|
Forum: Pentax K-1 & K-1 II
06-08-2018, 03:10 AM
|
| |
The "accelerator" processor is not an analogue unit. There is no "boosting" (as in amplification).
Fourier transform (taking the image from the spatial domain into the frequency domain) analyses show that there is high-frequency attenuation starting with ISO 640. This does not happen with regular noise reduction (such as dark frame substraction). This only happens when denoising techniques are applied that use smoothing which itself is based on determining the value of a pixel depending on its local context (cf. "nearest neighbour" approaches).
One can debate the visual impact all day long and even come to the conclusion that one prefers the treated output, but objective measurements don't lie.
|
|
Forum: Pentax K-1 & K-1 II
06-07-2018, 05:01 PM
|
| |
The difference can become significant when using image stacking (e.g., in astrophotography). Stacking noisy but otherwise intact images will result in a final image with proper detail. Stacking already denoised images cannot perform the same feat.
Furthermore, future RAW developers may offer advanced (not yet existing) denoising methods that will depend on unmodified data.
Quite possibly many are fine with the RAW manipulations. However, we do not know how many opted for the conversion without knowing about the consequences.
For sure.
I wonder whether DPReview's planned comparison between K-1 and K-1 II images will account for that. If they do, they potentially may find that their initial assessment may not be as defensible as they thought it would be (depending on how robust the K-1 II processing really is).
Yes, undoubtedly.
I have no problem at all with people who prefer the K-1 II output.
I still maintain, though, that it was unwise to give DPReview that kind of ammunition. Even if no single user would ever be negatively affected by the "accelerator" processing (most likely a purely hypothetical scenario) then DPReview's criticism of the mandatory denoising still nevertheless hurts Pentax and Pentaxians. For that reason alone, the feature should have been made optional to start with.
|
|
Forum: Pentax K-1 & K-1 II
06-07-2018, 04:45 PM
|
| |
I do.
Different cameras produce different images of the same scene. It doesn't make sense to then let some adaptive processing potentially homogenise differences.
Of course, you can. But then you are suggesting to move away from Pentax. This is not the first thing that springs to mind to most members of this forum.
The point is that it is completely unnecessary for Ricoh to alienate some subset of Pentax users. All they have to do is to make the processing optional.
You seem to miss that ACR does not "manipulate the raw data".
Nothing "manipulates the raw data", except potentially the camera.
Also, the fact remains that I can replace any RAW developer I don't like with another one. I cannot do the same with the camera without having to replace all my lenses, potentially start using an EVF or lose image stabilisation or weather-sealing, etc.
Is that so? Do all makers manipulate the sensor data?
Do you have anything to corroborate that view?
Note that the idea of all manipulations being created equal is of course nonsense. It is not the case that all RAW files are all lying only in different ways. This discussion is not as subjective as you may think it is. DxoMark and PhotonsToPhotos have measured RAW file properties. It is not uncommon across brands to do some manipulation at ridiculously high ISO levels. Personally, I don't care if any processing is applied to an ISO 102,400 image as it always going to lack any reasonable sense of fidelity.
Pentax seems to be unique in starting processing at ISO 1600, and now with recent camera models at ISO 640 already. The respective analyses show that the processing is not just any kind of data massaging but destructively smoothes out higher frequencies, i.e., performs denoising.
So, no, it isn't the case that everyone does the same thing anyhow and that everyone is assuming what they want.
|
|
Forum: Pentax K-1 & K-1 II
06-07-2018, 04:26 PM
|
| |
Perhaps, but maybe not.
I was impressed by the detail you recovered from one K-1 II image, but I'm sceptical that the same technique will be successful with different scenes where the "accelerator" may have smoothed irregular details (such as the dust particles on Adam's electronic component images).
I also noticed blotches in your processed K-1 II image that weren't present in the K-1 version; I guess that's what DPReview refers to as "artefacts".
I find some noise to be quite natural but smoothed surfaces and artefacts to be rather unnatural.
|
|
Forum: Pentax K-1 & K-1 II
06-07-2018, 03:50 AM
|
| |
You should tell that DPReview.
Rishi Sanyal asked whether it would be better to use dcraw for the camera comparisons. I answered "yes" and so should many more.
The trouble with ACR (Lightroom and Photoshop) is not just all the "under the hood" processing that happens even if you think you are using "neutral" settings, it is also highly problematic that they change their processing depending on the camera and that their processing is "adaptive", i.e. depends on image content. ACR is just a bad, bad choice for doing comparisons.
I disagree.
Although it would be desirable if Adobe gave one a way to turn off all their processing, there isn't necessarily an expectation towards a RAW processor that one is able to see a neutral version of the RAW file; the latter would look very unappealing. RAW processors are typically not meant to be used for scientific measurements or to support sound comparisons of cameras. If someone doesn't like what Adobe is doing, they are free to choose another RAW converter. There are excellent alternatives available.
It is an entirely different matter for RAW files. These come with the promise that they contain all information the sensor captured. Some kinds of modification to the data are fine, as long as they are not destructive. If a RAW file is in some way compromised, however -- say due to lossy compression, clipping black levels, smoothing out noise (independently of whether it is system-generated or scene-inherent noise) -- then one has no recourse at all. One cannot simply choose another software to circumvent the issue.
I generally value your contributions, but I feel with this one, you are not living up to your usual standard.
This statement seems to be expressing that some people are complaining because that's what they enjoy doing, implying that there is no real cause for complaint. I don't know whether Sony's lossless compression algorithm is sub-par, but assuming it isn't then of course the "complainers" have nothing to complain about. However, anyone complaining about lossy compression or mandatory RAW denoising has a point. One may not feel that these concerns cross one's own threshold for complaining, but one should be tolerant about the needs and desires of others and not label them as "complainers".
|
|
Forum: Pentax K-1 & K-1 II
06-04-2018, 07:24 PM
|
| |
You have that impression because apparently you don't know how to read the curves.
Rest assured that a lot of people find these measurements to be very useful and that they do correspond to photographically relevant properties of a sensor.
Of course forum discussion reference "dynamic range" all the time. It is also true that an increased dynamic range can improve how much one can push shadows. However, the latter is not necessarily the case. If one increases dynamic range by increasing the full well capacity of sensels on a sensor then this won't necessarily have a beneficial effect on how far one can push shadows (given the same exposure).
The property you are interested in is "SNR" (signal to noise ratio) at low signal levels.
First, DR curves are not meant to show the property you are looking for.
You need to look at the "Full NR" curves which show the SNR for the full range of light levels and then want to focus on low light levels (as you are interested in shadows, or at least pushing low light levels).
You'll find, for instance, that when the K-1 (click on "Full SNR") reaches 6dB at ISO 100, the D810 (click on "Full SNR") has not reached 6dB at the same light level yet (you can use your mouse cursor as a reference point when flicking between two browser windows). This means that at ISO 100, the K-1 supports a bit more shadow pushing. However, at ISO 64 (a setting not available to the K-1), the D810 has a better SNR at the same light level.
Second, you shouldn't take DPReview's DR shots at face value. We know that they initially made a "processing" mistake with their ISO 12,800 shot for the camera comparison tool. While they have updated the "camera comparison shots", they haven't updated the "DR" shots for the K-1 II. The latter are still the result of using a "1.00" firmware and unknown (and potentially incorrect) post-processing. Perhaps they unintentionally used noise reduction for the K-1 II? Something along these lines must have happened for their initial ISO 12,800 image on which they partly based their original verdict.
Using DPReview studio scene shots as your ultimate source of truth is not a good idea. They are simply not done with a level of rigour that justifies such a level of trust. Just note the magenta tinge in the K-1 II DR shots. None of the other cameras (including the K-1) have such a tinge problem and I maintain that a proper processing of the K-1 II files should not produce a magenta tinge.
|
|
Forum: Pentax K-1 & K-1 II
06-03-2018, 10:34 PM
|
| |
Surely it is possible to evaluate images visually.
That shouldn't stop one, however, to use technical terms correctly.
If you are using terms like "supple", "rich", "lustrous", you are in the artistic domain and have to talk to artists of how to use that vocabulary.
If you are using terms like "dynamice range", "signal to noise ratio", "focal length", "exposure", you are in the technical domain and ought to use the terms in accordance with their definitions.
As photographers we want to communicate with our images. So let's do it with our written word as well. This won't work, though, if some people decide to reference a spade by using "fork". The term "fork" already has a defined meaning and using it for a spade is misleading, so let's keep calling a spade a spade.
I don't think you have the numbers to back up the claim that the majority of people confuse "dynamic range" with "signal to noise ratio".
What you are looking at -- noise levels at very low light levels -- has much more to do with SNR (signal to noise ratio) than it has with "dynamic range".
SNR and DR are not decoupled entities so I know where you are coming from, but still you are using the vocabulary incorrectly.
The definition I'm using is not my personal choice. It has been established by experts.
You shouldn't make your own personal choices either, as to what a technical term means. Again, use "richness" or "cleanliness" with your own artistic license, but when you use a well-defined technical term like "dynamic range" it would be fitting to not stray from the well-defined original meaning.
Again, I don't think you have the numbers to back up the claim "most people". Even if it were the case, then "most people" shouldn't hijack technical terms to let them mean whatever they please.
I know what you are talking about, you are just using the wrong language. On top of that, I'm convinced that you are not seeing a real phenomenon, i.e., nothing that can be related to "dynamic range" as realised through physical hardware, but rather a processing result. There is no way Ricoh found a way to increase the SNR of the K-1's sensor to this degree. I don't know what is going on exactly (could be a number of things) but I sure know that your notion of the K-1 II being a "dynamic range miracle" is wrong.
Guess who built the camera you are talking about?
Guess what kind of terminology they used when designing it?
These "worthless" curves objectively measure properties of the hardware.
You are just failing to correctly interpret them and use the terminology correctly.
Do you really think you are smarter than the people who do this for a living?
Perhaps you should use a non-technical term like "pushability". Then we can agree that you like the "pushability" of the K-1 II and there is no tension with technical terms and the scientific measurements done by DxOMark.
|
|
Forum: Pentax K-1 & K-1 II
06-03-2018, 12:27 PM
|
| |
I'm afraid it doesn't work that way in a technical domain. :)
Certain terms have a defined semantics. You can choose to make your own definitions but that will create a communication problem with people using the standard meanings.
An increased DR can lead to an increased ability to push shadows, but you shouldn't make the reverse reasoning that anything that looks like it supports pushing shadows is a symptom of increased DR. The difference between the darkest and brightest perceptible points in an image (-> dynamic range) is not affected by post-denoising.
I think your conceptualisation of ISO is incorrect. There are a number of alternative ways in which one can determine an ISO equivalent value for an imaging system. The one that is typically used for digital sensors is based on saturation capacity. A lower ISO for a digital sensor does not mean "less sensitive to light". It just means "more capacity to deal with bright light". In other words, the lower the minimum ISO value, the more dynamic range a sensor offers. These facts are not debatable. Again, you can choose to attach different meanings to technical terms, but you won't change how engineers use them. :)
I don't think so. I'd say "different vocabulary".
It is fine AFAIC if you prefer the K-1 II ISO 100 images, but you must not conclude that the K-1 II actually offers more DR. If you visit DxOMark you can see how the D810 beats the K-1 (barely) due to its lower minimum ISO setting.
|
|
Forum: Pentax K-1 & K-1 II
06-03-2018, 04:13 AM
|
| |
What we are seeing is not "dynamic range" per se but noise levels after pushing in post.
I highly doubt that the K-1 II has more dynamic range than any of the cameras you compared it to.
It rather seems obvious that noise reduction is applied even at ISO 100.
Apparently the treatment differs from what is done after ISO 640 because bclaff did not detect any loss of details at lower ISO levels.
However, note that the image you are looking at was taken with firmware 1.00. This firmware is older than the regular release firmware. Note also, that the K-1 II shows a magenta tinge. This suggests that either the firmware version wasn't quite up to scratch yet and/or ACR did not properly deal with the K-1 II files. I wonder why DPReview never mentions that they are using a converter which has no dedicated camera profile for the tested camera yet.
In any event, unless someone goes through the effort of pushing both files from a K-1 and a K-1 II then -- using DPReview's tool only -- we don't know whether the K-1 II applies NR at ISO 100 already.
The D810 has a better sensor with true ISO 64 capabilities.
The K-1's sensor is probably the same as the one in the D800 (or very similar to it).
I think you meant it as a rhetorical question, but the obvious answer is that RAW data manipulation happens across the board.
DPReview is specifically taking issue with the low threshold (ISO 640) for manipulation and the fact that details appear to be lost and artefacts are introduced.
Interestingly, the K-1 II review specifically advises against using the K-1 II for astrophotography while no such warning is made for the Sony A7RIII or A7RII, both of which eat single pixel stars. I once confronted Rishi Sanyal with similar inconsistencies in evaluation, and he insisted that with sufficient levels of nitpicking one could construct a "DPReview bias" case for or against any manufacturer.
|
|
Forum: Pentax K-1 & K-1 II
06-03-2018, 01:08 AM
|
| |
Well, I do have trouble believing that it is the "chroma noise reduction" that brings back the detail.
Outer edges have much more contrast in your processed K-1 II image and that is normally a response to increased sharpening.
If Darktable's chroma noise reduction really increases edge definition in that manner, perhaps it features a small amount of sharpening to counter a loss of structure? I don't actually believe that and unless convinced otherwise assume that the main contributor to the recovery of detail is the extra sharpening.
Definitely interesting work, though!
We should DPReview know, who are preparing a piece in which they are going to demonstrate (no doubt using ACR) that processing K-1 II files makes matters even worse.
Of course they do, just as well as Nikon. Both have had "star eater" issues.
High-ISO RAW denoising is also quite common, with the significant difference to Pentax being that the other companies apply it at ridiculously high ISO settings only, whereas Pentax has a history of starting at ISO 1600 (not with the K-1, though) and opened themselves up to a debate by starting at ISO 640 with the K-1 II.
As to whether Pentax is dumb due to using an "accelerator" approach, I'd say that they were rewarded for this particular approach in the case of the KP. Back then no one seemed to care about the negative consequences. It should also be observed that more expensive sensors than the one used in the K-1 II support a dual-gain approach for reducing noise in high-ISO images. One cannot fault Ricoh for trying to emulate the same advantages through software (the algorithm that runs on the "accelerator" chip) that others can obtain through sensor hardware.
It would all be moot, if Ricoh gave users a choice. Some Pentax camera models, at some point in time, had a mandatory dark frame substraction at higher ISO settings. This was a disaster for astrophotographers and luckily that limitation was lifted in later models. I hope that Ricoh will do the same for the currently mandatory RAW denoising.
|
|
Forum: Pentax K-1 & K-1 II
06-03-2018, 12:41 AM
|
| |
I didn't take BigMackCam much time at all to produce a slightly better K-1 result using Darktable. There might be subtle differences regarding the blues in the K-1 II shot, but I have no doubts whatsoever that this effect could be mimicked in post-production for the K-1 as well. Regarding chroma noise and detail, I prefer the result from the K-1.
EDIT: Meanwhile BigMackCam posted another version of the K-1 II shot that shows more detail. On the one hand I agree that it counts what one can achieve in the end, even if it involves extra sharpening. On the other hand, I'm unsure whether this will always work, e.g., with the smudged dust on the electronic part featuring in the pentaxforums K-1 vs K-1 II image comparison.
|
|
Forum: Pentax K-1 & K-1 II
06-03-2018, 12:26 AM
|
| |
Thanks a lot for posting the result of your experiment.
This is what I expected.
+1
Do you think we can organise a request to Ricoh to make the RAW denoising optional?
DPReview claim they are in "discussion" with Ricoh regarding the "accelerator" processing but every time I ask them about the status of this "discussion" I get the sounds of crickets only. It may well be the case that they mentioned the possibility of making the processing optional once but never received a response from Ricoh.
I think it would be good if perhaps Adam could use his weight as the pentaxforums.com owner to let Ricoh know that there are many Pentaxians who would wish for the RAW denoising to be optional (if only to remove ammunition from DPReview and other sites, e.g., Lenstip).
This goes to show how "thorough" DPReview's reviewing is.
They definitely noticed the unusually low noise levels of the KP but overall highly praised the KP for "squeezing more out of the sensor" than competitors.
One wonders what would have happened if they hadn't botched up their initial evaluation of the K-1 II. They may not have noticed the loss of detail at all. It surely is not that easy to detect from a mere visual inspection of the current sample images. Like monochrome, I agree with DPReviews overall criticism, but not with their standard of work, the attitude of some of their staff towards valid criticism, and their lack of consistency which unfortunately typically translates into disadvantaging Pentax whether through charged language or very questionable numerical scores.
There is no way the value for money proposition of the K-1 II is as low as DPReview indicates and an IQ comparison to cameras whose current (not introduction) prices are 160% (A7R III) and 165% (D850) of the K-1 II is dubious.
|
