[photoptimist]

I agree with the overall premiss that one will be hard pressed to find any RAW output that hasn't been processed to look better, however, let's not lose sight of the difference between non-contentious processing and information-destroying NR.

But what about information-preserving or even information-enhancing NR? An NR algorithm that removes dark current, corrects for pixel-to-pixel sensitivity variations, or removes cross-talk both reduces noise and enhances information in the sense that the resulting RAW file is a more accurate representation of the light levels in the original scene.

Of course a sensor will have internal stages, some of them concerned with reducing noise. Most of these are completely fine, however, as they reliably only remove noise that would otherwise be added by the chip. For instance, there are schemes to ensure that it doesn't matter whether any residual electrons are left in sensels before they are exposed again, or schemes that help to avoid differences in amplification whenever parallel conversion is involved. All these NR measures are completely fine and no one in their right mind would dispute them.

Great points! You are right.

In contrast, we have schemes like the "accelerator chip" or some "BIONZ processing" that take the data from the sensor chip and then try to make it look better, according to some notion of "better". These are the NR stages that amount to "cooking" and should be purely optional. Unlike the sensor chip internal stages, these "afterburners" have no access to sensor-internal data/properties, i.e., they could be just as well performed outside the camera. That's why they should be available in-camera for those who are happy with what they are doing, but it should be possible to disable them for those who feel they want to run a different/better processing, potentially in the future with methods that aren't available yet.

This is where we diverge because the digital chips on the camera certainly do have access to some sensor-internal data/properties from a number of sources. They potentially have in-factory or post-factory calibration files (such as can be created by the "pixel mapping" function) that have measured some aspect of that camera's specific sensor properties. That data is available to the in-camera accelerator and PRIME chips but not available to an post-procesing software.

What's interesting is that the "future methods" are a bit of a false hope. Every digital camera that I have owned has shown some evidence of sensor aging. Someone might create a future method for getting a more accurate, more pure RAW file but if that requires calibrating the camera to measure it's pixel-by-pixel artifacts, then that future method is of no use to old RAW files because one can't go back in time to get the calibration data.

------------

Of course, all of this is speculation because we don't actually know what noise is being reduced by the accelerator and whether the chip is destroying any information at all. That said, I do agree with you that we should worry about what the chip is doing and understand how it affects our images.

Comparing K-1 to K-1ii images is useful first step to detect some of the effects of accelerator chip but it is not a conclusive proof that the chip is damaging the image. Any difference between a K-1 and K-1ii image may be evidence of image damage by the K-1ii chip or image damage due to uncorrected noise in the K-1. (It may also be evidence of camera-to-camera sensor quality differences, too.) Even difference between K-1ii @ ISO 100 and K-1ii @ ISO 800 have interpretation difficulties. (It may also be evidence of simple shot-to-shot noise differences, too.)

The gold standard test is how closely do the K-1ii RAW files measure the true R, G, and B light levels of the scene. Measuring the exact number of electrons in each pixel is irrelevant -- electrons are not the image. The real number that matters and the true "pure" RAW file would tell us the photon flux at each pixel (which is actually different from the photon count!). There are clever statistical analyses that can spot anomalies in the RAW file to see what the chip might be doing. And there's always the possibility of cross-comparing an ISO 800 single-shot image from a K-1ii to a 3x3 Brenizer method ISO 100 pixel shift image.

Personally, I would suspect that a careful analysis of K-1 RAW and K-1ii RAW files will find: 1) the K-1ii files are more pure; 2) the two files both have artifacts but of somewhat different kinds.

[Class A]

Standard Sony sensors are not invulnerable to external noise sources. OTOH standard Sony sensors are oblivious to external noise sources.

Sorry, I don't understand what you are trying to say.

Digital data already contains noise.

Obviously it does but previously you said "Many sources of noise are outside the sensor... maybe it tries to compensate for that." and my point was that "outside the sensor" it is impossible for further noise to be added. Once the data is available in digital form -- and it is outside a Sony sensor -- then there cannot be any further "sources of noise".

If the K-1 II is as good or better, I'll have no reason to complain.

Please note that I'm not complaining either.

I just consider the (rather likely) possibility that the "accelerator" processing may destroy some information and that some people won't like that.

Others will surely debate about "cooking" regardless of the actual results.

It is not enough to have processing that is better than "no processing". In order to apply any processing to RAW data (do anything you want to JPGs; no one should complain) you have to be sure that your processing will be better than any one that can be invented in the future. If you don't then you are robbing RAW data one of its benefits, i.e., that you will be able to redevelop it with superior RAW development techniques in the future that don't exist yet.

What makes you so confident? Even algorithms cannot be replicated if you don't have the same data to work on.

I'm confident because I have an engineering background.
From the nature of the components used one can make very educated guesses what is reasonable to assume and what is not.

If I took the time to lookup the datasheet of the Sony sensor used in the K-1 (II) then I could most likely tell with absolute certainty that the accelerator chip cannot have access to anything that any external software wouldn't be able to have access to as well. I'm not bothered enough to do it but I'd bet money on the fact that there is no "secret sauce" to the "accelerator" processing.

Have you ever wondered why they call it "accelerator" instead of "sensor enhancer" or similar?
This chip is most likely a DSP running some image processing algorithm whose parameters depend on the shooting data. Nothing you couldn't do later on as well, if you knew the algorithm.

This is just an assumption; basically you're assuming you are right and interpret whatever they said in this light

You are right, without opening up the camera and examining the board and consulting data sheets I cannot know for sure 100% what is going on. But my take on the matter is not just any opinion. It is a well-informed one.

I'm not just assuming "I'm right". I'm considering the parts involved and what makes sense from an engineering perspective. I don't know how much you enjoy getting into the nitty-gritty of technological details, but I think if you did it as much as I do, you would most likely agree with me that I don't make that many assumptions and where there is guesswork involved, my guesses are very well-founded in what makes sense technology-wise.

RAW converters also work with non-demosaiced data (until they apply demosaicing). I assume the meaning is, such processing cannot be replicated with the data from a RAW file.

Yes, of course RAW converters work with non-demosaiced data internally but that is no good to you as a user of a RAW converter, is it? No RAW converter I know allows you to plug in a new NR component into the rendering pipeline before the data is demosaiced. It is in that sense, that Ricoh may be correct in stating that what the accelerator does cannot be replicated with (standard/existing) software.

Who's doing that?
Ricoh Imaging is by far the best source of information we have on this subject though.

You are asking me who is taking marketing statements as gospel on what happens inside the K-1 II?
Are you not doing that when stating "But the point is, they're talking about a kind of processing which cannot replicate in software, after the fact (without detail loss).". Are you not taking their statement at face value, even though there are many reasons why this is probably a bad idea?

Also, I disagree with you regarding Ricoh's marketing statements being the "best source of information". Some engineering knowledge goes a long way in having a way to understand when marketing statements may not hold up to scrutiny.

I'm not saying "Ricoh is lying". Not at all. I'm just saying don't quote non-technical marketing speak in a discussion about what makes technological sense.

There is no such thing as an original RAW data. The sensor outputs an already processed data; and if you somehow manage to disable all noise reduction methods you'd probably get something ugly. Noise suppression starts with hardware.

I will say it for the last time:
Of course there is something like "original RAW data". Most manufacturers don't give you access to it, because they don't want you to see the untreated data or feel RAW converters wouldn't be able to handle it well enough. However, the notion of "original RAW" data surely exists.

Some "processing" is absolutely fine as it doesn't imply any information loss. The discussion is about lossy processing. Surely you must understand that just because there is always some non-contentious processing, you cannot make the jump to stating that therefore any processing is just fine.

I don't think it is possible, by hardware design.

You are entitled to that opinion. In my view it doesn't make any sense, given how Sony sensors work and that there is no reason why it should be impossible to not apply the processing.

[Class A]

But what about information-preserving or even information-enhancing NR?

That would be completely fine, of course.

Note that I consider the possibility that the accelerator chip only performs information-preserving transformations. The jury is still out.

This is where we diverge because the digital chips on the camera certainly do have access to some sensor-internal data/properties from a number of sources.

In theory, that's a possibility. Given the nature of Sony's "all-in-one" sensors, it is highly unlikely. Of course one can consider all kinds of possibilities but I'd love to see some shred of evidence which lends credibility to the idea that the accelerator chip has some kind of privileged access to the sensor. I'll eat my words, if anyone comes up with such evidence. To be clear, I'm not talking about data that could very well be packaged up for post-processing (such as a dark frame).

If Ricoh wants to be taken seriously regarding "kind of signal processing which cannot be obtained by just software processing mechanism" then they must mean "cannot be replicated by post-processing in principle" not just "you cannot do it in post-processing because we withhold the data to do so from you".

They potentially have in-factory or post-factory calibration files (such as can be created by the "pixel mapping" function) that have measured some aspect of that camera's specific sensor properties.

That is a possibility.

However, the better approach would then be to make the in-camera processing optional and provide some way to deliver the data to the user as part of the image or separately.

As an analogy, let me point out that the long exposure NR is not mandatory on the K-1. There are cameras (IIRC, even from Pentax) that make a dark-frame subtraction mandatory. Instead of enforcing this kind of processing, Ricoh allows you to take an alternative path (e.g., take your own dark frames and substract them in post-processing) because there are sometimes advantages to this path.

In the same spirit, I believe any kind of image processing that may imply any kind of disadvantage should be optional. Again, I do not know that the accelerator processing is destructive in any way and I'm not making claims either way.

What's interesting is that the "future methods" are a bit of a false hope.

That is most certainly not true. There has been a time when demosaicing algorithms made big strides and re-processing files certainly made sense. Regarding the subject at hand, if the NR is destructive in any way (and the OP "knows" that it is, doesn't he?) then it is almost certain that future NR methods could do a better job with the unmodified data.

Someone might create a future method for getting a more accurate, more pure RAW file but if that requires calibrating the camera to measure it's pixel-by-pixel artifacts, then that future method is of no use to old RAW files because one can't go back in time to get the calibration data.

True, hence my suggestion to provide the calibration data separately (instead of baking it in somehow).

We agree that if there is only one (non-destructive) way the calibration data can be used to optimal effect then applying the respective algorithms to the RAW data is just fine. I have my doubts that this is the case with the K-1 II's processing, but I'm open to any kind of outcome of future investigations. As I said before, I don't care that much; my K-1 is all the camera I need and want. I would prefer Ricoh not to make any kind of destructive processing mandatory in future cameras, but to be frank, I think there is next to 0% influence we have on the matter. I'm confident, though, that Ricoh will make competent decisions.

The gold standard test is how closely do the K-1ii RAW files measure the true R, G, and B light levels of the scene.

Yes, that's why I suggested it would be ideal to have a PixelShift image for comparison. That wouldn't amount to the "truth" either, but it would be much better than comparing a pushed ISO 100 shot with an ISO 800 shot.

There are clever statistical analyses that can spot anomalies in the RAW file to see what the chip might be doing.

Yes, DxOMark detected certain noise-reduction (smoothing) RAW cooking using such analyses.

Personally, I would suspect that a careful analysis of K-1 RAW and K-1ii RAW files will find: 1) the K-1ii files are more pure; 2) the two files both have artifacts but of somewhat different kinds.

Quite possible. I can't say.

I think it is time for me to bow out of this discussion. I'm really not that bothered what the exact K-1 II story is (it is a fabulous camera either way) and before proper comparisons are made, there cannot be any real progress.

[Kunzite]

Sorry, I don't understand what you are trying to say.

Obviously it does but previously you said "Many sources of noise are outside the sensor... maybe it tries to compensate for that." and my point was that "outside the sensor" it is impossible for further noise to be added. Once the data is available in digital form -- and it is outside a Sony sensor -- then there cannot be any further "sources of noise".

What I mean is that sources of noise could be external to the sensor - for example, noise in power supply, introduced by the SR system, etc - but the Sony chip might not know how to identify and correct for them. Said noise would appear in the digital output.

Please note that I'm not complaining either.

I just consider the (rather likely) possibility that the "accelerator" processing may destroy some information and that some people won't like that.

Noted. I wasn't thinking about you, by the way. You're one of the reasonable folks.
At some point yes, detail will be affected - Pentax did that in past models. There is probably such processing in the K-1, too - except that I don't care after looking at the results. There is probably such processing in the K-70 and KP - except I'm too busy WOW!'ing at the images to notice.
It seems rather a philosophical problem than a technical one; "I want raw RAWs" instead of "I want better results". In our discussion, I'm merely pushing for the latter.

It is not enough to have processing that is better than "no processing". In order to apply any processing to RAW data (do anything you want to JPGs; no one should complain) you have to be sure that your processing will be better than any one that can be invented in the future. If you don't then you are robbing RAW data one of its benefits, i.e., that you will be able to redevelop it with superior RAW development techniques in the future that don't exist yet.

That is raising the bar a bit too high.
There are many reasons to perform in-camera processing, even in cases where it could be matched by (future) external software applications. There is stuff you want it done anyway, and better done consistently (rather than hoping the RAW converters would save the day).
Then, there are the RAW converters which assumes a certain level of RAW quality. No banding due to line/column parallel reading, for example. Feed them a "raw RAW" and they won't know how to cope with it.

I'm confident because I have an engineering background.
From the nature of the components used one can make very educated guesses what is reasonable to assume and what is not.

If I took the time to lookup the datasheet of the Sony sensor used in the K-1 (II) then I could most likely tell with absolute certainty that the accelerator chip cannot have access to anything that any external software wouldn't be able to have access to as well. I'm not bothered enough to do it but I'd bet money on the fact that there is no "secret sauce" to the "accelerator" processing.

Have you ever wondered why they call it "accelerator" instead of "sensor enhancer" or similar?
This chip is most likely a DSP running some image processing algorithm whose parameters depend on the shooting data. Nothing you couldn't do later on as well, if you knew the algorithm.

One thing I've learned in my profession is that you can only assume one thing: that assuming things will get you to the wrong conclusion.

External software would have access to such data only if the data is saved in the RAW file. Perhaps some very detailed and pretty much secret information about the camera would have to be revealed... and even if everything's available, there's no guarantee it would be implemented, correctly and consistently and in a timely fashion.
Lightroom 6 users definitely won't benefit from it.

I would not make assumptions about what the accelerator does based on its name, especially as it might slow down the flow of data sent to the image processor.

You are right, without opening up the camera and examining the board and consulting data sheets I cannot know for sure 100% what is going on. But my take on the matter is not just any opinion. It is a well-informed one.

I'm not just assuming "I'm right". I'm considering the parts involved and what makes sense from an engineering perspective. I don't know how much you enjoy getting into the nitty-gritty of technological details, but I think if you did it as much as I do, you would most likely agree with me that I don't make that many assumptions and where there is guesswork involved, my guesses are very well-founded in what makes sense technology-wise.

Now, this is another assumption - what I did and to which extent, and how it would influence my thinking.

Yes, of course RAW converters work with non-demosaiced data internally but that is no good to you as a user of a RAW converter, is it? No RAW converter I know allows you to plug in a new NR component into the rendering pipeline before the data is demosaiced. It is in that sense, that Ricoh may be correct in stating that what the accelerator does cannot be replicated with (standard/existing) software.

Some RAW converters are more transparent than others, in telling you what they're doing at demosaicing. I believe that quite a few of them are doing pre-demosaicing NR.
My point is, the option is there.

You are asking me who is taking marketing statements as gospel on what happens inside the K-1 II?
Are you not doing that when stating "But the point is, they're talking about a kind of processing which cannot replicate in software, after the fact (without detail loss).". Are you not taking their statement at face value, even though there are many reasons why this is probably a bad idea?

Also, I disagree with you regarding Ricoh's marketing statements being the "best source of information". Some engineering knowledge goes a long way in having a way to understand when marketing statements may not hold up to scrutiny.

I'm not saying "Ricoh is lying". Not at all. I'm just saying don't quote non-technical marketing speak in a discussion about what makes technological sense.

Yes, that's exactly what I'm asking.
No, this is not what I'm doing. I'm not even taking their statements at face value - however, they are from Ricoh Imaging and supposed to know better than us who are merely guessing. If they offer information, too bad I can't double check it but I'm not going to ignore it either.

I will say it for the last time:
Of course there is something like "original RAW data". Most manufacturers don't give you access to it, because they don't want you to see the untreated data or feel RAW converters wouldn't be able to handle it well enough. However, the notion of "original RAW" data surely exists.

Some "processing" is absolutely fine as it doesn't imply any information loss. The discussion is about lossy processing. Surely you must understand that just because there is always some non-contentious processing, you cannot make the jump to stating that therefore any processing is just fine.


You are entitled to that opinion. In my view it doesn't make any sense, given how Sony sensors work and that there is no reason why it should be impossible to not apply the processing.

How do you define "original RAW data"? Is it the data as read from the sensor? Banding, all sort of issues? How about all the processing done on the sensor? How about done directly on hardware, where do you stop?
How do you convince all RAW software developers to process said data to a level of quality you're happy with? What do you do when your competition offers less "original" RAW data which looks much better?
That's the dilemma with pursuing a philosophical goal rather than a practical one. You don't even know what you're after.

I never stated that any processing is just fine. I'm not even saying that the K-1 II's processing is completely without loss of detail. I'm saying that, from what I've seen so far, there's no reason for concern - no smoothing, no star eater.
I'm looking forward to a proper K-1 - K-1 II comparison.

[MJKoski]

It is good idea to try Nik Raw Presharpener plugin. It can be set to edge enhancement mode and results are similar depending on amount of sharpening.

[reh321]

Please also note that when I state that there should be an option to get original RAW data, I'm talking about a matter of principle. I think this should be true for all manufacturers. As all manufacturers appear to be doing some level of "beautification", I'm not pointing at Pentax specifically. However, Pentax has always struck me as a brand that wants to get things right as opposed to appealing to the masses and in that sense my expectations towards Pentax are a bit higher.

I'm not even certain what this means. I am quite certain that sensor designers 'tune' output by resistance, capacitance, inductance, and routing decisions with respect to signal line(s) within the sensor itself. There is no way to get to the 'original data' other than by going through those lines. That is why BSI is such a big deal - it affects their approach to those lines.

Even if future comparisons reveal that everything the accelerator does is fine, where would be the harm to allow disabling it for those who think that they'll be able to beat it in the future?

From the talk, I am fairly confident Pentax engineers consider the 'accelerator' to be an extension to the sensor rather than an extension to the processor. Do you want to 'disable' anything in the sensor that affects quality? What does that even mean?

[MarkJerling]

From the talk, I am fairly confident Pentax engineers consider the 'accelerator' to be an extension to the sensor rather than an extension to the processor. Do you want to 'disable' anything in the sensor that affects quality? What does that even mean?

I agree. I do wonder if we're not overthinking this.


According to this site, Pentax K-1 Mark II: todas las novedades - Blog de Tienda Pentaxeros, the chip does this:
"The main and most interesting is the aforementioned accelerator for the PRIME lV image processor, which promises a higher processing speed and an excellent noise reduction, improving image quality. Pentax claims to have drastically improved noise management to ISO 819,200 to expand the creative possibilities in high sensitivity shots."



The site goes on to say:
"This accelerator was already implemented in the Pentax KP and K-70 , in which the processing and noise containment are excellent for a sensor of APS-C format of 24 megapixels, reaching to equal - and even exceed - in this section the performance of many full-frame sensors."

From what I'm reading, it appears that we are seeing a greater computing capability in order to help improve what leaves the camera in terms of RAW image. I can't see a benefit in having the user choice to enable or disable the accelerator chip. It's not possible in the KP or in the K-70 and no users of those cameras have made mention of the need to have a choice.

So, are we overthinking this? Attempting to view the accelerator chip as some sort of separate process happening? Because to me, it seems it's more of an enhancement to the PRIME IV chip. In effect, the ability to change the output from that chip for the better, without building this added computing capability into that chip so that we still only see one large chip on the board, rather than two.

[clickclick]

There's been a lot of really good discussion here, and I respect the points of view expressed, but really, until we have some very well controlled comparisons in a variety of conditions between the mark I and mark II models, I think we're at the point of just splitting hairs.

[reh321]

I agree with the overall premiss that one will be hard pressed to find any RAW output that hasn't been processed to look better, however, let's not lose sight of the difference between non-contentious processing and information-destroying NR.

Of course a sensor will have internal stages, some of them concerned with reducing noise. Most of these are completely fine, however, as they reliably only remove noise that would otherwise be added by the chip. For instance, there are schemes to ensure that it doesn't matter whether any residual electrons are left in sensels before they are exposed again, or schemes that help to avoid differences in amplification whenever parallel conversion is involved. All these NR measures are completely fine and no one in their right mind would dispute them.

In contrast, we have schemes like the "accelerator chip" or some "BIONZ processing" that take the data from the sensor chip and then try to make it look better, according to some notion of "better". These are the NR stages that amount to "cooking" and should be purely optional.

This, ultimately, is where we disagree. The only "betterment" sought by the 'accelerator' is the removal of noise - that is what we have been told and all that we have seen; I also would object if the 'accelerator' were adding anything that wasn't in the scene, but I think it is great if they remove something that wasn't there in the first place. By definition 'noise' is something 'added' by the sensor. The stars which were 'consumed' by that Sony camera were real artifacts, but noise is the result of heat, or crosstalk, or some other phenomenon within the system - and since the sensor is the only component in the system before the 'accelerator', it must be generated by activity within the sensor. Sony could simply add this circuitry as another layer of the sensor, as another one of those 'internal stages' you mention, but since Pentax doesn't manufacture sensors, they have to add it as another chip, as an 'external stage'.

So, are we overthinking this? Attempting to view the accelerator chip as some sort of separate process happening? Because to me, it seems it's more of an enhancement to the PRIME IV chip. In effect, the ability to change the output from that chip for the better, without building this added computing capability into that chip so that we still only see one large chip on the board, rather than two.

I agree with everything you say except I view the 'accelerator' as enhancing the sensor, rather than the processor.

[normhead]

There's been a lot of really good discussion here, and I respect the points of view expressed, but really, until we have some very well controlled comparisons in a variety of conditions between the mark I and mark II models, I think we're at the point of just splitting hairs.

IR will have their test results out in a month or two, and you'll be able to compare the K-1ii to anything you want.Well, with the exception of maybe a 100 ISO image underexposed 3 stops then augmented in post production. Most of us just look at 800 ISO images and say "good enough or not good enough". That works for me. If the K-1ii is 2 stops better at higher ISO settings that's great, that is the difference between 1600 ISO and 6400 ISO for bird images, and the difference between 1/500s (barely adequate) and 1/2000s for my shutter speed. Now that's incredible if it holds up in testing. As for not shooting 800 ISO for landscape.... really? What were you thinking?

[clickclick]

IR will have their test results out in a month or two, and you'll be able to compare the K-1ii to anything you want.Well, with the exception of maybe a 100 ISO image underexposed 3 stops then augmented in post production. Most of us just look at 800 ISO images and say "good enough or not good enough". That works for me. If the K-1ii is 2 stops better at higher ISO settings that's great, that is the difference between 1600 ISO and 6400 ISO for bird images, and the difference between 1/500s (barely adequate) and 1/2000s for my shutter speed. Now that's incredible if it holds up in testing. As for not shooting 800 ISO for landscape.... really? What were you thinking?

Yep, and I think I'm in a similar camp as you. While I use my K3II for most of my birding, the combination of being able to get that shutter speed up combined with imrpoved AF, which sounds like it might be pretty significant, makes the mark II upgrade a really appealing step for me. And I figure when I'm serious on landscape photography, I'll get the tripod out, and if no tripod, the hand held pixel shift looks like another win win situation.

Really looking forward to the start of the upgrade program here in the U.S.

[MJKoski]

It bothers me how Pentax pulls the improvement out of the 36MP chip. I have used every camera model which has the chip and only D810A (which really has improved high ISO compared to standard D810) has its base ISO changed to ISO200 instead of ISO64 so there is no free lunch. But now there are such promises.

[photoptimist]

It bothers me how Pentax pulls the improvement out of the 36MP chip. I have used every camera model which has the chip and only D810A (which really has improved high ISO compared to standard D810) has its base ISO changed to ISO200 instead of ISO64 so there is no free lunch. But now there are such promises.

Right, there no free lunch.

But that does not mean that Ricoh can't invest money in R&D to build non-destructive noise reduction processes in a custom chip and then sell you that lunch.

[reh321]

It bothers me how Pentax pulls the improvement out of the 36MP chip.

What does this even mean? Improvement over what?

[Rondec]

What does this even mean? Improvement over what?

I think he is comparing it to the D800, D810, and A7r, all of which have some version of the 36 megapixel sensor. I forget why, but there was some issue with the original K-1 that he didn't like -- something about color shifts with image stacking of 200 images -- something like that.