 puchasers feel good about their investement
- less apparent noise in low ISO shadows that will prevent people from over-exposing, therefore blowing highlights
Everything is relative Gabor. You are only dissatisfied with the shadows of the A900 because you have found that they could have been closer to your taste (meaning noisy but more detailed). All those users who don't know this are just happy about having clean shadows... See above for some ideas on the value of clean shadows.
Don't get me wrong, I am with you here, noise reduction in camera is not something I find to be a good idea, but I am just trying to put things in perspective.
I agree with you Bernard, you put in perspective the well known fact that many people will be happy without knowing that his wife or her husband is seeing another person, or will pick the blue pill and enjoy eating the fake beefsteak in Matrix the movie. But some other people prefer to know what's really under the hood, that's why I think Gabor's analysis are very valuable.
I also made a list of advantages of all these tests and findings:
- The user has the power to _decide_. To decide if switching on/off camera's NR. To decide if purchasing or not a camera where NR cannot be switched off.
- Camera vendors can find out not everyone out there are just followers willing to buy any new camera on the market that _seems_ to perform better than others, so maybe they finally concentrate on making cameras that _do perform_ better than others.
- Sites like DxO Mark, whose RAW analysis are the best found on any public site, can admit their quality measuring criteria is weak against camera vendor RAW tricks, and mend this.
...
-
I'm curious whether it's being applied on-chip in the second NR portion, or in the BIONZ??
The on-chip noise reduction is rather noise avoidance. It cares for single pixels, it does not create blobs.
Moreover, the noise reduction we can see here is selective: in ISO 1600 (and above, I guess) it can be turned on to one of three levels or off. The signs of that actuion are identical to those observable in ine lower ISO shots, but not selectable.
-
There is no healthy patient, only incompetent diagnosis.
and there are plenty of doctors who do not listen.
-
FWIW, Iliah Borg has weighed in on this issue a bit. Maybe you could get involved in this thread, Gabor:
:link to Iliah's comments: (http://forums.dpreview.com/forums/readflat.asp?forum=1037&message=31285521)
-
Gabor, following the use of FFT to find traces of noise reduction made by Emil Martinec here (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p4.html), I calculated the FFT distribution of noise found in the cup of post #19 (http://luminous-landscape.com/forum/index.php?s=&showtopic=32777&view=findpost&p=266346):
(http://img24.imageshack.us/img24/6760/compd.jpg)
There is no evidence, from the FFT perspective, that any noise reduction has been applied to the RAW data in the 'OFF' setting. So would it be logical that NR is applied to low ISOs from 100 to 800 in the 'OFF' setting?
On the other side for any other setting a clear spatial pattern arises in the FFT (in fact is the same pattern for 'LOW', 'MEDIUM' and 'HIGH', just with different amplitudes) that makes think of some noise reduction algorithm based not only in adjacent pixels but also pixels at greater distances.
Perhaps Emil can give his interpretation.
BR
-
Gabor, following the use of FFT to find traces of noise reduction made by Emil Martinec here (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p4.html), I calculated the FFT distribution of noise found in the cup of post #19 (http://luminous-landscape.com/forum/index.php?s=&showtopic=32777&view=findpost&p=266346):
http://img24.imageshack.us/img24/6760/compd.jpg (http://img24.imageshack.us/img24/6760/compd.jpg)
There is no evidence, from the FFT perspective, that any noise reduction has been applied to the RAW data in the 'OFF' setting. So would it be logical that NR is applied to low ISOs from 100 to 800 in the 'OFF' setting?
On the other side for any other setting a clear spatial pattern arises in the FFT (in fact is the same pattern for 'LOW', 'MEDIUM' and 'HIGH', just with different amplitudes) that makes think of some noise reduction algorithm based not only in adjacent pixels but also pixels at greater distances.
Perhaps Emil can give his interpretation.
BR
The uniform appearance of the FT of the image with NR off supports the notion that no noise reduction is being used. It's not a guarantee, more the other way around -- if one *does* see the appearance of a pattern, with the noise power reduced at high spatial frequency, then that is strong evidence that there *is* NR taking place. (High spatial frequency is located at the edges of the FT image; reduced noise power at high frequency means that the speckling is darker there.) It could be that there is still NR being done even when the speckling is uniform, but that it's of a more subtle form than a simple finite range spatial filter.
Iliah seems to think that the issue is black clipping:
http://forums.dpreview.com/forums/read.asp...essage=31285521 (http://forums.dpreview.com/forums/read.asp?forum=1037&message=31285521)
Why oh why don't camera manufacturers offset their data so that the noise spectrum is not corrupted?
-
There is no evidence, from the FFT perspective, that any noise reduction has been applied to the RAW data in the 'OFF' setting. So would it be logical that NR is applied to low ISOs from 100 to 800 in the 'OFF' setting?
Actually, this is what I stated in the initial post: the NR setting does apply to ISO 1600 (and I guess higher, but I did not care for that), but it has no effect at lower ISOs; some NR *always* takes place.
On the other side for any other setting a clear spatial pattern arises in the FFT (in fact is the same pattern for 'LOW', 'MEDIUM' and 'HIGH', just with different amplitudes) that makes think of some noise reduction algorithm based not only in adjacent pixels but also pixels at greater distances
The actual noise reduction is very primitive. I am working on the demonstration (in fact on a dozen demonstrations, I have created already hundreds of captures). The result of the NR is the reduction of the number of levels and "blobs" of pixels in the same level.
The attachments in my post #38 should be clear enough, I don't see much room for speculations.
-
Actually, this is what I stated in the initial post: the NR setting does apply to ISO 1600 (and I guess higher, but I did not care for that), but it has no effect at lower ISOs; some NR *always* takes place.
The actual noise reduction is very primitive. I am working on the demonstration (in fact on a dozen demonstrations, I have created already hundreds of captures). The result of the NR is the reduction of the number of levels and "blobs" of pixels in the same level.
The attachments in my post #38 should be clear enough, I don't see much room for speculations.
What to you think in regards to Iliah's black clipping explanation?
-
I have found the *perceived* noise characteristics of *my* D3x to be really strange: At 1600 it's unbelievably good, more than enough for fashion/portrait use. After 1600 it's like going off a cliff.
Edmund
Part of it is the fact the noise in the D3x that is visible is so damn attractive compared to various other cameras.
-
What to you think in regards to Iliah's black clipping explanation?
It is NOT black cipping.
-
Part of it is the fact the noise in the D3x that is visible is so damn attractive compared to various other cameras.
wow! what does make D3X's noise so attractive? I thought digital noise was just a gaussian distribution on any camera sensor, with no spatial structure and which amplitude solely depends on the number of converted photons and the ISO gain setting.
-
It is NOT black cipping.
Could explain a bit more in detail? Iliah wrote some pretty interesting stuff in that thread.
-
Could explain a bit more in detail? Iliah wrote some pretty interesting stuff in that thread.
I am working on it. It is not a simple, straightforward issue; I am demonstrating it through a whole bunch of observations, all heavy duty pixel peeping.
-
wow! what does make D3X's noise so attractive? I thought digital noise was just a gaussian distribution on any camera sensor, with no spatial structure and which amplitude solely depends on the number of converted photons and the ISO gain setting.
There is also pattern noise, see here for an example from a 20D:
http://theory.uchicago.edu/~ejm/pix/20d/te...e/#patternnoise (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/#patternnoise)
these sorts of noises have a visual impact that exceeds their contribution to noise measures like the standard deviation, because even though they may be small in magnitude for any given pixel, they are coherent across many pixels and therefore get picked up by visual processing that is looking for patterns.
The D3x seems to have such noises well under control. Also the read noise std dev of the D3x at base ISO is quite low compared to other DSLRs at base ISO.
-
I am working on it. It is not a simple, straightforward issue; I am demonstrating it through a whole bunch of observations, all heavy duty pixel peeping.
Practically, nothing that has to do with photography.
-
Practically, nothing that has to do with photography.
Of course, if you shoot jpegs.
-
Guillermo,
I can send you the other images if you are interested, I have similar images from 100 to 1600 ISO with different noise reductions.
Best regards
Erik
Gabor, following the use of FFT to find traces of noise reduction made by Emil Martinec here (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p4.html), I calculated the FFT distribution of noise found in the cup of post #19 (http://luminous-landscape.com/forum/index.php?s=&showtopic=32777&view=findpost&p=266346):
(http://img24.imageshack.us/img24/6760/compd.jpg)
There is no evidence, from the FFT perspective, that any noise reduction has been applied to the RAW data in the 'OFF' setting. So would it be logical that NR is applied to low ISOs from 100 to 800 in the 'OFF' setting?
On the other side for any other setting a clear spatial pattern arises in the FFT (in fact is the same pattern for 'LOW', 'MEDIUM' and 'HIGH', just with different amplitudes) that makes think of some noise reduction algorithm based not only in adjacent pixels but also pixels at greater distances.
Perhaps Emil can give his interpretation.
BR
-
To see if there is noise reduction, clipping, or other shadow treatment; or simple in-camera flare one needs to perform resolution tests. Shots of a newspaper from a proper distance (so that the text is easily readable if the newspaper is exposed according to the meter), taken with a good prime lens with as few elements as possible (think Tessar) stopped down to f/5.6 - f/8 and viewfinder covered, -2 to -6 EV, at various ISO settings, from a steady tripod, - that is a useful photographic test, especially when done under incandescent, fluorescent, flash, and daylight.
-
I can send you the other images if you are interested, I have similar images from 100 to 1600 ISO with different noise reductions.
ISO100 with NR set to OFF and HIGH would be the most interesting. However let's wait for Gabor's results first.
-
ISO100 with NR set to OFF and HIGH would be the most interesting. However let's wait for Gabor's results first.
I believe that Gabor is saying that NR is always on at low ISO, and it doesn't change, regardless of the high ISO NR setting. Granted, he's got a lot of convincing to do, since he doesn't really have proper test shots like was mentioned above.
-
Hi,
I'd suggest that Gabor is having almost all test shots he needs. By the way, I can see the patterns Gabor found in my 100 ISO shots using the DxO converter. Gabor has invested a lot of work on this issue an I have contributed some.
Best regards
Erik
I believe that Gabor is saying that NR is always on at low ISO, and it doesn't change, regardless of the high ISO NR setting. Granted, he's got a lot of convincing to do, since he doesn't really have proper test shots like was mentioned above.
-
I'm not saying that there hasn't been a lot of work done. I appreciate the effort. The question is whether the work is being done correctly, which I believe is what Iliah implied above.
-
Hi,
I guess that we can see different aspects depending the tests/evaluations we have made. We can see some artifacts in raw-images from the A900, that's pretty certain. Now there are some questions:
- Are these artifacts relevant?
- If so under which conditions?
- What can we do about them?
It is unfortunate if Sony gets some bad reputation because of these discussions. The artifacts we see are not normally visible under low ISO conditions. It is conventional wisdom that the A900 performs worse on high ISO noise than the competition, possibly exaggerated by unadvantageous processing in ACR/LightRoom.
Regarding the noise reduction at low ISO I do have 100-1600 ISO shots like the ones Guillermo analyzed. So I plan to go ahead and check FFTs on all those images, but I don't have image analysis software on my primary workstation (Apple Imac using Tiger), so I need to revert to another platform, or find good software for the Mac. The need is essentially to be able to read unprocessed images in TIFF or similar format and perform FFT on them.
I also have a comment on Michaels Reichmann's findings on the Canon 5DII, A900 and Nikon D3x. What he says that thet are virtually indistinguishable up to 800 ISO but above that the Nikon and the Canon have a one step advantage. MR also very clearly stated that he was not discussing DR (Dynamic Range), because that was beyond his means to evaluate. Michael's findings are essentially consistent with DxO-mark. In my view the simple truth is probably that at ISO 100-200 all the cameras are good enough. There may be a theoretical advantage to the D3x under lab conditions but they may or may not be relevant under real life conditions. Bernard has shown some pretty impressive images by his Nikon D3x which may demonstrate the need for DR at low ISO, but for my picture taking DR at ISO 200 seldom was a limiting factor.
Best regards
Erik
I'm not saying that there hasn't been a lot of work done. I appreciate the effort. The question is whether the work is being done correctly, which I believe is what Iliah implied above.
-
Hi,
I guess that we can see different aspects depending the tests/evaluations we have made. We can see some artifacts in raw-images from the A900, that's pretty certain. Now there are some questions:
- Are these artifacts relevant?
- If so under which conditions?
- What can we do about them?
It is unfortunate if Sony gets some bad reputation because of these discussions. The artifacts we see are not normally visible under low ISO conditions. It is conventional wisdom that the A900 performs worse on high ISO noise than the competition, possibly exaggerated by unadvantageous processing in ACR/LightRoom.
Regarding the noise reduction at low ISO I do have 100-1600 ISO shots like the ones Guillermo analyzed. So I plan to go ahead and check FFTs on all those images, but I don't have image analysis software on my primary workstation (Apple Imac using Tiger), so I need to revert to another platform, or find good software for the Mac. The need is essentially to be able to read unprocessed images in TIFF or similar format and perform FFT on them. I can write that kind of software, by the way, but it's better to use something that is well tested and proven.
I also have a comment on Michaels Reichmann's findings on the Canon 5DII, A900 and Nikon D3x. What he says that thet are virtually indistinguishable up to 800 ISO but above that the Nikon and the Canon have a one step advantage. MR also very clearly stated that he was not discussing DR (Dynamic Range), because that was beyond his means to evaluate. Michael's findings are essentially consistent with DxO-mark. In my view the simple truth is probably that at ISO 100-200 all the cameras are good enough. There may be a theoretical advantage to the D3x under lab conditions but they may or may not be relevant under real life conditions. Bernard has shown some pretty impressive images by his Nikon D3x which may demonstrate the need for DR at low ISO, but for my picture taking DR at ISO 200 seldom was a limiting factor.
Best regards
Erik
-
I get a sense that discussions like this drift too far from practical, real world photos, which are surely the main purpose of photography.
Dynamic range has always been an issue with photography. The camera with a single shot has never been able to capture the dynamic range that the eye sees, because the camera has a fixed aperture at the time of exposure, whereas the eye's pupil contracts and dilates almost instantly as its gaze changes from shadows to highlights.
Here's a shot taken years ago with my Canon D60. Exposure was very much to the right, requiring a -1.1EV adjustment in ACR as well as a 95 recovery of highlights. I suspect that Gabor, using Rawlanyze could point out that the following shot is in fact overexposed. Yet the shadows are awful, as you can see from the 100% crop of the bottom right corner.
This scene was simply beyond the DR capability of the Canon D60. My D700 would have probably have managed it with aplomb.
[attachment=12188:full_image.jpg] [attachment=12189:Bottom_r...t_corner.jpg]
-
Hi,
I'd suggest that Gabor is having almost all test shots he needs. By the way, I can see the patterns Gabor found in my 100 ISO shots using the DxO converter. Gabor has invested a lot of work on this issue an I have contributed some.
Best regards
Erik
No wonder, DxO applies noise reduction during the conversion.
-
Ray,
I see two reasons for testing, one is curiousity the other the need to explore the limitations of a system in order to learn living with it.
Erik
I get a sense that discussions like this drift too far from practical, real world photos, which are surely the main purpose of photography.
Dynamic range has always been an issue with photography. The camera with a single shot has never been able to capture the dynamic range that the eye sees, because the camera has a fixed aperture at the time of exposure, whereas the eye's pupil contracts and dilates almost instantly as its gaze changes from shadows to highlights.
Here's a shot taken years ago with my Canon D60. Exposure was very much to the right, requiring a -1.1EV adjustment in ACR as well as a 95 recovery of highlights. I suspect that Gabor, using Rawlanyze could point out that the following shot is in fact overexposed. Yet the shadows are awful, as you can see from the 100% crop of the bottom right corner.
This scene was simply beyond the DR capability of the Canon D60. My D700 would have probably have managed it with aplomb.
[attachment=12188:full_image.jpg] [attachment=12189:Bottom_r...t_corner.jpg]
-
I plan to go ahead and check FFTs on all those images, but I don't have image analysis software on my primary workstation (Apple Imac using Tiger)
ImageJ works on Mac, you may need Java upgrade. For Tiger you may need to use 32-bit version, worth a try IMHO. http://rsb.info.nih.gov/ij/docs/install/osx.html (http://rsb.info.nih.gov/ij/docs/install/osx.html)
To extract raw channels from a raw file you can use command-line utility, 4channels, pre-compiled for Mac in LibRaw http://www.libraw.org/data/LibRaw-0.7.0-MacOSX.zip (http://www.libraw.org/data/LibRaw-0.7.0-MacOSX.zip) If you will have problems running it on Tiger please let us know.
-
Ray,
I see two reasons for testing, one is curiousity the other the need to explore the limitations of a system in order to learn living with it.
Erik
Curiosity is fine. But how are you going to translate the data that Gabor produces to real world scenes in a way that is practically useful?
We're into hair-splitting here, aren't we?
-
Curiosity is fine. But how are you going to translate the data that Gabor produces to real world scenes in a way that is practically useful?
We're into hair-splitting here, aren't we?
No, we are not into hair-splitting. Translation of raw data into the images of real world scenes is done with the help of those who develop raw converters. The more we know about raw data the better are the raw converters we code and the better are the converters we use.
-
No, we are not into hair-splitting. Translation of raw data into the images of real world scenes is done with the help of those who develop raw converters. The more we know about raw data the better are the raw converters we code and the better are the converters we use.
Well that's a different matter. Is this what the thread's really about? Prising information about the structure of the RAW file in order to produce a better RAW converter? I'm all in favour of that, but I find it strange that any company such as Sony would not cooperate with developers of RAW converters who wish to enhance the images from Sony cameras.
-
Well that's a different matter. Is this what the thread's really about? Prising information about the structure of the RAW file in order to produce a better RAW converter? I'm all in favour of that, but I find it strange that any company such as Sony would not cooperate with developers of RAW converters who wish to enhance the images from Sony cameras.
From what I understand, none of the camera companies provide disclosure on their raw "recipe," and that's part of the problem for raw converter developers.
-
From what I understand, none of the camera companies provide disclosure on their raw "recipe," and that's part of the problem for raw converter developers.
Actually, "recipe" is a good term. Structure of raw is not very hard to decode. It is parametadata that is missing.
-
Regarding the noise reduction at low ISO I do have 100-1600 ISO shots like the ones Guillermo analyzed. So I plan to go ahead and check FFTs on all those images, but I don't have image analysis software on my primary workstation (Apple Imac using Tiger), so I need to revert to another platform, or find good software for the Mac. The need is essentially to be able to read unprocessed images in TIFF or similar format and perform FFT on them.
Erik I use a PS plugin for FFT by Alexandre V. Chirokov. You can download it from here (http://www.guillermoluijk.com/download/fft.zip). Just save it in the usual path for plugins and apply it to the noise image having been converted to 8 bit. Channel G of the resulting image is the FFT amplitude.
For extracting the RAW data from each channel, you can use DCRAW in Mac. I would recommend:
dcraw -v -d -r 1 1 1 1 -T -4 -t 0 file.cr2
The resulting TIFF will be a grayscale RGGB pattern with RAW data. It was simply rescaled and fitted in the 16-bit range.
To isolate each of the R, G1, G2 and B channels you don't need to leave PS, just rescale to 50% using the 'Aproximation' (no idea if the name changes for the English versions of PS) method. This performs a nearest neighbour rescaling which will pick only one out of each 4 channels:
(http://www.guillermoluijk.com/article/rawnoise/extraccion.gif)
(http://www.guillermoluijk.com/article/rawnoise/extraccion2.gif)
For a RGGB pattern:
1. Straigt rescaling provides the B channel (like in the previous image).
2. If you add 1 pixel of canvas on both sides prior to rescaling it provides the G2 channel
3. If you add 1 pixel of canvas up and down prior to rescaling it provides the G1 channel
4. If you add 1 pixel of canvas on both sides so as up and down prior to rescaling it provides the R channel
Once you have the desired channel, crop an area with uniform noise (no texture), converto to 8 bit and apply the FFT as explained above.
I would like to see the FFT at ISO100 with NR OFF and HIGH. Remember after FFT, the G channel of the image contains the FFT modulus, which can be enhanced using a contrast curve.
BR
-
For extracting the RAW data from each channel, you can use DCRAW in Mac. I would recommend:
dcraw -v -d -r 1 1 1 1 -T -4 -t 0 file.cr2
The resulting TIFF will be a grayscale RGGB pattern with RAW data. It was simply rescaled and fitted in the 16-bit range.
To isolate each of the R, G1, G2 and B channels you don't need to leave PS, just rescale to 50% using the 'Aproximation' (no idea if the name changes for the English versions of PS) method. This performs a nearest neighbour rescaling which will pick only one out of each 4 channels:
(http://www.guillermoluijk.com/article/rawnoise/extraccion.gif)
(http://www.guillermoluijk.com/article/rawnoise/extraccion2.gif)
For a RGGB pattern:
1. Straigt rescaling provides the B channel (like in the previous image).
2. If you add 1 pixel of canvas on both sides prior to rescaling it provides the G2 channel
3. If you add 1 pixel of canvas up and down prior to rescaling it provides the G1 channel
4. If you add 1 pixel of canvas on both sides so as up and down prior to rescaling it provides the R channel
Once you have the desired channel...
4channels is easier, takes only file name or a wildcard as a parameter and you will get 4 totally unmodified raw planes as TIFF files.
-
Guillermo and Iliah,
Thanks for all good suggestions. I finally used 4Channels and ImageJ under Leopard. 4Channels would SIGV under Tiger.
I tested three different images from the same series that were analyzed by Guillermo, ISO 100 with NR OFF and HIGH and ISO 1600 with NR HIGH.
The result was that I could not see anything significant in the FFTs of the 100 ISO-images while there were patterns in the 1600 ISO image. According to the manual high ISO NR is off from 100 ISO to 800 ISO.
Left image ISO 100, NR HIGH. Right image ISO 1600, NR HIGH.
Best regards
Erik
-
Sorry for the delay, but this has proven to be much more laborous than I had thought. I have been experiencing with different ways of demonstrating that, what is obvious for me from analysis conducted by Rawnalyze.
Sony A900 - the effect of noise reduction on the raw data (http://www.cryptobola.com/PhotoBola/SonyA900/SonyA900_NR.htm)
-
Is there any way to be 100% positive that it's not one of the converters or programs that's doing this? If so, this is a huge problem with the Sony. RAW should be exactly that and not what appears to me to be JPEG-like NR and compression in a RAW format.
-
Is there any way to be 100% positive that it's not one of the converters or programs that's doing this? If so, this is a huge problem with the Sony. RAW should be exactly that and not what appears to me to be JPEG-like NR and compression in a RAW format.
It is not noise reduction, it is a crude analog to digital conversion that drops levels and clips to black forming clusters and causing posterization in shadows. That, indeed, is some problem; and Sony is well aware of it. By the way, the problem exists not only in Sony cameras.
-
It is not noise reduction, it is a crude analog to digital conversion that drops levels and clips to black forming clusters and causing posterization in shadows. That, indeed, is some problem; and Sony is well aware of it. By the way, the problem exists not only in Sony cameras.
Is this not a case of 'you get what you pay for'? When you say that the problem is due to a crude A/D conversion, that implies that the solution to the problem is a more sophisticated and therefore more expensive A/D converter, or a completely redesigned A/D converter costing lots of research dollars which would have to be recouped with a higher price for the camera.
-
Hi,
The problem with this discussion is that it develops to a lot of Sony bashing. The problems that Panopeeper see show up in well designed tests, but not really in normal pictures. That said there are probably other cameras offering better noise performance like the Nikon D3X or the Canon 5DII. Some of he problems get exaggerated if the ACR engine is used for RAW-conversion. Panopeeper and Iliah don't agree on the cause of the problem. In my view Gabor did an excellent work in analyzing the images, but the interpretation of his results is subject to discussion. It seems that the Canon images are void of this problem. I obviously cannot compare the three top tier FF cameras, owning only one.
The two images below are from a test shoot of my new Sigma 12-24/4.5-5.6 lens, surprisingly good, BTW. One is the full image, the other one a detail from the shadows with "Light and Shadows" applied over part of the image. In my view this image is quite demanding regarding DR. The highlight include sun visible trough clouding and it is of course clipped, recovery turns it into a grey area.
Sorry for having problems with putting attachements in the right place...
Erik
[attachment=12203:20090315...748_Edit.jpg]
It is not noise reduction, it is a crude analog to digital conversion that drops levels and clips to black forming clusters and causing posterization in shadows. That, indeed, is some problem; and Sony is well aware of it. By the way, the problem exists not only in Sony cameras.
-
Is this not a case of 'you get what you pay for'? When you say that the problem is due to a crude A/D conversion, that implies that the solution to the problem is a more sophisticated and therefore more expensive A/D converter, or a completely redesigned A/D converter costing lots of research dollars which would have to be recouped with a higher price for the camera.
It's a little more complicated than that, because there are many ADCs, and they're all on the chip. The EXMOR is a completely different design than what we've seen from other cmos makers, and there are good and bad things about it. It seems that, however the D3x is attaining it's 14bit processing, it partly shows why the camera is so expensive. We've known all along that the D3x is better in the shadows.
-
It's a little more complicated than that, because there are many ADCs, and they're all on the chip. The EXMOR is a completely different design than what we've seen from other cmos makers, and there are good and bad things about it. It seems that, however the D3x is attaining it's 14bit processing, it partly shows why the camera is so expensive. We've known all along that the D3x is better in the shadows.
Can you expand upon that? Are you saying that each of the 24m pixels has its own A/D converter? If that's the case, presumably they are all the same on the same design of chip. The D3X presumably has a different design and quality of A/D converter because it's 14 bit instead of 12, which presumably contributes to the extra cost of the D3X.
Are you implying that the problem with the A900 is basically due to its 12 bit processing?
-
Ray,
The one you should ask is Iliah, because he discussed the raw-conversion problem. The Sony has not 24 million ADCs but well over 6000, one for each column. Both the A900 and the D3X seems to use on chip AD-converters, with the Nikon being able to use 12-bit or 14-bit ADC conversion, 12-bit conversion is 5 FPS whereas 14-bit conversion is 1.8 FPS. Now, the FPS difference doesn't count if you are shooting with self timer and mirror lock up. I don't know how Nikon achieves 14-bit conversion, could be that they have 14-bit converters on chip or that they use existing 12-bit converters in a smart way. To my understanding the D3 does have external ADCs on the circuit board but these are absent from the D3X.
If you dig into into DxO-mark you can find the following figures:
[attachment=12207:Alpha900.jpg]
[attachment=12206:Canon5DII.jpg]
[attachment=12208:Nikon3DX.jpg]
You see that Nikon and Alpha are similar and Canon is different, why I don't know.
Best regards
Erik
Can you expand upon that? Are you saying that each of the 24m pixels has its own A/D converter? If that's the case, presumably they are all the same on the same design of chip. The D3X presumably has a different design and quality of A/D converter because it's 14 bit instead of 12, which presumably contributes to the extra cost of the D3X.
Are you implying that the problem with the A900 is basically due to its 12 bit processing?
-
The Sony has not 24 million ADCs but well over 6000, one for each column.
Ah! That makes more sense. But it's still a puzzle to me how any design of a RAW converter could undo what is a fait accompli due to sloppy hardware A/D conversion.
-
But it's still a puzzle to me how any design of a RAW converter could undo what is a fait accompli due to sloppy hardware A/D conversion.
It is not sloppy (crude and sloppy have different meanings IMHO), it is state of the art, meaning the camera successfully competes over my D3X and 1Ds MkIII, rendering sometimes colour detail separation superior to D3X but for quite a lot less money. In raw converter void shadows and void highlights in one of the channels are treated equally - through additional interpolation.
-
The result was that I could not see anything significant in the FFTs of the 100 ISO-images while there were patterns in the 1600 ISO image. According to the manual high ISO NR is off from 100 ISO to 800 ISO.
Left image ISO 100, NR HIGH. Right image ISO 1600, NR HIGH.
I wouldn't be so sure. There seems to be 'something' happening at ISO100:
(http://img22.imageshack.us/img22/8186/900fft.gif)
This could agree with the soft NR Gabor observed. I'll try to read his article now but I am working on a Jurassic connection speed these days.
BR
-
There seems to be 'something' happening at ISO100:
(http://img22.imageshack.us/img22/8186/900fft.gif)
Something is definitely happening at low ISO, especially with those weaker channels, red and blue (daylight shooting) - no question about that. If you derive the looks of FFT for the type of ADC behavior I described you can see similarities. Furthermore, if you examine raw files taken at different ISO settings including intermediate, like 320, you will see that the effect is in fact weaker at ISO settings higher than base. Comparative study with other cameras, 5D original on the first place, will show more. Attributing the effect to pixel-level noise reduction one needs to come up with an explanation how it should work to produce the distributions we observe.
-
In my view this image is quite demanding regarding DR
Erik, what about uploading this raw file?
-
It is not noise reduction
1. "It" does reduce the noise,
2. "it" can be influenced by NR selection,
3. Sony calls "it" noise reduction.
Reasons enough to call it noise reduction.
it is a crude analog to digital conversion that drops levels
1. I find it highly improbable, that the analog circuitry is capable of doing this on *five levels, selectable by the user*.
2. The green channels would be equally affected.
and clips to black
It does not.
-
> "it" can be influenced by NR selection
Please demonstrate it.
> The green channels would be equally affected.
Please demonstrate this too.
Gabor, what I read in your post is a lot of statements without any proof, model, all subjective interpretation, and you refuse to perform experiments that will show you why and where your methods and conclusions are wrong. OK, send your study to Sony and see what they will answer you.
-
Yeah but where? I can send it to you or anyone asking for it.
Best Regards
Erik
Erik, what about uploading this raw file?
-
Yeah but where?
Yousendit, as usual. If you post the link here, then everyone can download it; if you don't want that, then send the link to me and to those specifically asking for it.
-
> "it" can be influenced by NR selection
Please demonstrate it.
> The green channels would be equally affected.
Please demonstrate this too
Sony A900 - the effect of noise reduction on the raw data (http://www.cryptobola.com/PhotoBola/SonyA900/SonyA900_NR.htm) is doing just that.
what I read in your post is a lot of statements without any proof
Interesting; I see every statement pretty well proven.
all subjective interpretation
Really? What about some specifics?
My qualification of the noise reduction as crude is certainly subjective. Isn't it interesting that you used the same term?
send your study to Sony and see what they will answer you
Why would I? I am not working for them, nor do I own the camera.
-
I appreciate all the hard work, Gabor, and I'm really interested in what's going on with the A900 at low ISO, but since Iliah seems to be implying that what we're seeing is caused by hardware, and, in the past you've said, "I have zero competence on the hardware area," I'm having a hard time with your assertions, and consider them dubious without further proof. If you are correct, then Sony should take care of this ASAP.
-
http://www.dyxum.com/dforum/forum_posts.as...=43626&PN=6 (http://www.dyxum.com/dforum/forum_posts.asp?TID=43626&PN=6)
Originally posted by qbic
Is it possible that "blobs" are only effects of data quantization at low bit resolutions? When you amplify low resolution signal you will get "blocks"
Your reply:
What you are saying is, that the sensor yields data enough only for let's say ten bits, and that gets "blown up". It's a bit thin for a top of the line camera.
Anyway, if that were so, then
a. the same phenomenon would be observable in the brighter areas as well,
b. the green channel would exhibit the same phenomenon,
c. the effect would not depend on the NR selection.
Now, from the current thread:
Your statement:
The green channels would be equally affected.
Me:
Please demonstrate this too
You:
Sony A900 - the effect of noise reduction on the raw data is doing just that.
Something does not compute here.
And of course I see you are not looking analytically on your own histogram. The histograms, by the way, are typical for level drops and posterization.
-
We sent to Sony engineers a brief summary of findings from practical shooting situations about a month ago, and detailed analysis based on shooting targets under strictly controlled conditions yesterday. I suggest once again Gabor to contact Sony directly.
-
Hi,
Here is a link to the "Lighthouse image" in RAW
https://rcpt.yousendit.com/664943793/8e93f5...e5b8adac573e337 (https://rcpt.yousendit.com/664943793/8e93f532c0d2d0de6e5b8adac573e337)
Erik, what about uploading this raw file?
-
Hi,
Here is a link to the "Lighthouse image" in RAW
https://rcpt.yousendit.com/664943793/8e93f5...e5b8adac573e337 (https://rcpt.yousendit.com/664943793/8e93f532c0d2d0de6e5b8adac573e337)
Please set you camera to ISO 320 instead of 200.
-
Iliah,
I have read your recommendation but I don't really understand it. I'm prepared to listen. Can you explain in what way 320 ISO would benefit? I also don't really see the benefit of intermediary ISO. A pointer to some other discussion would be just fine!
Best regards
Erik
Please set you camera to ISO 320 instead of 200.
-
Erik,
regarding the lighthouse shot:
1. if you want to see or demonstrate the camera's DR and noise capability, you should *always* set the Black to 0 in ACR; otherwise you are simply cutting off the most important parts. Pure black is always noiseless.
2. with blacks = 0, pls load the image in PS and take a look at the really dark areas, like the bottom right corner. Switch to channel mode and look at only the red and the green channel, alternating. You may want to boost the brightness before. You will notice, how much more detail is in the green channel.
I think Sony chose this way of noise reduction (in this case the implicite NR) with good reason:
a. the green channel, which is higher than the others in the majority of cases and thus less noisy, remains unchanged; it carries the details,
b. the other channels, which are more often noisy, get "smoothened", leading to posterization/lack of details, but the green will make good for much of it.
The downside is blotchiness, but only in the very shadows, and visible only on smooth areas.
-
Now, from the current thread:
Your statement:
The green channels would be equally affected.
Me:
Please demonstrate this too
You:
Sony A900 - the effect of noise reduction on the raw data is doing just that.
Something does not compute here
Well, I have no reason to convince you to actually *read* that paper instead of only talking about it.
And of course I see you are not looking analytically on your own histogram. The histograms, by the way, are typical for level drops and posterization.
It is simply amazing (and amusing), that you want to know what I see in the histograms, even without having read what I had to say about them.
-
a. the same phenomenon would be observable in the brighter areas as well,
What we are seeing appears to be the reverse of highlight clipping. ie - A typical sensor blows out the highlights and puts 255 for brightness. Game over. This appears to be that when there's too little data in dark areas, it goes off the deep end and adds in noise by mistake. Over-aggressive NR trying to reduce stuff that's really not there, maybe?
b. the green channel would exhibit the same phenomenon,
There are twice as many green sensors in a Bayer pattern, so the effect would be about 1/4 as pronounced. Which is what we see. There's enough data to keep the green channel "clean" at low levels. This points to the sensor either not being sensitive enough in low light or a problem with the internal software/converters.
c. the effect would not depend on the NR selection.
NR appears to be non-functional and a gimmick for anything under ISO800, at least according to the photos and examples that I have seen. Usually NR makes an enourmous and immediate impact when pixel-peeping. Yet we see virtually no change at all at ISO100, 200, or 400. What gives?
Perhaps the issue is simpler, then? Maybe Sony shipped the first batch of cameras with software that automatically has the NR set to maximum no matter what you're actually telling it to do? At least I hope it's as simple as this.
I'd like to also add that the clipped histograms are immediately obvious. The D200 and the A900 both have this chunky clipped pattern that jumps out at you. (as said, it's not only Sony that appears to have bad in-camera processing from time to time)
-
I appreciate all the hard work, Gabor, and I'm really interested in what's going on with the A900 at low ISO, but since Iliah seems to be implying that what we're seeing is caused by hardware, and, in the past you've said, "I have zero competence on the hardware area," I'm having a hard time with your assertions, and consider them dubious without further proof
That's ok with me; never mind, that the validity of none of my assertions in that paper depends on the question if the actions are performed in firmware or in hardware. As the matter of fact, I have not mentioned this aspect at all.
-
Hi Gabor,
The idea with the picture was not to demonstrate anything but to show a simple example of a very high contrast target with normal processing. Now that I have published the raw image anyone can interprete it to her/his own liking.
Best regards
Erik
Erik,
regarding the lighthouse shot:
1. if you want to see or demonstrate the camera's DR and noise capability, you should *always* set the Black to 0 in ACR; otherwise you are simply cutting off the most important parts. Pure black is always noiseless.
2. with blacks = 0, pls load the image in PS and take a look at the really dark areas, like the bottom right corner. Switch to channel mode and look at only the red and the green channel, alternating. You may want to boost the brightness before. You will notice, how much more detail is in the green channel.
I think Sony chose this way of noise reduction (in this case the implicite NR) with good reason:
a. the green channel, which is higher than the others in the majority of cases and thus less noisy, remains unchanged; it carries the details,
b. the other channels, which are more often noisy, get "smoothened", leading to posterization/lack of details, but the green will make good for much of it.
The downside is blotchiness, but only in the very shadows, and visible only on smooth areas.
-
Iliah,
I have read your recommendation but I don't really understand it. I'm prepared to listen. Can you explain in what way 320 ISO would benefit? I also don't really see the benefit of intermediary ISO. A pointer to some other discussion would be just fine!
Best regards
Erik
We use to say "try and see" The explanation as it seems is that if the ADC is fed with pre-amplified signal it copes better. From our experiments on 3 cameras we see IS) 320 to be consistently the optimum. We push further in raw converter as needed.
-
Well, I have no reason to convince you to actually *read* that paper instead of only talking about it.
It is simply amazing (and amusing), that you want to know what I see in the histograms, even without having read what I had to say about them.
We read your paper carefully, several times, and spent 2 hours discussing it.
-
This appears to be that when there's too little data in dark areas, it goes off the deep end and adds in noise by mistake
No, this is not the case. As I posted already, it is not black clipping.
There are twice as many green sensors in a Bayer pattern, so the effect would be about 1/4 as pronounced. Which is what we see
I don't see that, and you don't either. Btw, the proportion of the green pixels has nothing to with that aspect. (Though it does have to do with my suspicion posted above to Erik.)
NR appears to be non-functional and a gimmick for anything under ISO800, at least according to the photos and examples that I have seen. Usually NR makes an enourmous and immediate impact when pixel-peeping. Yet we see virtually no change at all at ISO100, 200, or 400. What gives?
A change between what? Between the omnipresent NR and the omnipresent NR?
Anyway, you see the same pictures again differently than I see them, for example in the samples rendered by ACR (http://www.cryptobola.com/photobola/SonyA900/SonyA900_NR_ISO100-800_ACR_Samples.htm)
Look at the fourth (the blue) and the last two strips.
Perhaps the issue is simpler, then? Maybe Sony shipped the first batch of cameras with software that automatically has the NR set to maximum no matter what you're actually telling it to do? At least I hope it's as simple as this
This is not impossible., but the difference should be obvious in the firmware release. Is there any new firmware out there? I have raw files from five cameras, all are version 1.
The D200 and the A900 both have this chunky clipped pattern that jumps out at you
I really don't know what you are referring to. I have not observed anything like that on D200 raw data, but my collection is quite limited.
-
OK, I'll check, I changed two om my presets to ISO 320. That said I personally never had any problem with noise at low ISO on the A900.
BTW, in some of your writings I have seen that you were of the opinion that the biggest difference between the Nikon D3x and the A900 that the Nikon captures more photons. If we assume that the sensels are similar I would presume that the difference would depend on the color filter array (better transmission on the Nikon) or the microlenses which my differ. What's your taken on this?
By the way I downloaded RPP and play around with it, I think I also made a donation but I'm not really sure. For the time being I'll stay with LR as main application in many ways it's the application I would have written if my time and knowledge has allowed me.
Best regards
Erik
We use to say "try and see" The explanation as it seems is that if the ADC is fed with pre-amplified signal it copes better. From our experiments on 3 cameras we see IS) 320 to be consistently the optimum. We push further in raw converter as needed.
-
I felt the same way about lightroom, Erik, but I've finally given it up completely. LR/ACR mishandles A900 like no other, and I figure there's no reason to be interested in the nuances of all these FF cameras if I'm using a poor conversion.
-
So, what are you using? Aperture?
Best regards
Erik
I felt the same way about lightroom, Erik, but I've finally given it up completely. LR/ACR mishandles A900 like no other, and I figure there's no reason to be interested in the nuances of all these FF cameras if I'm using a poor conversion.
-
So, what are you using? Aperture?
Best regards
Erik
I'm on PC, so I'm using RAW Therapee and C1. The bulk of my work is with C1, because, while it may give up a little to RAW Therapee, it is still much better than LR, and it's workflow is not too bad. Anyways, sorry to get off topic...
-
Dear Panelpeeper,
I have just read this thread and its attachments with interest, since I have an A900 and I have a possible explanation based on my experience in medical imaging.
In medical image processing we sometimes use a technique called "adaptive smoothing". This is a non-linear process that clips off outlying values from a stream of digital data. It is very helpful to fix occasional hardware blips where the ADC sends out a wrong value. It can also be used to reduce noise. In this process the value of the current reading is compared to the value of the previous and following pixel (or pixels) and the 2nd derivative computed. If the value exceeds the expected trend by a particular amount (say 1 or 2 std. dev.), then it is clipped to the expected value, say the mean slope adjusted value. In a uniform region such as with your color rectangles, the clipping would be to the mean of the adjacent values.
The shape of your histograms in the red and blue channels is highly suggestive of this kind of processing, as you can see the histogram width is reduced by clipping in from the tails. In medical imaging this works fairly well because the image is monochromatic, and the process selectively works on the noisier parts of the image (dark areas) where statistical noise is greatest. But in digital photography this would be a terrible method of noise reduction, since the intensity of the blue, red, and green channels, and hence noise, depends on color as well as intensity. Thus if imaging a red rectangle, the blue and green channels would be noisy even if the intensity is not low. This will cause blotchiness in the image as you have described, not only in the shadows, but also in regions with near saturation of red, blue, or green.
A second problem of this kind of noise reduction should occur at edges. An "outlyer reduction" algorithm, whether it tests 2nd derivative or something else can easily be fooled at a sharp edge and cause artifacts. This would be something to look for in images, but as far as I know, this has not been seen yet. Probably the best way to look for it would be with ISO1600 and NR set to high and an image of a wire against the sky.
In any case I love the image quality I am getting from my A900, and the noise cleans up very well in Neat Image. By the way, according to my understanding Neat Image works by filtering the image with the inverse of the FFT pattern that Gluijk has shown elswhere in this thread. But of course Neat can not remove the blotchiness and posterization caused by A900 noise reduction. Thankfully this is usually a very tiny effect, but I agree it would be good to reduce it even further.
==Doug
Here are the findings in condensed form:
1. Noise reduction Off, Low, Normal and High are identical from ISO 100 to 800, ON THE RAW DATA. They correspond something like NR Low @ ISO 1600.
In other words: all reviews used pre-NRed samples.
However, NR Off means almost Off with ISO 1600 (this explains the "sudden" increase of noisiness at ISO 1600).
2. The noise reduction affects mainly the red and the blue channels, much less the green. This is a very primitive noise reduction; it simply eliminates some pixel levels in the affected areas.
3. Due to the nature of this noise reduction, the very dark, noise reduced areas become darker. The consequence is, that not only the visual appearance but the noise measurement too indicates not only lower noise that it would be without NR, but that in deeper shadow, suggesting a greater dynamic range.
The magnitude of this shift is about 0.6-0.7 EV between the red/blue and the green channel. As the green channel itself is not virgin either, the shift must be even greater.
I estimate that in end effect the noise gets shifted and the DR "enhanced" by at least one full stop.
A side effect of the above is the non-linearity of the pixel values, and a WB shift in the low ranges: if the WB is correct in the "normal" ranges, it is certainly incorrect in the very dark range.
The nature of this noise reduction explaines the blotchiness as well. This will be appearant from the coming demonstrations.
-
In any case I love the image quality I am getting from my A900, and the noise cleans up very well in Neat Image
Doug,
let's make something clear: I have nothing for or against the A900; I am neutral even towards my own camera (in fact the only reason I have that is, that I have a much larger investment in lenses). I have no reason to make the A900 appear better or worse than it is; my interest is purely professional. I am analyzing raw images from dozens of camera models several hours long a day, and the A900 is something special.
Now back to the question if this is done in the A/D circuitry or later. Again this question is irrelevant regarding the validity of my assertions, but it is 9an interesting question.
It is simple to say that this is done by the A/D converter, but you have to think about details: what exactly is done.
1. Action on five levels, selectable from outside.
2. Performing a different action (this is the sixth version, much different from the others) on every second pixel: the A/D is column oriented, and every second pixel is green in every column. The green pixels are treated very differently from the others.
3. Watching the neighboring column's A/D converter: the blobs are not column or row oriented but twodimensional, context sensitive.
This is such an honorous task that Sony could have implemented a much better electronics, like some MFDBs have, still much cheaper. On the other hand, this is nothing special for the on-board processor. This may be hardware (i.e. not replacable by firmware version), but that is a far cry from being done by the A/D converter.
Those knowing how exactly such circuitries are implemented may have a different opinion, and I would not dispute that after they have considered the above points.
-
We read your paper carefully, several times, and spent 2 hours discussing it
Well... One of the statements in the abstract is
The green channel is affected to a much lesser degree than the other channels, with all ISOs and all NR settings.
Thus there is no contradiction between the paper and the statement
The green channels would be equally affected (namely if qbic's speculation were fitting).
-
Doug,
Now back to the question if this is done in the A/D circuitry or later. Again this question is irrelevant regarding the validity of my assertions, but it is 9an interesting question.
It is simple to say that this is done by the A/D converter, but you have to think about details: what exactly is done.
1. Action on five levels, selectable from outside.
2. Performing a different action (this is the sixth version, much different from the others) on every second pixel: the A/D is column oriented, and every second pixel is green in every column. The green pixels are treated very differently from the others.
3. Watching the neighboring column's A/D converter: the blobs are not column or row oriented but twodimensional, context sensitive.
Dear Gabor,
Yes, I agree that the kind of processing I suspect would be done by software. It is after all an algorithm. But I can see two dimensional blobs forming even if the processing algorithm is applied to columns only.
By the way, the black dots seen in Canon5DII would be easily removed by this kind of processing....
This may also be the reason that the blotchiness is prominently seen in GREEN grass (blue and red values low).
==Doug
-
I really don't know what you are referring to. I have not observed anything like that on D200 raw data, but my collection is quite limited.
I think it was the D20. One of the other cameras also was visibly clipping its histogram.
EDIT:
The green channel is affected to a much lesser degree than the other channels, with all ISOs and all NR settings.
If the camera is trying to overcompensate for low levels(ie - "I see black - oh no! not enough photons!") and failing miserably, then it would stand to reason that because there are as many green pixels as the blue and red combined, it would mess up less often on the green channel. The explanation Doug gave makes a lot of sense. What happens if the sensor has almost no data on one of the channels - to the point where the random noise itself is considerably brighter?
I could see it getting confused. It looks to me like it might be trying to NR the black areas and treat the noise as real data.
-
This may also be the reason that the blotchiness is prominently seen in GREEN grass (blue and red values low).
The blotches appear the best (though this should be calle the "worst") in areas, where
a. the green is relative low (this does not cause any trouble but weekens the mitigating effect of the healthy green),
b. either red or blue is not low, i.e. not causing trouble,
c. the other one is very low; those blobs will be translated by the rendering directly in blotches, if not sppressed by Blacks.
Now the effect is at the maximum. Example the blue and red patches on the color checker. See
Blotches and Blobs (http://www.cryptobola.com/PhotoBola/SonyA900/SonyA900_NR_ISO1600_BlotchesAndBlobs.htm)
Another aspect against A/D caussing this, though it is not an absolute proof: some pixel values appear only in the left half of the image. See the attached captures: the orangy vertical line is just in the middle of the sensor. The images are in pairs: the first of a pair with red, blue respectively green pixels only in the left half, the next pixel level is overall. (The cross like form is from the black separator between the color patches.)
Note, that this super A/D converter would have to take care of this separately for the red, green and blue (the "levels of skipping" are not identical).
-
If the camera is trying to overcompensate for low levels(ie - "I see black - oh no! not enough photons!") and failing miserably, then it would stand to reason that because there are as many green pixels as the blue and red combined, it would mess up less often on the green channel
LOL, you are attributing to the on-sensor A/D superhuman (or superelectronics) capabilities, combined with super mistakes.
The facts are not supporting this:
1. the green does not exhibit the blobs even in a black frame, where there is no difference between the intensity of the channels,
2. the red and blue blobs appear even at -5 EV, which is about 5-6 stops brighter than the black frame.
I could see it getting confused
Yes, I see that :-)
It looks to me like it might be trying to NR the black areas and treat the noise as real data.
This "black nonsense" (like "black clipping") is somehow firmly anchored in some minds. See #2 above: there is no black anywhere.
-
Still to the "black myth":
1. This raw file (36 MB) (http://www.panopeeper.com/download/SonyA900_ISO0100_CCC_DSC01194.ARW) is well exposed; take a look at it. The blue channel on the red patch is blobby; that at ISO 100. Load the image in PS and take a look at the red channel (blobby on the green and blue) and the blue channel (blobby for example on the yellow).
2. Look at following histograms, with ISO 1600. The first one is NR Off, the second is NR High. The red channel is not near black, still strongly affected, while the green is only slightly. (The green is only from one channel in both histograms, so that effect is shown in more clarity.)
-
Well... One of the statements in the abstract is
The green channel is affected to a much lesser degree than the other channels, with all ISOs and all NR settings.
Thus there is no contradiction between the paper and the statement
The green channels would be equally affected (namely if qbic's speculation were fitting).
Interesting that you do not see the contradiction between two of your statements; because I was not speculating.
Again, here you are telling me (not qbic) that your article demonstrates the green channel to be equally affected:
Your statement:
The green channels would be equally affected.
Me:
Please demonstrate this too
You:
Sony A900 - the effect of noise reduction on the raw data is doing just that.
More interesting that you ignore the fact that in daylight green channel registers much more photons. Repeat under the incandescent light and see the results.
But most telling is your refusal to send your study to Sony. One of the possibilities is that at this stage you must know that it is not worth it - the study is one big error. In this case it is hard to understand why you still you spread it. The other, even less pleasnt possibility is that you are not attempting to help Sony users here.
-
Hi,
Just for your information, the pictures I have taken were made under incandescent light. Two simple halogen lamps on either side.
Best regards
Erik
Interesting that you do not see the contradiction between two of your statements; because I was not speculating.
Again, here you are telling me (not qbic) that your article demonstrates the green channel to be equally affected:
Your statement:
The green channels would be equally affected.
Me:
Please demonstrate this too
You:
Sony A900 - the effect of noise reduction on the raw data is doing just that.
More interesting that you ignore the fact that in daylight green channel registers much more photons. Repeat under the incandescent light and see the results.
-
Just for your information, the pictures I have taken were made under incandescent light. Two simple halogen lamps on either side.
I'm not talking about your images, but about tests shots making grey step wedge. Minimum is newspaper shots as I descrbed.
-
Interesting that you do not see the contradiction between two of your statements; because I was not speculating.
Again, here you are telling me (not qbic) that your article demonstrates the green channel to be equally affected:
Your statement:
The green channels would be equally affected
I really don't understand your problem.
If qbic's suggestion was right THEN the green channel would be equally affected.
However, I have shown, that the green channel's processing is vastly different from the other channels.
Thus qbic's suggestion can not be right.
More interesting that you ignore the fact that in daylight green channel registers much more photons. Repeat under the incandescent light and see the results.
It is *totally* irrelevant, which pixels collect more light in which illumination generally; only the *actual light capture* counts. The attached histograms are from the red patch in an ISO 1600 shot, first NR Off, then NR High. The average of the red on that patch is 1.2 EV HIGHER than the green; still, the red undergoes a much-much greater change (between NR Off and NR High) than the green.
But most telling is your refusal to send your study to Sony. One of the possibilities is that at this stage you must know that it is not worth it - the study is one big error
I think it's time for you to get serious; joking has its limits.
Anyway, I wonder if you are really so naive as to believe, that Sony does not know this issue better. I am convinced, that this is not only known, but it is well-planned. In fact, it is a very good idea, as long as the blotches are suppressed. The customers are happy with their cameras, most reviewers have been fooled in publishing unrealistic figures, so what's the problem? Though I made a search and found an article (some printed photography related paper's site) from last year, saying that the raw data is massively NRed.
-
However, I have shown, that the green channel's processing is vastly different from the other channels.
Not so. You ignore different effective amount of light. Results in the green channels are different (btw you show 3 histograms instead of four), but the explanation is not different processing, but different captured signal. And we are not talking about noise reduction at high ISO here.
Sony engineers and management need to have feedback on the perception of the product. That helps improving. Your position here is beyond my understanding.
-
LOL, you are attributing to the on-sensor A/D superhuman (or superelectronics) capabilities, combined with super mistakes.
The facts are not supporting this:
1. the green does not exhibit the blobs even in a black frame, where there is no difference between the intensity of the channels,
2. the red and blue blobs appear even at -5 EV, which is about 5-6 stops brighter than the black frame.
This "black nonsense" (like "black clipping") is somehow firmly anchored in some minds. See #2 above: there is no black anywhere.
I didn't technically mean "black" so much as too low of a signal at the sensor. Maybe they've set the NR to a value where it's thinking "too dark - apply maximum NR" when it's still 5-6 stops before true black.
In any case, the on-board software looks to be bugged and set up wrong. That red/blue artifacting shouldn't be so bad. Maybe the green is fine but something about the sensor itself is mangling the red/blue channels and the NR is their attempt at "fixing" the problem - like adding a software micro-lens. Since it's not applied to all three channels, of course you'd get artifacts. Is there a way you can rule out Sony possibly applying a second level of NR to only one or two of the channels?
P.S. What this really reminds me of is over-amplification of a TV signal. You get artifacts, ghosting, and noise because you're also amplifying the noise as well.
-
I didn't technically mean "black" so much as too low of a signal at the sensor. Maybe they've set the NR to a value where it's thinking "too dark - apply maximum NR" when it's still 5-6 stops before true black
Fact is, that even NR Low affects already the fifth-sixth stop at ISO 1600, where the noise is only 15% (SNR > 6).
Maybe the green is fine but something about the sensor itself is mangling the red/blue channels and the NR is their attempt at "fixing" the problem - like adding a software micro-lens
Apparently this nonsense with "the sensor is doing it" won't go away.
Is there a way you can rule out Sony possibly applying a second level of NR to only one or two of the channels?
What would be the "first level"?
-
Fact is, that even NR Low affects already the fifth-sixth stop at ISO 1600, where the noise is only 15% (SNR > 6).
If it's true, then something is badly wrong here.
What would be the "first level"?
I meant that they are doing some fiddling and pre-processing with the data from the sensor before the NR is applied. Kind of like a hidden extra level of adjustment that's not accessible to the user. ie - "We screwed up with the sensor - quick - tweak it and kludge a fix in software before it ships!" Is it the sensor/chips/hardware or the software that's causing this? The data from the three channels is obviously not close - that green channel is just way off - like it's coming from a completely different camera.
-
"We screwed up with the sensor - quick - tweak it and kludge a fix in software before it ships!"
I don't think the sensor is doing anything wrong (except for not being as good as Sony wants it to appear). I find the method quite smart, though it starts too "soon", i.e. at noise levels, which should be left for the later processing.
Basically, the method of keeping the green intact (well, not really, but compared to red and blue) preserves much detail, and the noise will be much less due to the red and blue adjustment. However, there are several downsides:
- loss of details, when not the green is dominant,
- color blotches (but obviously this is not a widely seen problem, I guess the "Blacks" takes care of it),
- off colors; this is very easy to see, I am surprized that we don't hear many complaints.
The attached captures show a crop from the same setting, ISO 1600, one NR Off, the other NR Normal, one pair sharpened, the other not. NR was set to 0 in ACR. Look at the red details: this is the consequence of the red mutilation (I added some extra contrast in PS to enhance the differences). Note, that this these reds are in the fifths and sixth stop of the DR, i.e. not dark.
-
Thanks,
I'll check out a bit more! I like your processing. Highlight recovery seems to be better than in Lightroom. Would be interesting to see a full size JPEG or TIFF or some detail.
Erik
Yes, Aperture. I put your image through it and gave it some mild Photoshop touch-up afterwards. No noise reduction. I could squeeze a lot of information out of your ARW that held up pretty well when pushed a bit.
[attachment=12260:20090315_DSC01748.jpg]
-
Gabor,
I don't really think that Sony manipulates the images just to get better reviews. For one they get bad reviews anyway. Most reviews are about JPEG and there is no need to manipulate "raw" to get good JPEGs. I see that there are some issues with Sony ARWs at high ISO with blochiness, these can be caused by the artifacts you see or something else. I think you demonstrated quite clearly that the artifacts are also present at low ISO.
I appreciate your efforts even if I don't necessarily share your opinion of Sony purposely manipulating the images just to get god DR.
Best regards
Erik
I really don't understand your problem.
If qbic's suggestion was right THEN the green channel would be equally affected.
However, I have shown, that the green channel's processing is vastly different from the other channels.
Thus qbic's suggestion can not be right.
It is *totally* irrelevant, which pixels collect more light in which illumination generally; only the *actual light capture* counts. The attached histograms are from the red patch in an ISO 1600 shot, first NR Off, then NR High. The average of the red on that patch is 1.2 EV HIGHER than the green; still, the red undergoes a much-much greater change (between NR Off and NR High) than the green.
I think it's time for you to get serious; joking has its limits.
Anyway, I wonder if you are really so naive as to believe, that Sony does not know this issue better. I am convinced, that this is not only known, but it is well-planned. In fact, it is a very good idea, as long as the blotches are suppressed. The customers are happy with their cameras, most reviewers have been fooled in publishing unrealistic figures, so what's the problem? Though I made a search and found an article (some printed photography related paper's site) from last year, saying that the raw data is massively NRed.
-
I don't really think that Sony manipulates the images just to get better reviews
I don't believe that they are doing this for better reviews; that's a side effect, I would not even say that it is calculated.
As I posted, some reviewer found out the heavy NR on the raw data already last year. One does not need to go deep in the raw data; I presented samples from ACR prooving the NR effect, the details are only for the geeks. Thus the reviewer's oversight was not a calculable, except for DPReview, who don't understand anything of the issues past JPEG.
Btw, the apparent increasing of the DR through the darkening of the deepest shadows is not done in order to fool the measurements, it is inherent to the applied method. However, that should have been noted by everyone, who was *measuring* DR and noise instead of *prattling* about it.
-
As I posted, some reviewer found out the heavy NR on the raw data already last year. One does not need to go deep in the raw data; I presented samples from ACR prooving the NR effect, the details are only for the geeks. Thus the reviewer's oversight was not a calculable, except for DPReview, who don't understand anything of the issues past JPEG.
Are you finding the same issues with non-Adobe RAW converters? Could we see some examples with RAW Therapee, RPP, etc.
-
Are you finding the same issues with non-Adobe RAW converters? Could we see some examples with RAW Therapee, RPP, etc.
I base my findings alone on the raw data before raw conversion. The ACR samples demonstrate only, that these effects do appear in the converted image as well. I don't work with any other raw converter, but I think Erik would not mind that I pass some of his raw files, if someone wants to experiment on them.
The above posted crops showing the loss of detail in the red stoff are from Imaging Resources, one can download the raw files and interpret them at will. I am not authorized to use their images in publications; I hope they pardon the trespassing with the crops (anyway I did not put them in my HTML pages).
The final effect can be even worse with certain other converters without noise reduction, for ACR always carries out some NR even if you select 0.
-
As I posted, some reviewer found out the heavy NR on the raw data already last year. One does not need to go deep in the raw data; I presented samples from ACR prooving the NR effect, the details are only for the geeks. Thus the reviewer's oversight was not a calculable, except for DPReview, who don't understand anything of the issues past JPEG.
Btw, the apparent increasing of the DR through the darkening of the deepest shadows is not done in order to fool the measurements, it is inherent to the applied method. However, that should have been noted by everyone, who was *measuring* DR and noise instead of *prattling* about it.
How it is proved that the artefacts are result of noise reduction, and not of something else, like multiplication of data coming from red and blue channels accomplishe in digital domain? )
ACR is not a tool to measure anything, its processing chain is not intended for data analysis.
Darkening of shadows decreases measurable dynamic range and can't fool the measurements.
-
I base my findings alone on the raw data before raw conversion. The ACR samples demonstrate only, that these effects do appear in the converted image as well. I don't work with any other raw converter, but I think Erik would not mind that I pass some of his raw files, if someone wants to experiment on them.
The above posted crops showing the loss of detail in the red stoff are from Imaging Resources, one can download the raw files and interpret them at will. I am not authorized to use their images in publications; I hope they pardon the trespassing with the crops (anyway I did not put them in my HTML pages).
The final effect can be even worse with certain other converters without noise reduction, for ACR always carries out some NR even if you select 0.
Gotcha. I realize that your findings are based on raw data before conversion, but I don't believe what you're showing in ACR backs that up, because ACR is awful for the A900. Interestingly, I just saw this little website today, and, although I realize that DxO is far from the purest of raw converters, I thought that is was appropriate to post here, because the author says in reference to the A900 and ACR, "Dark areas often take on a blotchy appearance and it is difficult to increase shadow detail without turning the blacks to gray and ruining the contrast." Regardless, take this site for what it's worth: a900 DxO vs. ACR (http://www.brettdeacon.com/gallery.php?gallery=RAWConverters)
-
I don't believe what you're showing in ACR backs that up, because ACR is awful for the A900
You are misleading yourself if you believe that this is an issue with ACR. The details are destroyed by the noise reduction, no software will recover that.
Again: the green channel is supposed to carry the details in this schema. This does work as long as the green is dominant. I attach crops from the same images I posted above, but this time from Rawnalyze, no demosaicing, no conversion, this is the raw data. First NR Off, then NR Normal (NR OFF appears in the file name).
1. The green channel does not contribute to the details on the red cloth (the green is more than a stop darker than the red), but it contributes a bit to the muddiness.
2. The blue does not contribute to the details (it is 1.5 EV darker than the red), but it does contribute a lot to the muddiness.
3. The red carries all details - but the red is seriously damaged by NR Normal.
The last pair is the composite color image.
Now, do show please how much details DXO can extract from the NR Normal version. The raw files are at Imaging Resources (http://www.imaging-resource.com/PRODS/AA900/AA900THMB.HTM), look for AA900hSLI1600.ARW and AA900hSLI1600_NR_OFF.ARW.
Dark areas often take on a blotchy appearance and it is difficult to increase shadow detail without turning the blacks to gray and ruining the contrast
That's it! Though I have not seen the claim, that DxO solves this problem.
-
You are misleading yourself if you believe that this is an issue with ACR. The details are destroyed by the noise reduction, no software will recover that.
Again: the green channel is supposed to carry the details in this schema. This does work as long as the green is dominant. I attach crops from the same images I posted above, but this time from Rawnalyze, no demosaicing, no conversion, this is the raw data. First NR Off, then NR Normal (NR OFF appears in the file name).
1. The green channel does not contribute to the details on the red cloth (the green is more than a stop darker than the red), but it contributes a bit to the muddiness.
2. The blue does not contribute to the details (it is 1.5 EV darker than the red), but it does contribute a lot to the muddiness.
3. The red carries all details - but the red is seriously damaged by NR Normal.
The last pair is the composite color image.
Now, do show please how much details DXO can extract from the NR Normal version. The raw files are at Imaging Resources (http://www.imaging-resource.com/PRODS/AA900/AA900THMB.HTM), look for AA900hSLI1600.ARW and AA900hSLI1600_NR_OFF.ARW.
That's it! Though I have not seen the claim, that DxO solves this problem.
I was referring to the low ISO shots, not high. My mistake. That being said, ACR slaughters A900 low ISO shots as well, while other converters don't.
I realize that NR on ISO 1600+ should be left off. I guess my confusion is still in how you haven't really proven that the problem of A900 isn't hardware based like Iliah mentioned.
-
I was referring to the low ISO shots, not high
I have given plenty of proof, that this is happening with ISO 100 to 800 as well, but to a lesser degree than the above demonstration, which was NR Normal. As this can not be turned off, a direct comparison like above is not possible.
I realize that NR on ISO 1600+ should be left off
It IS off, if you turn it off (well, almost). The issue is not ISO 1600.
I guess my confusion is still in how you haven't really proven that the problem of A900 isn't hardware based like Iliah mentioned.
You are forcing me to repeat myself.
1. NONE of my assertions depends on this aspect; as the matter of fact, I don't care. The question is relevant only from the perspective if a change can be expected or not.
2. I don't see any need to prove that such a function is not realized in the A/D converter. As this would be extreme (a super-duper A/D converter but what for?), one needs to prove, that Sony wasted the resources for this nonsense (for it can be done much cheaper outside of the sensor).
-
Well, you have absolutely no proof that low ISO artefacts are due to noise reduction. You do not realize that ADC used in A900 is unstable at low values, introducing arbitrary error in deep shadows, one of the reason of this error being heat build-up. One absolutely needs to understand hardware issues to attribute artefacts properly.
-
I did mention in the past that it looked like over-amplification to me. Similar to what you'd see if you over-amplified a TV signal(at least the older analog ones). Like it's just creating random garbage and ghosting in the signal path and the NR is trying to deal with it. To me, the blotches and artifacts look too uniform, in a way, to be only something cause by the apparently crummy software.(which we can all agree has serious issues with the NR routines) WHERE in the signal path it's getting messed up and corrupted, I don't know. It could honestly be anything in the chain between the sensor and the software.
Every shot, they're there, and almost like a piece of wallpaper or an image etched on glass that's been overlaid on top of the sensor. The NR seems to alter it, but it's always present. This suggests another problem is also happening as well.
-
Like it's just creating random garbage and ghosting in the signal path and the NR is trying to deal with it
And this is happening at ISO 100 and 200, but not at ISO 1600? A really strange circuitry.
To me, the blotches and artifacts look too uniform, in a way, to be only something cause by the apparently crummy software
This is the first time I see a claim, that a mess is "too big to be created by software".
The NR seems to alter it, but it's always present. This suggests another problem is also happening as well.
Not at ISO 1600, NR Off. Although there are clear signs of adjustment even @ 1600 with NR Off, but the blobs disappear, taking the blotches with them from the converted image.
-
Interesting discussion thus far.
However, sometimes I get the feeling that in any such situation, one can get better results (assuming the intention is constructive) if one does not go ahead and create admittedly provocative threads in every popular photography related forum around, and instead, directly contact Sony, and ask for their explanation on the how/why something was done the way it was. There are extremely smart people in the team of over 200+ Sony (previously Konica-Minolta) engineers involved in the development of the A900, and I would not assume that I or anyone else I know, is automatically smarter than their collective knowledge.
Many of these things have input from the hardware side and also from the firmware side and without understanding the entire cradle-to-grave scenario and the inter-relationships between all these factors, one could be jumping to conclusions and providing fodder to the less knowledgeable, to diss a product unfairly. This is the internet after-all.
-
Interesting discussion thus far.
However, sometimes I get the feeling that in any such situation, one can get better results (assuming the intention is constructive) if one does not go ahead and create admittedly provocative threads in every popular photography related forum around, and instead, directly contact Sony, and ask for their explanation on the how/why something was done the way it was. There are extremely smart people in the team of over 200+ Sony (previously Konica-Minolta) engineers involved in the development of the A900, and I would not assume that I or anyone else I know, is automatically smarter than their collective knowledge.
Many of these things have input from the hardware side and also from the firmware side and without understanding the entire cradle-to-grave scenario and the inter-relationships between all these factors, one could be jumping to conclusions and providing fodder to the less knowledgeable, to diss a product unfairly. This is the internet after-all.
I think you are being naive to think that Sony engineers are going to be allowed to express their personal opinion on these issues raised by Gabor, or that Sony will admit details of their processes into the public domain, other than details already released for purposes of 3rd party design of RAW converters. At best, there will be an official Sony response dominated by marketing spin.
-
I think you are being naive to think ... or that Sony will admit details of their processes into the public domain... At best, there will be an official Sony response dominated by marketing spin.
It took a while, but finally Sony confirmed the NR in the A700 raw data A700 and issued a firmware update.
one could be jumping to conclusions and providing fodder to the less knowledgeable, to diss a product unfairly
I see the expectation is an official statement from Sony, if the blotches are blotches, if the loss of details is real and the color shift is not due to the inferiority of the monitor. Perhaps the answer is "those are perceptions only or artifacts caused by your monitor. If you were using a Sony Trinitron monitor, you would not see those phenomena".
-
And this is happening at ISO 100 and 200, but not at ISO 1600? A really strange circuitry.
I have an odd theooy/question to propose. What if the "ISO1600" is the actual native resolution/sensitivity of the sensor and they are artificially tweaking and boosting the gain to simulate lower ISO? So NR off at 1600 would look clean because it doesn't need NR, but turning up the gain as you get to lower ISOs would require more aggressive NR at each level to combat the increased background noise?
-
It took a while, but finally Sony confirmed the NR in the A700 raw data A700 and issued a firmware update.
They did not have to "finally confirm" about the NR in the A700. Sony had freely published that they were doing NR on RAW, when they released the A700. After being universally panned for that strategy, they came out with a firmware fix, that completely de-activated RAW NR in the A700 (via the "NR off" setting), along with a bunch of other improvements.
One curious factor in the Canon 5DII, is that Canon swiftly came out with a firmware fix, that masked the "black dot" phenomenon (not present at all in the 21MP FF sensor in the 1DSMKIII). Is that not a clear indication that the RAW in the Canon is very well cooked and that the "black dot" masking, was again a further cooking of the RAW ? Any thoughts on that ?
-
(not present at all in the 21MP FF sensor in the 1DSMKIII)
The 5D MKII and the Canon 1DS MK3 do not use the same sensor - the 5D MKII is a re-design - hence the reason it was and isnt an issue on the 1DSMK3 . ChuckWestfall wrote about this in one of his tech tips.
Looking at the file differences between a 5D MKII and a 1DS MK3 it does appear that the 5D MK2 is cooking more into the RAW. I wouldnt say one is any better than the other - they are just different.
-
I have an odd theooy/question to propose. What if the "ISO1600" is the actual native resolution/sensitivity of the sensor and they are artificially tweaking and boosting the gain to simulate lower ISO?
Your theories are getting more and more exotic. If the base ISO were 1600, then ISO 100 would cause burning out the top FOUR stops.
Think what is happening at ISOs lower than the base ISO: plain overexposure; for example the A900 overexposes by 2/3 EV at ISO 100, indicating that the native ISO is around 160.
They did not have to "finally confirm" about the NR in the A700. Sony had freely published that they were doing NR on RAW, when they released the A700
That's right. I looked it up; the issue was, that there was no "off" setting.
One curious factor in the Canon 5DII, is that Canon swiftly came out with a firmware fix, that masked the "black dot" phenomenon (not present at all in the 21MP FF sensor in the 1DSMKIII). Is that not a clear indication that the RAW in the Canon is very well cooked and that the "black dot" masking, was again a further cooking of the RAW ? Any thoughts on that ?
The questions are justified, even though this has nothing to do with the actions of the A900.
1. The black dots have been caused by "stray" pixels. This needs an explanation.
Canon's CR2 (and at least one Olympus) creates a raw file containing the pixel values with an offset. This means, that a "black" pixel has the value around 128, 256, 512 or 1024, depending on the model. One theory is, that the A/D increases the voltage by a certain fixed amount, in order to catch irate values (negative charges) as well. Probably the other CMOS sensors are doing the same, but are not passing on the data. (The Nikon D300 is a notable exception.)
In this setting there can be and are pixels far under the "black level"; they will be changed to zero by the raw processing. This is happening all the time; there are a few hundred such pixels in the very dark areas, usually with high ISO.
What was happening with the 5D2? Someone made a test shot directly before and after loading the new firmware version; these shots demonstrate the source of the black dots and the solution. Anyway, the cause of the black dots were very stray pixels: even with value 0 at the average black level of 1024, i.e. those pixel values represented -1024; they occured always directly at the right end of saturated pixels. These stray pixel values were overall in the negative range but concentrated at -1024, which indicates that they would be even lower but could not be represented.
This looked pretty much like overflow caused by sensor bleeding. I guess this is a limitation of the hardware; other sensors (like the D3/D700) create different phenomena caused by bleeding.
Now, back to the issue: the new firmware "puts these pixels into their context", probably by looking further to the right and using those pixel values. This is clearly "cooking" of the raw data; the same can be achieved in raw processing (and is done in DPP). I do not see any likeness to the noise reduction on raw data; I don't have any problem with *this* cooking (the problem is somewhere else: where the stray pixels arize). However, there *is* some other cooking in the 5D2 files, which I can not explain.
Looking at the file differences between a 5D MKII and a 1DS MK3 it does appear that the 5D MK2 is cooking more into the RAW
I am interested to see, how you came to this result.
-
While I find these shots very good, I think it is important to note, that your (or mine or anyone else's) judgement on the dynamic range of a scenery is of secondary relevance compared to the actual measurement.
I can't agree with that. Actual measurement is subject to a raft of false science and bogus measurements, in my opinion. Its the reason serious hi fi is not judged by measurement but by ear.
My judgment of the A900 is that it has very wide dynamic range. Not quite as wide as medium format, but wider than most dslr's. Having determined that for myself through my own observation and use, all else is irrelevant.
Quentin
-
I can't agree with that. Actual measurement is subject to a raft of false science and bogus measurements, in my opinion. Its the reason serious hi fi is not judged by measurement but by ear.
My judgment of the A900 is that it has very wide dynamic range. Not quite as wide as medium format, but wider than most dslr's. Having determined that for myself through my own observation and use, all else is irrelevant.
Quentin
I agree. Eye never lies.
What the eye can see, it is all that matter.
All the rest is good stuff for spending time in front of the PC.
-
My judgment of the A900 is that it has very wide dynamic range. Not quite as wide as medium format, but wider than most dslr's. Having determined that for myself through my own observation and use, all else is irrelevant.
Dynamic range through NR reduction (i.e. through detail loss) is not real DR. If you think it is, then I will give you a secret recipe to maximize your DR: apply to all your pictures a gaussian blur filter with 20 pixels radius. All noise will disappear at any ISO so your DR will be stunning. But please don't complain about loss of detail because that kind of DR is what you wanted.
BR
-
Dynamic range through NR reduction (i.e. through detail loss) is not real DR
There is more to it. This particular method of noise reduction reduces the average pixel values close to black, and that causes the appearance of higher dynamic range: not only that the noise is lower, but that noise appears in deeper shadows, where higher noise is expected.
Even the semi-professional measurements, like DxO, DPR, etc. have been fooled by this effect. Imatest is helpless in this situation.
Btw, I saw a link to a new review, at Digital Camera Info; see the Noise section (http://www.digitalcamerainfo.com/content/Sony-A900-Digital-Camera-Review-20324/Noise.htm), particularly following graphs:
(http://images.digitalcamerainfo.com/images/upload/Image/new/reviews/2009/SONY/A900/tests/SONY-A900-noise-nr-levels.gif)
(I have no idea, what the percentages mean.)
Notice, that "Low" over ISO 800 is the plain continuation of the line from 200 to 800, not "Off"? These graphs confirm just what I have proven, that the implicite noise reduction under ISO 1600 is comparable to "NR Low" at 1600.
The guy, who created this review does not understand his own graphs:
As shown above, noise reduction processing didn't kick in until ISO 800, at which point it provided a substantial improvement over results with noise reduction turned off
-
There is more to it. This particular method of noise reduction reduces the average pixel values close to black, and that causes the appearance of higher dynamic range:
What f-stop referred to saturation are we talking about here? Did you manage to quantify the shift in EV of the average signal value with respect to the expected value because of the NR? this can be calculated for ISO1600 (NR OFF vs others). Few levels in the deep shadows mean an important amount in EV.
The guy, who created this review does not understand his own graphs
lol
-
What f-stop referred to saturation are we talking about here?
I have no idea, nor do I understand what bearing that has on the subject.
Did you manage to quantify the shift in EV of the average signal value with respect to the expected value because of the NR? this can be calculated for ISO1600 (NR OFF vs others)
I do not make such cross-calculations, for they are based on unproven assumptions. I asked Erik to make parallel shots with a Canon camera to croos-check the exposure; if he is making some, I will be able to calculate the differences.
Few levels in the deep shadows mean an important amount in EV
Yes, that's the issue. Changing the average by two levels, later on even by one level can "add" a stop to the DR. Of course, this is all imaginary DR. This has nothing to do with Sony's trick; some cameras have relative clean "base noise", like the old 5D, simply due to the lack of levels. I always felt the 5D is far overrated; you don't see much noise between 0 and 1, do you?
-
I have no idea, nor do I understand what bearing that has on the subject.
If the exposure shift is only noticeable from let's say -10EV, we wouldn't care at all. Noise make pixels jump by more than 1 f-stop in such dark areas. If exposure shift is happening in -7EV for example, that's a different story and would definitively make DR measures wrong with a conservative SNR=12dB criteria.
I always felt the 5D is far overrated; you don't see much noise between 0 and 1, do you?
I think the 5D is a very good camera for steady shooting, working only in the higher f-stops through multiexposure. If I found one now at 1000€ I would go for it, but it has become impossible to find old stocks. I am a bit scared of the 20Mpx of the new model.
Its good behaviour at high ISOs made people generally consider it a low noise winner. But for tripod high DR applications, were noise in the deep shadows matters, the 5D was just an average camera. I don't think the lack of levels was really the problem, it was simply quite noisy.
-
Dynamic range through NR reduction (i.e. through detail loss) is not real DR. If you think it is, then I will give you a secret recipe to maximize your DR: apply to all your pictures a gaussian blur filter with 20 pixels radius. All noise will disappear at any ISO so your DR will be stunning. But please don't complain about loss of detail because that kind of DR is what you wanted.
BR
Guillermo,
That's true, and this is where Quentin's method of visual comparison of real-world images can be effective, provided identical scenes are used under the same lighting conditions. If Camera A produces noisier but more detailed shadows than Camera B, then this fact should be apparent in real-world images. If it isn't, then the fact that Camera A is noisier but more detailed, is of no consequence, although it may be of academic interest to some and it may well be essential information for those who are producing their own RAW converter or noise reduction program.
I'm reminded here of Jonathan Wienke's DR test target (downsized version below) which I downloaded over a year ago, took a few shots comparing the DR of my 20D and 5D, then spent hours arguing about the necessity of comparing DR at the pixel level as opposed to the scene level. My view was, and still is, that DR at the scene level is what's important to me, although pixel DR is of some academic interest.
For example, I find it interesting that, according to DXOmark tests, the SNR and DR of the 5D MkII and 20D are approximately the same, at the pixel level. In fact, the 5D2 is sometimes very slightly better in DR by amounts of approximately 1/4 EV, which are not significant in my view. What is significant is the substantially better SNR and DR of the 5D2 in normalised prints, or same size and same FoV images on monitor.
The pixel size of the 5D2 is very close to the pixel size of the 20D. It's interesting that, after a number of years of development between the release of the 20D and the 5D2, so little progress has been made in improving the SNR and DR of the pixel. I would deduce from the DXOmark results that any image taken with the 5D2 that was cropped to the same FoV one would get if one had used a 20D with the same lens, would be of approximately equal quality to the 20D shot. Whether or not this is true can also be determined by visual inspection of real-world images.
[attachment=12482:Jonathan...t_target.jpg]
-
That's true, and this is where Quentin's method of visual comparison of real-world images can be effective, provided identical scenes are used under the same lighting conditions.
I completely agree. The point is that Quentin didn't state he did any comparision camera to camera. He was just happy with the DR results of his A900 without comparing them to any other machine tested under the same scene. That's why I said he could be just enjoying an imaginary DR (I like this term Gabor introduced) which actually consisted in NR with detail being destroyed in the shadows.
-
It's interesting that, after a number of years of development between the release of the 20D and the 5D2, so little progress has been made in improving the SNR and DR of the pixel.
I suspect that this is a built-in limitation with the Bayer pattern itself. Beyond a certain point, you will need different technology, because software and chips can only do so much to clean it up.
-
What f-stop referred to saturation are we talking about here?
If the exposure shift is only noticeable from let's say -10EV, we wouldn't care at all
You asked about f-stop, but now I see you were thinking of exposure value. Quite a bit difference.
This depends on the unadultered noise, which depends on the ISO. It starts at about -8.8 EV with ISO 200; with ISO 800 it starts at about -7 EV.
I find it interesting that, according to DXOmark tests, the SNR and DR of the 5D MkII and 20D are approximately the same, at the pixel level
The "black noise" of the 20D (measured on the masked pixels) is almost twice as high than that of the 5D2 at ISO 100, and more than twice at ISO 1600. The effective noise of the 20D is correspondingly higher.
In other words: the DxO numbers are incorrect; there did occur some noticable development in the meantime.
-
The "black noise" of the 20D (measured on the masked pixels) is almost twice as high than that of the 5D2 at ISO 100, and more than twice at ISO 1600. The effective noise of the 20D is correspondingly higher.
In other words: the DxO numbers are incorrect; there did occur some noticable development in the meantime.
Well, that's interesting. Could this be the subject of another topic? Perhaps someone who owns both a 20D and 5D2 could post some sample comparisons of images with equal pixel count, employing the normal flawless technique of all good photographers.
-
I completely agree. The point is that Quentin didn't state he did any comparision camera to camera. He was just happy with the DR results of his A900 without comparing them to any other machine tested under the same scene. That's why I said he could be just enjoying an imaginary DR (I like this term Gabor introduced) which actually consisted in NR with detail being destroyed in the shadows.
Maybe Quentin had other dSLRs in the past, and in similar shooting scenes as the A900, never displayed the wide DR range that the A900 displays ? Just a thought.
I personally am amazed at the level of detail the A900 retains, down into the deep darker regions of the image, while not blowing out the highlights. Maybe the "Sony trick" of expanding the DR, is working, regardless of our "pixel-peeping" assumptions/findings ?
-
Maybe Quentin had other dSLRs in the past, and in similar shooting scenes as the A900, never displayed the wide DR range that the A900 displays ?
I envy those photogs who have such ability; they don't need any metering, neither auto nor external, for they simply know or see the required exposure.
I personally am amazed at the level of detail the A900 retains, down into the deep darker regions of the image, while not blowing out the highlights
I have no reason to shake your feeling or belief, but I do wonder how you want to know this, when
a. you don't (can't) know, how dark those regions are if the camera changes the intensities,
b. you don't know how much detail is really captured on sensor level, for you see only what the camera passed in the raw file.
See post #137.
Maybe the "Sony trick" of expanding the DR, is working, regardless of our "pixel-peeping" assumptions/findings ?
The Sony trick is not in expanding the DR but in creating the impression of an expanded DR. However, I do not think that this was intended; it is a side effect of the method.
As such, the unintended trick is a success, as your example proves it.
-
Perhaps someone who owns both a 20D and 5D2 could post some sample comparisons of images with equal pixel count, employing the normal flawless technique of all good photographers
That's easy. Someone posts a shot of his/her cat, by a 5D2 and I post a shot of my dog made by the 20D. Then the normal flowless technique of all good photographers will immediately show, which sensor has higher noise on the pixel level.
-
I completely agree. The point is that Quentin didn't state he did any comparision camera to camera. He was just happy with the DR results of his A900 without comparing them to any other machine tested under the same scene. That's why I said he could be just enjoying an imaginary DR (I like this term Gabor introduced) which actually consisted in NR with detail being destroyed in the shadows.
I've used quite a few different dslr's so I have my own personal baseline for comparison (is there any better?).
I honestly can't be bothered measuring and "proving" one way or another which camera has better dynamic range or lower noise. You get a far better feel for these things though use. The observations of others are of course useful and interesting, but ultimately they take a back seat to your own experience.
Thus, a camera I though clipped quite suddenly was the Olympus E3. Cameras I found which had excellent dynamic range included the Kodak 14n, the DCS Pro 760C, the Nikon D700 and the Mamiya ZD. And so on...
Tested against these and other cameras I have owned, used or tried, I remain of the view that the A900 is very good. I can't compare with Canon as I have an irrational dislike of the brand
YMMV
Quentin