Here's a 5D II shot I processed in LR 4.1 RC2, C1 6.3 and DPP 3.11.26. It's a bit of a torture test as I zoomed in on the cave and tried to pull out some detail. I didn't push this too far as I wanted some semblance of contrast, but in LR I lifted shadows by 65. I tried various settings with and without NR and decided to leave it on. The DPP seemed to have less noise but I think it must be doing NR when it's set to zero. No matter what I tried however DPP could not extract the very fine detail of the C1 or LR, the difference is quite prominent. I even tried the DLO plugin but it didn't make much if any difference.
One thing I'll say about LR is this took me about a tenth of the time it did in DPP and was also much quicker than C1 to get something similar.
speedmaster20d wrote:
I am not sure if I agree, to my eyes DPP has just as much detail if not more, look at the stripes on the shirt. The difference in sat. is just the default profile, you can increase it.
I don't like LR rendering t all, it generates coarse grain with inferior fine detail IMO...
Here is one example that shows this.
If you are interested here is a field example ISO 3200, 5D3 with DPP, a result like this is just not possible with ACR.
dkmiles1 wrote:
If I recall, the shadow noise recovery issue rears its head in LOW ISO situations (ISO 100-800) and not in high ISO examples like this one at 3200.
It would be interesting to see the same test but with lens performance neutralized (i.e. shot @ f9 or so). To my eyes the DPP image has more detail both in the wood fence as well as floor tiles.
As others have said, it is spliting hairs in today's hardware / software. All are capable of remarkable results, the limitation is the person behind the gear ;-)
thw2 wrote:
OK. As they say, trust no one but yourself.
So, I decided to grab an intentionally underexposed 5D3 shot (50mm, f/1.4 , 1/800 sec, IS 100) and process it with DPP 3.11.26 and ACR 6.7 (just released). ALL NR (luminance as well as chroma) set to zero. Increased exposure by 2 eV and compensated for heavy vignetting (expected at f/1.4) in both programs.
Daan B wrote:
All I am seeing is that the DPP conversion has more sharpening and some NR applied. I don't see more detail.
That's incorrect, it has no NR applied. sharpening was 3/10 (default) for DPP and 25 in LR (default). If you increase sharpening in LR noise gets worse.
As I have motioned before DPP uses proprietary Canon algorithm for demsoaic. This includes characteristics of each image sensor used, including its spectral response and strength of AA filter. Adobe use generic algorithms with some modifications.
DPP also offers chroma and luminance NR separately. The chroma NR removes color noise w/o affecting detail. Luminance NR in DPP is not good, it smoothens the image.
The grain in DPP output is very fine (almost single pixel) pushing noise power to around Nyquist limit. This allows noise suppression w/o affecting details later in the process flow (not in DPP itself) since it reduces noise and detail spectral overlap.
Also for DPP it is best to use "sharpness" in the RAW tab as opposed to unsharp mask. Unsharp mask mimics the in-camera processing, creating halos with coarse details like the cave example above which looks like in-camera JPEG. "Sharpness" will extract fine details.
Any ways, use whatever you like. As I said it doesn't matter how your process your images, the end photo is what matters.
speedmaster20d wrote:
FPN cannot be removed by subtracting dark frames, pattern is fixed but values are random (White). It will get worse by subtraction because noise power is additive.
It depends, from what I've read people have had some success removing certain forms of the banding since it has been, to some extent, constant enough frame to frame so that it actually did improve things overall. Apparently none of the horizontal banding is amenable to this but certain forms of the vertical banding are. There has been some speculation that the type of vertical banding the 5D3 may well be the type somewhat amenable to this. I'll see.
speedmaster20d wrote:
I am not sure if I agree, to my eyes DPP has just as much detail if not more, look at the stripes on the shirt. The difference in sat. is just the default profile, you can increase it.
I don't like LR rendering t all, it generates coarse grain with inferior fine detail IMO...
Here is one example that shows this.
If you are interested here is a field example ISO 3200, 5D3 with DPP, a result like this is just not possible with ACR.
To me it is NOT DPP that is showing tight, fine, natural 'grain'. The ACR samples there have tight, film-like, naturally looking 1 pixel-sized 'grain' of an even nature while DPP has that mix of smoothed out and tight 'grain' and large blotch 'grain' look. I suppose you may say that lets it NR better since it does it more heavily on flat surfaces, but I find that weird semi-smooth to suddenly harsh edges and random bits of 'grain' look quite unnatural looking and bothersome and prefer more overall noise, if need be, to that sort of look myself.
And what settings did you use for ACR pre-sharpening and what post sharpening did you use? The ACR defaults for sharpening I do not find lead to the best performance for fine details.
thw2 wrote:
I honestly cannot find any (noise) advantage in shooting with DPP. Maybe folks forgot to turn off their NR setting in DPP?
+1
In general, and apart from issues with integration with the rest of one's workflow, you can largely get to the same results with a range of raw conversion software. In some cases, the very same NR engine is used, as in LR and ACR.
skibum5 wrote:
It depends, from what I've read people have had some success removing certain forms of the banding since it has been, to some extent, constant enough frame to frame so that it actually did improve things overall. Apparently none of the horizontal banding is amenable to this but certain forms of the vertical banding are. There has been some speculation that the type of vertical banding the 5D3 may well be the type somewhat amenable to this. I'll see.
Noise is random, there is no such thing as "constant noise" that is called an offset. FPN cannot be reduced by subtraction of dark frames. There might be some cross-talk between readout channels in old cameras you maybe able to reduce by dark frames.
skibum5 wrote:
To me it is NOT DPP that is showing tight, fine, natural 'grain'. The ACR samples there have tight, film-like, naturally looking 1 pixel-sized 'grain' of an even nature while DPP has that mix of smoothed out and tight 'grain' and large blotch 'grain' look. I suppose you may say that lets it NR better since it does it more heavily on flat surfaces, but I find that weird semi-smooth to suddenly harsh edges and random bits of 'grain' look quite unnatural looking and bothersome and prefer more overall noise, if need be, to that sort of look myself.
And what settings did you use for ACR pre-sharpening and what post sharpening did you use? The ACR defaults for sharpening I do not find lead to the best performance for fine details....Show more →
I have no idea what you are talking about in terms of "large blotch" etc. These things are subjective. To my eyes LR looks grainy and coarse, DPP looks natural and detailed.
The grain in ACR output has an average radius of 3-4 pixels in DPP is 1-2. Unless you use wrong settings, such as USM or luminance NR etc. which is what most people do.
Anyways, we cannot agree on this topic. I expressed my opinion and you did yours. time to move on.
gdanmitchell wrote:
In general, and apart from issues with integration with the rest of one's workflow, you can largely get to the same results with a range of raw conversion software. In some cases, the very same NR engine is used, as in LR and ACR.
I don't know if that's a guess on your part or a fact, but even if it were the same NR "engine," each software's interpretation of Raw image code varies. Canon's Raw code is proprietary, and their algorithms and understanding of their products largely is too. Adobe, et. al. have to backwards engineer every camera manufacturer's Raw code, so even if the engine is the same, its effect on an image obviously are not.
Todd Klassy wrote:
...each software's interpretation of Raw image code varies. Canon's Raw code is proprietary, and their algorithms and understanding of their products largely is too. Adobe, et. al. have to backwards engineer every camera manufacturer's Raw code...
Um. Not exactly.
A RAW image does not contain code. It contains data which you can directly see by using a variety of programs (dcraw, RawTherapee, etc.)
Out of this raw data a converter (re)constructs an image. There are several steps involved, from debayering to imposing a tonal curve. Different converters do this in different ways -- but there is no need to reverse engineer the sensor manufacturer's converter because there are many valid ways to go about it. Different converters make different choices and pick different trade-offs, and I don't see that Canon has any built-in advantage except that it gets a head start when a new camera is released.
Wiki says part of the RAW file is encrypted. (So does this from 2005) http://www.dpreview.com/news/2005/4/27/davecoffininterview
From that link: 4. I take it that reverse-engineering the metadata out of the RAW file is just as complicated (if not more so) than the actual sensor data itself, is this correct?
Yes, the metadata is much more complicated. That's why dcraw reads only metadata necessary to decode the image, and ignores the rest.
If any of the metadata that can help in interpreting the file - for instance, I can imagine the metadata can help guide which of a choice of algorithms to use - then Canon will have a built-in advantage.
The main issue in demosaic is figuring out the correct correct matrix for combining red, green and blue values to arrive at the RGB vector for each pixel. If the spectral response of the CFA is not know these values have to be found by trial and error using a calibrated color chart. But the problem is that CFA response is not linear and calibration is difficult.
The result is chroma noise and color blotches that are heavily correlated in adjacent pixels plus larger grain size. This increases noise PSD overlap with spatial detail. Attenuating such tones will suppress detail. This is why NR often smoothens details.
For best results deconvolution during demosaic needs to take into account the features of the low pass filter. This was very nicely demonstrated in a separate thread on this forum with synthetic images. Adobe just apply USM to the output, this is the same type of sharpening used in Photoshop on JPEG files etc. On the other hand, if set to "sharpness", DPP performs particular deconvolution calculation during demosaic itself which squeezes out every bit of detail that is in the RAW data.
Canon DPP also offers DLO which corrects optical aberration effects based on the characteristics of the lens used. There has been numerous threads on this last one. It has proven effective in many cases. This again is not simple sharpening.
All of this relies on propitiatory information about image sensor and optics that is not available to third-party developers.
Canon has offered development modules to Adobe and others in the form of SDK. However Adobe cannot use the SDK in their software because it will invalidate their exclusive-rights IP by the venue of prior art.
The same is true about Nikon. CNX2 produces best results with Nikon cameras and for a long time Adobe had problem decoding WB with D700/D3/D3s cameras. This lead to color artifacts. There were many complaints and I think finally the reached an agreement to fix it. This was the official statement from Adobe to try to clear things up
It somewhat states that Nikon will continue developing their own software for NEF but they will support Adobe's effort while obviously not sharing their secrets.
Adobe's main advantage is their experience in integration and user interface.
Any ways, these discussions are a bit redundant to photography. It is best to go out there take pictures and see which camera/software works best for you. It is the talent of the a photographer that matters not the equipment or RAW converter they choose.
skibum5 wrote:
ACR defaults to a too large sharpening radius to extract maximum fine detail IMO.
DPP defaults to too much blurring, odd-looking NR at default IMO.
Who uses ACR defaults? It's still a shame to see Adobe won't allow radius below 0.5, but the mask helps enormously to just target areas of real detail.
The whole RAW code thing really needs to go..
If Canon swapped like Leica to defaulting to using DNG it would be a big step in the right direction..
Been playing and there is a definite difference in the created files.
I did a massive batch of hi ISO stuff at between 3200 and 8000 processed a few files with LR4/PS5 and DPP as well as I could to see which gave the best results.
Letting em sir for a day before I look at them so I can see them with more independent eyes..
Todd Klassy wrote:
I don't know if that's a guess on your part or a fact, but even if it were the same NR "engine," each software's interpretation of Raw image code varies. Canon's Raw code is proprietary, and their algorithms and understanding of their products largely is too. Adobe, et. al. have to backwards engineer every camera manufacturer's Raw code, so even if the engine is the same, its effect on an image obviously are not.
It isn't a guess. Adobe has said so. I've been at Adobe and heard the Adobe LR and PS engineers say so.
Raw data are basically just the "raw" sensor data - luminosity values of different photo sites. It starts out the same no matter what conversion software you use.
There are differences in how the various programs process the data, and there are differences in the user interfaces among the programs - what things are called, what settings are editable, what "smart" adjustments the application designers have provided, how the files interact among programs.
My point was that in the end, you can get great converted files from a whole range of programs, and many of the purported differences I've seen have been more about the way the programs were used than about their intrinsic abilities of process file data. A skilled operator who understands the variables provided by the software and who understands the photographic effects of applying them can get great conversion results with a range of programs.
I'm still partial to ACR, though I can understand why quite a few folks have come to prefer LR. (Here, the raw conversion abilities are very comparable, as the two programs share code from what I have been told.) My reasons for using ACR include: it does a very fine job (equivalent to LR though interface isn't the same), I'm deeply familiar with it from long use and can quickly and effectively get the file to the point I want, and it integrates very tightly with Photoshop, which is my preferred post-processing tool. (I especially like the smart layers feature that allows me to easily and non-destructively return to the raw convertor even after working in PS.)
By the way, I agree with the person who called for camera manufacturers to adopt the DNG format for raw files. The whole proprietary thing is unnecessary (and dng can accommodate differences in camera files) and it should concern us all that the format of our creative work is locked up in proprietary formats when a non-proprietary one is available.
