This content has been marked as final. Show 37 replies
[ Before reading this post, please make sure you have read the one immediately preceding this! ]
First off, apologies, as this is quite a long post. System info: MacBook Pro C2D (3Gb RAM), OS X v10.4.9, Photoshop CS3 with ACR v4.1, Lightroom v1.0. Tests done with Canon EOS-30D & Powershot Pro1.
While there are a lot of things I really like about the image processing in the new version 4.1 of ACR, like the new optional edge-defringing, the more flexible sharpening controls (with no black halos around specular highlights!), less white-speckling in dark areas of higher ISO shots etc., I have come across a very disturbing fact which I believe has already been noted by a few others as well.
The fact being that v4.1 of ACR is now seemingly applying a certain degree of luminance noise-smoothing to raw files that is camera and ISO specific and, as far as I can tell, cannot be turned off. In my humble opinion, this is a bad bad thing! ACR has been known in the past as having one of the most "hands-off" approaches to raw image processing and this new smoothing effect is going against the grain! Pardon the pun... :-)
I first noticed this when I was processing ISO 50 files from my PowerShot Pro1. Being a P&S, even at ISO 50, there was a certain amount of noise texture which I found rather pleasing but with ACR v4.1, even with all noise reduction and sharpening set to zero, images that I post-sharpened in Photoshop had a certain watercolor-like look, somewhat similar to what, for example, Genuine Fractals might give one upon upsizing an image. When I went back to the Lightroom processed versions, which is using the older raw rendering engine, I did not see this effect and indeed, there was a dramatic difference between the two.
I had extensively tested ACR v4.1 with ISO 100 files from my EOS-30D and 20D, and did find that with certain settings, the new sharpening controls would render a similar watercolor type look when one is viewing files at 100%, but when sharpening was dialed right down and the image was post-sharpened in Photoshop, the differences between the old and new ACR, while visible, were not too dramatic and I did actually prefer the slightly cleaner rendering of skies and shadow areas in v4.1.
However after seeing what v4.1 ACR did with my Pro1 files, I decided to look at some ISO 1600 shots from my 30D and compare them to the previous version of ACR. Whoa! What a dramatic difference! There seems to be a considerable amount of luminance smoothing going on, even with all sharpening and noise reduction controls at zero, which wipes out subtle surface textures and can make edges of fine detail sort of "blend" together. Please do note that I realize that on anything other than the biggest prints or enlargements (or viewing at 100% zoom), one probably would not notice the effect of this new image processing, however...
I primarily do landscape photography and for me, preserving every last bit of texture detail is very important in making an image look natural. I much prefer a more "film-like" rendering of a raw file, even if that means a bit of graininess, rather than a smoothed image that looks artificial and more processed. In addition, this new image processing in ACR v4.1 limits the amount of post-sharpening that can be done to an image since too much sharpening starts to bring out edge-artifacts where ACR has made "decisions" on what was legitimate detail and what was sensor noise.
What I will say, is that this new processing in ACR v4.1 actually seems very sophisticated, and reasonably benign in many cases, low ISO digital SLR shots for example, so for certain types of images and applications, it may well be a very good thing, however certainly not for everything. So this brings me to my request...
I humbly request that the developers of ACR put in the option to turn any and all noise smoothing off in the next version of Camera Raw. Just a checkbox in the Detail Pane labeled "Preserve All Texture" maybe, or have luminance smoothing truly at zero when the slider is at zero. As I mentioned, I think the new processing might be preferable for certain types of images and workflow, so I don't want to completely go back to the old system, but I really do want the option to preserve as much image detail as possible when needed. I expect that there are many pros out there that will agree with me.
As a bit of background, I have been working in the professional imaging industry now for well over 10 years, having started off back in the days of the Kodak DCS-420 and Nikon E2. I manage the digital imaging department in a high-end professional camera shop and have extensive experience with all manner of digital SLRs, medium format digital backs, film scanners and so on.
I always have been a strong proponent of shooting raw and, in the past, have always felt that Adobe Camera Raw gave, by far, the best looking rendering of raw files when it came to detail and a general lack of demosaic artifacts. I have experience using a large number of other raw conversion programs available (Capture One, Raw Developer, Canon DPP, Nikon Capture, Hasselblad FlexColor, SilkyPix, LightZone, Aperture etc.), and while I haven't always found the color rendering to be best in ACR, as far as image detail and creating a clean, natural looking file that doesn't look "digital", I have always found ACR to be absolutely top-notch. Color rendering one can fix in post... however when a raw converter does detail destroying smoothing, creates nasty demosaic artifacts that pop out when sharpening etc., well that is something than cannot be fixed or recovered in Photoshop afterwards! Well, at least not without an inordinate amount of work or trying to mask the effects by adding noise etc.
In fact, a number of revisions ago (maybe v3.4?), Adobe subtly changed the raw rendering of ACR and noted at the time that the demosaic algorithm had changed in order to create slightly sharper looking images. At that point, I already felt that a slight step backward had been taken (with respect to generating the most "natural" looking digital images), however with judicious adjustment of my post-sharpening technique, I was able to accommodate that change in my workflow and I think many others did not even notice it. This new version of ACR, however, has gone too far IMO...
Since I have recently dedicated myself to using Adobe Lightroom, I am also quite concerned that the next version will have this same new raw rendering engine. After recommending LR to many other professional photographers, I am now worried at this development since with a new version of LR, there will probably not be any way to downgrade the raw engine as one can with the plugin architecture of Photoshop.
I may not have time to check this forum on a regular basis, so if anyone from Adobe would like further clarification regarding my opinion on this matter, please feel free to email me directly. An email link can be found on my personal website:
or just email "mander" at the above domain.
Thanks for letting me rant on about this!
Please come back to the forum to look for replies.
Email responses would defeat the purpose of the forum, which is for everybody to benefit from all questions and answers, not just the original poster.
Additionally, we are all users just like you, just trying to help each other out. You are not addressing Adobe here.
Thanks for the note. You are, of course, correct. I actually should have edited the last part of my original post since I did notice, after I had written it, that one has the ability to be informed via email of postings to the discussion thread.
Well, I just installed Lightroom v1.1 and indeed, the same smoothing is happening in high ISO files from my Canon 30D. Argh.
Opened a support case with Adobe but based on the first response, I don't think the person who replied really understands what it is I am talking about so far.
Time to give feedback to the Lightroom team as well to see if I can make them understand the issue better.
Might have to downgrade ACR and Lightroom again... but that is also frustrating given that there are indeed many real improvements in ACR v4.1 and especially in LR 1.1.
Mike, I guess I have to agree with you. Just did a simple comparison of a 20D file at ISO 1600 with all sharpning and noise reduction set to 0. ACR 4.1 looks different than previous version. Looks like some kind of noise reduction is going on and if I zoom in VERY much it starts to look painterly in comparison to previous versions. Also edges in the picture tend to look very harsh. BTW,Mike, you might have a look at the following discussion:
I just want to add my voice to yours. Like you, there are a lot of things that I like about the change in algorithms in LR 1.1. On the other hand, the results I get at higher ISO (say, ISO 1600), while they show considerably less high frequency luminance noise than they used to when I used 1.0, display artifacts that could be compared to a "watercolor" or "painting" effect and look unnatural, maybe more so than the grain I used to have.
This is on a Canon EOS 5D, and my only gripes is that this effect is visible even when no noise reduction should be applied (both NR sliders to zero). It definitely leads to reduced detail in hair, in my opinion, and seems similar to noise reduction algorithms that are used in some digicams. I would prefer by far to have an option to get an image without noise reduction applied at all, in cases where I want to apply more sophisticated noise reduction myself.
[A version of this message originally posted to the Lightroom forum thread titled "NOT happy with image quality of Lightroom 1.1" to help answer someone's question as to whether the smoothing in LR v1.1 is equal at the same ISO from different cameras. Since ACR v4.0 and LR v1.0 have the equivalent raw engine, as do ACR v4.1 and LR v1.1, I thought this a valid comparison on that forum]
Okay, I have assembled a file with numerous comparisons here (15.78 Mb):
Basically I have the same image from my EOS-30D at ISO 1600 and from a Fuji S5 Pro at ISO 1600. One set of images was converted with Adobe Camera Raw v4.0 and the other with v4.1.
I converted the Fuji S5 file in 6.2 Mp mode, so ACR is not upsampling the file. Noise gets clumpy looking in the 12.1 Mp mode on Fuji raw files. An additional note is that ACR v4.0 rendered these Fuji files *way* darker than ACR v4.1, presumably due to the fact that the S5 support was not yet tweaked in v4.0. What I did is add 1 stop of exposure compensation in ACR v4.0 and then tweak the levels in PS to make the two layers visually as close as possible. On the 30D layers, I did not have to make any exposure adjustments between ACR versions.
On all images, I converted them in ACR with sharpening slider(s) at zero, luminance NR at zero and chroma NR at the default of 25 on the ISO 1600 samples and at 5 for the 30D ISO 100 sample. In all cases I did one pass of the "Sharpen" filter in PS after cdonversion to emphasize the differences and "pop" the grain a little more. Even without the sharpen pass though, the differences are obvious.
There are two groups of layers, labeled ACR v4.0 and ACR v4.1. Zoom the image in to 100% or higher and turn the top ACR v4.1 layer group on and off to see the differences.
On the Fuji S5 sample at the top, notice that there is effectively no difference to the luminance noise at all, essentially every pixel is identical, edges look identical etc. There is only a slight change in color/contrast due to the differences in rendering of the two ACR versions. I didn't bother including it, but the Fuji at ISO 100 is the same... no difference in smoothing between ACR v4.0 and ACR v4.1.
However, on the EOS-30D ISO 1600 samples, you can see a fairly dramatic difference as there is an obvious smoothing of noise. To my eye, the ACR v4.0 images seems more natural whereas certain areas on the ACR v4.1 samples look somewhat artificial. This isn't the "worst" example of smoothing that I've seen, but is a useful comparison since I have the identical shot on a Fuji S5.
The differences between v4.0 and v4.1 with the EOS-30D raw sample at ISO 100 though is almost imperceptible, and there is only the very slightest hint of any sort of luminance smoothing... certainly not enough to worry me in any way! So obviously there is an ISO dependent smoothing going on with the 30D but not with the S5 Pro.
I wish I had the same comparison with a D2X or D200 as well, and depending on my time, I may shoot some samples. However as I mentioned before, in my limited testing of the D2X (and of the Leica M8) I did not see any ISO dependent differences between the two ACR versions.
Here are links to the original raw files if anyone wants to try this themselves:
Fuji S5 Pro zipped ISO 1600 raw (8.96 Mb):
Canon EOS-30D zipped ISO 1600 raw (8.31 Mb):
Thanks, Mike, for all the time you put in to your investigation. Your results strongly suggest that indeed ACR4.1 is applying an extra baseline luminance noise reduction based on the ISO, but only to Canon raw files (or perhaps only to certain Canon models).
I apologize if this is an inappropriate request (I'm still pretty new to these forums), but would it be possible for someone on or close to the ACR team to post a statement about this issue? Specifically, which of the following statements is true:
A. It's a bug that only certain cameras have this extra noise reduction. We plan on addressing by a.)removing it, b.)making it selectable, or c.)making it apply to all cameras in release x. (or I suppose d.)we're not addressing it.)
B. This feature was deliberate and the reason we did it was x.
Any clarification would be very helpful.
BTW, there's another interesting comparison that you can do with Mike's latest sample. If you use either of the ISO 100 crops as the reference for how much detail was actually seen by the camera, you should be able to swag how much of the detail in the 1600 crops is real vs. how much is inferred by the noise. The point being, when you compare the ACR 4.0 and 4.1 1600 crops, you perceive a loss of detail, but by looking at the 100 crops you can try to figure out if any of the lost detail was actually seen by the camera in the first place.
For instance, look at the 4.0 100 crop and find a spot with a lot of detail. Now look at the same spot in the 4.0 1600 crop and try to get a feel for how much of the detail that's actually in the 100 crop you can see in the 4.0 1600 crop. Now compare that same spot between the 100 crop and the 4.1 1600 crop and try to get a feel for how much of the detail that's actually in the 100 crop you can see in the 4.1 1600 crop. Now that you have a feel for each, does the 4.1 1600 crop have any less detail than was actually in the 100 crop?
To my eyes, both the 4.0 and 4.1 ISO 1600 crops preserve about the same amount of detail that was actually seen by the camera. But, as someone mentioned earlier, this might be a bit academic, as the extra noise in the 4.0 1600 crop makes you perceive that there's more detail (real or not) and isn't it the perception of more detail that actually counts? And of course, this is pixel peeping in the spirit of "knowing your tool," as none of the differences we're looking at here would be perceptible on a 4x6. Larger prints, not so sure yet...
Thomas Knoll himself, the creator of Photoshop and ACR monitors the ACR forum regularly and, from time to time, replies with very brief but definitive notes.
He'll see your post(s).
[This cross-posted to the Lightroom forum too]
As I mentioned in my previous post I believe, that particular example is not the best (worst). I have seen more "destructive" baseline NR in other shots, but with those I do not have another camera's shot to compare with. I am not necessarily suggesting that the NR being performed is bad... on the contrary, it seems generally an "intelligent" algorithm compared to some I've seen. However in some instances it does indeed seem to obliterate fine detail. Sure, this detail is at the borderline of resolution but to my eye, the older raw conversions yielded a more natural look.
In any case, the real issue is that when I put a NR slider to zero, I generally expect the software to give me ZERO, no matter how good the implementation is. This is what I liked about ACR in the past, compared to let's say Capture One, where even with all the sliders at zero, there was a huge amount of NR being done on longer exposures or higher ISOs. Not sure if Capture One still does that, but that neither here nor there.
Zero should be zero, and it is indeed seems to be zero on other cameras, but not on Canon raw files. That is my issue.
For all those who like the baseline NR, just have it at the current level when, say, the slider is at 5 (not zero), and all users will be happy. Whether or not there is any more real detail without NR is a moot point. If I wanted to use Noise Ninja, for example, to post process, then I would rather start with an image without any baseline NR applied.
Until we get a statement from an Adobe engineer, I can only speculate:
1) Is it a bug on Canon raw conversions?
2) Is it a bug on the Mac platform but not PC? No one has answered that one yet, I don't think...
3) Is there a Canon user on the ACR team that really likes the baseline smoothing? ;-)
4) Are the Canon files being processed as intended but the others aren't, so is it a bug in the opposite sense of point #1?
5) Did Adobe think that since Canon users bask in the knowledge that their high ISO files are cleaner than everyone else's, they wanted to make absolutely sure there were no complaints about noise? ;-)
6) Did Canon pay them to do it that way?!!! :-) :-)
Anyway, some valid possibilities and some not so much of course. I hope we get some sort of statement soon. If I hear anything definitive from my tech support case, I will post to this forum - and the Lightroom forum.
I have spent far too much time on this analysis in the last little while. I will now patiently wait for some real info. Luckily I am very happy with the ISO 100 results, and for the vast majority of shooting I have done lately, LR v1.1 and ACR v4.1 will work beautifully.
Just a couple of questions to help me understand the thrust of your questions:
1. Did you calibrate your camera(s) to ACR?
2. Do you find the same objectionable results on properly exposed images, or only on substantially underexposed images like your two samples?
Thanks in advance.
For my part, ACR has been calibrated for the cameras I'm testing. I took shots of a colorchart and used the Fors script.
The smoothing is just as noticeable on well-exposed shots. Exposure doesn't seem to impact the amount or end result of the automatic baseline luminance smoothing that's being applied to seemingly only Canon cameras.
Not sure I understand question 1? If you mean using one of the calibration scripts in conjunction with a MacBeth chart, well that I've tried. Found the results terrible. I am not interested in colorimetrically correct (arguably) photos for most of what I shoot - I used to love Velvia when shooting film, and that sure is not accurate either! So no, while I have done such a calibration in the past, I am not using it now.
Regarding question 2, I would say that both examples I have posted from my 30D are essentially good exposures, given the subject matter and contrast, although that is not to say I wouldn't tweak them a little. Here is a link to some screenshots from ACR, with highlight/shadow clipping warnings on and controls default/zeroed. Note the histograms... not a whole lot of room to move the exposure up on the outdoor shot without washing out the sides of the boats and the sky and with the statue in Vegas, I am already clipping the highlights to the max that I was comfortable with.
Screenshots (less than 250 Kb each):
However, the smoothing does not seem to be affected by the overall exposure. I have to emphasize this simple point again though: the NR slider at zero should be zero! It is on other non-Canon cameras I've tested and that is entirely my point. How bad you think my exposures might be has no bearing on that argument in the end.
Oh and if you are talking about my cluttered workshop example with the Leica M8, well that was underexposed on purpose since I was running through the ISO's as a test of the camera's shadow noise and detail levels. : and the M8 did not have the smoothing issue in the first place.
It's possible to try this on a PC if my acquaintance has finally upgraded but not installed ACR 4.1. I may be able to backtrack here, but at the moment, the only way I know is to install CS3 on a second computer.
We have access to images made on a 20D and a Nikon D80, which we shot together, or reasonably close together. We certainly can set up a couple of shots.
I probably will not be able to really attack this for at least a week. There are data taking traps here which have to be considered when doing this type of comparison.
One thing I have noticed, and this goes back to comparing the 20D to the Nikon 8400, is that the Canon is intrinsically smoother than either the 8400 or the D80, which, for my taste, the D80 is preferable as I can always smooth if I want to,as I did last night in a shot of some rocks in water.
I looked at DSF0462 and, if you consider prevailing opinion that one is to expose to the right on a histogram, I had to move the Expose slider to +0.80 to emulate the increase in exposure to do so. But in general the histogram as is looked pretty good for the impression I was getting of the image values.
Okay now this is interesting. Should have thought to try this earlier. Edit the EXIF iso-speed field down to 100 on an image shot at 1600. Guess what? ACR v4.1 clearly shows a difference in baseline smoothing (noticeably less - surprise, surprise!) on the same image when it is fooled into thinking it was shot at ISO 100 and not 1600.
I have to head off to a birthday party at the moment but will do some more testing tonight and post the samples as before. ACR v4.0 versus v4.1 and Canon 30D versus Fuji S5 Pro both with ISO 1600 converted normally and also with ACR fooled into thinking they are ISO 100 images. Will be interesting to see what happens...
I have spent too much time already... but I am intrigued.
A side note: you can't trust ACR's (or other raw processors') histogram regarding overexposure, because they show the result of WB application. Change the WB and you can get rid of the overexposure (or you may believe so), or you can "create" one. That is far from the truth, namely from clipping the raw pixels.
Check the Lightroom thread:
Thomas Knoll has cleared things up! It's not "Canon only." It's "Bayer array sensor only." What we're seeing are new demosaicing algorithms that have been implemented for cameras with Bayer array sensors, which includes at least Canon and maybe other manufactures as well. But, it turns out, not Fuji because they do not have Bayer array sensors. This means that Mike's samples are showing behavior exactly as expected.
Since this is the effect of a new demosaicing algorithm instead of some extra baseline luminance smoothing, this also means that there is no way to turn it off, short of going back to the older noisier algorithm. If you want more noise to infer more detail, you'll have to add it!
>If you want more noise to infer more detail, you'll have to add it!
Noise may "infer" details or it may not, but removing the noise removes details as well.
Noise removal should be left to the discretion of the refining post processing. For example one may want to apply noise removal only to certain areas.
The goal of recording raw images is to gain a higher level of control over the image processing, than one can have with in-camera conversion. Indiscriminate "enhancements" are going just in the opposite direction.
It's the "dumbing down" phenomenon creeping into graphics.
Adobe can start to claim the "You Push the Button, We Do the Rest" moniker.
Check out my reply to Thomas in the other forum. The short answer is that this smoothing also does not occur on the Leica M8 or on a Nikon D2X, and they both use a standard Bayer array sensors AFAIK. His explanation doesn't make sense to me so far...
I just tested a D2X ISO 1600 file again to be sure. Will do some more testing later tonight and/or tomorrow.
I suggest you to make a comparison by converting the raw file into DNG; pls post your finding.
Just curious: what will converting to DNG tell us that Mike's previous method of comparing ACR 4.0 and 4.1 won't?
It is pure speculation; there is no point to go into details, except if the noise removal is not happening with DNG.
I actually am importing and converting my Canon files to DNG in Lightroom (but also backing up the original CR2 files thank goodness!) and I have compared both CR2 and DNG files in the two ACR versions - no difference in behavior that is obvious to me.
This evening was spent processing a bunch of images of guests I photographed at the birthday party I attended. I used Lightroom and numerous shots were done at ISO 400-800. For the purpose of these images, I have no real issues but I really don't like what I am seeing. Out of focus backgrounds have a weird smoothed and almost posterized look due to variations in NR. Not bad enough that any non-photographer will really notice, so for these shots I will live with it. The outdoor shots at ISO 100 or 200 are fine, but then indoor ones in lower light, and thus higher ISO, do look a bit strange.
Anyway, no time today for more comparisons. Will try and get to that tomorrow.
I have started a new thread titled "ACR v4.1 vs v4.0: baseline high ISO NR - Part 1" in the ACR forum with my final analysis, so to speak, here:
The posting was so long I had to split it in two and the first reply to my original post contains links to the samples.
create a custom preset in LR all set to zero then apply this preset when importing, give this a try what is causing this is that ACR is applying a sharpening to the images
would like as well to See Acr fixed to be 300 dpi not default of 240 DPI
or at least read what the camera is doing
real issue since Lr does not allow us to change this in LR itself and lossing 60 dpi is not good on high qual images
You don't lose anything. Check the file size in actual dpi. 240 is probably a convenience for Epson printing. That's all I can imagine it's number choice for.
sigh... You are not losing anything with 240ppi vs 300ppi.
The number of pixels is information, the ppi is just some metadata.
I said that wrong, Pierre. Not ppi, but actual pixel count is what matters.:D
>would like as well to See Acr fixed to be 300 dpi not default of 240 DPI >
That is very easily accomplished:
Just open a file in ACR and click on what looks like a blue URL Link at the bottom of the ACR window.
Change the parameters in the dialog which then opens to whatever you like; and close the dialog.
Next time that you open an image in ACR, your new parameters will show in the "blue link".
I have a PC, and Yes, the bug is there. I have the same problem with my Canon Raw files. I don't like, since I m a wedding photographer and high ISO shots are common to my work (dim lights in churches/temples and venues alike force me to shoot at high ISO, even using flash). It is NOT a Mac only problem.
From the ReadMe file of ACR 4.2 released tonight:
Noise reduction adjustment for all cameras with Bayer Pattern sensor: The base point noise reduction applied at the demosaic stage of raw processing has been reduced. The resulting effect is that images with zero luminance noise reduction applied in Camera Raw 4.2 will contain more noise than the identical settings in Camera Raw 4.1 but less noise than identical settings in Camera Raw 4.0.
Lawrence, I was answering lowolf, I know that you know (and we cross-posted)
I'm VERY disappointed, because 4.2 works like 4.1 for me. There are topics like
In the latter one it really seemed that Adobe had noticed and nearly accepted the numerous complaints about new "quality".
Now I'm looking on my ISO 200 (at 300% zoom) image in ACR4.2 and it's just the same problem again (Canon 20D). Little patches, maze-like artifacts, watercolor look, less detailed, radius slider really funny (it probably sets how large details it should be killing-multiplied by 1.5).
And again, no possibility to turn this cheap-looking processing off!!!!!!! It makes me mad, sorry.
Adobe, please have a look at 4.0 conversion quality and detail and then at 4.1 & 4.2.
I'm going to reinstall 4.0 once again now :(