[SteveinSLC]

Ricoh may indeed have sourced the accelerator chip externally. As was pointed out earlier in the thread, chips such as these produced by Nvidia and ATI have long been used to enhanced graphics processing in computers.
More recently, they have been found to be exceptionally useful for certain types of artificial intelligence processing. I wouldn't be surprised if Nvidia sells more chips for that than for graphics. So its possible that if Ricoh came up with various types of AI algorithms to do things like post processing, focusing, or the dynamic pixel shift, the chip could be doing that too.

[Ian Stuart Forsyth]

I think everyone commenting on RAW files should specify which converter they are using and whether they checked that the default settings for various ISO settings are the same. In the case of ACR (Adobe Camera RAW, which is used in both Photoshop and Lightroom) there are likely manipulations going on under the hood, i.e., there will be some processing even if the sliders indicate a neutral setting.

I am using fast raw viewer

---------- Post added 04-23-2018 at 06:56 PM ----------

Having looked at the RAW conversions presented, I'd say this discussion is one of "splitting hairs" (can't argue with that can you? ).

What I am seeing is that the areas that are slightly out of focus with any tonal variations start to smear into 1 uniform glob with no detail, something that one would expect from rather generous NR while the small detail appears to have some sharpening a applied.

I know its minor but we are talking about iso 800 and why the need for noise reduction, and what is more important going from iso 100 to iso800 is where we would normally see the most reduction in noise as this range is where we see a decrease in read noise because the sensor isn't quite isoless.

It would be nice if you had the option to use it up to iso 800 better would be up to iso 1200 (without NR) as this is a way to bypass any NR in camera because pass iso 1200 really anything you need above that can be pushed afterwards without the worry of any extra read noise. Now if you want to avoid NR your are stuck to a lower iso and from the looks of it its below 800 somewhere

Looking at the raw file
iso 100 on the left ----iso 800 on the right


And the real problem is not the iso 800 shot, its the iso 800 shot when I want to ignore setting the iso any higher and setting the image brightness and clipping point after the fact, having the NR kicking in at or around the time the camera becomes isoless gives me choices as how to process the image, now your are stuck with baked in NR at iso 800. How that baked in NR will look when I lift that NR processing 1, 2 stop will only be more apparent as I proceed to use the camera as an isoless system.

[reh321]

It is surprising that the accelerator kicks in as early as iso 800 considering the quality of the sensor. Anyone looking should be able to see evidence of noise reduction in the samples provided. Cooked raws are increasingly common it seems. There must be something driving this development. Review sites or users? Who knows?

I guess the manufacturer have better data on sensor noise and are thus able to produce better denoise defaults. However this is only the case because the manufacturers fail to share the data. And the downside is that you can't turn it off when you prefer noise to avoid loosing detail. It's an unfortunate way to proceed but most camera brands seem to have adopted it.

Suppose the 'new and improved' sensor were just the previous generation sensor with a new layer added, and that new layer implemented the 'accelerator'. It would be a new piece of silicone which emitted pixels with less noise than the previous generation sensor had, and I'm not convinced you could demonstrate that "cooking" had occurred. People are too focused on 'pure' sometimes.

[Ian Stuart Forsyth]

I got my hands on some K1 mrk 2 files this evening from a friend and was able to build a profile for the mark 2
http://isfphotography.com/pentaxk1mr2ian1.dcp
Number 1 profile was created from a file that was underexposed for what I would like when creating a profile


This second is still underexposed when creating the profile but better than the first profile, normally I would use an exposure that is as large as I can but beggars cant be choosers,

http://isfphotography.com/pentaxk1mr2ian2.dcp

---

Last edited by Ian Stuart Forsyth; 04-23-2018 at 10:01 PM.

[acoufap]

My impression: depending on how I look on the comparison images I see slight differences or not. Focusing of my eyes seem to have a bit of impact - and of course my brain. Funny thing. The differences - nothing I‘d bother about.

Here something to think about.

[Class A]

here is no such thing in any RAW development software (RawTherapee included - and I have used it, occasionally still do for testing) as an import that doesn't involve a demosaicing algorithm (it has to use something, right?).

You could use dcraw to create an undemosaiced TIFF file from a RAW file. You wouldn't be able to visually inspect it in a meaningful way, but you could run analyses on it.

What I am seeing is that the areas that are slightly out of focus with any tonal variations start to smear into 1 uniform glob with no detail, something that one would expect from rather generous NR while the small detail appears to have some sharpening a applied.

We should not be comparing one image that hasn't received noise reduction (ISO 100) with one that has (ISO 800). Just the presence of noise alone in the ISO 100 shot could create the illusion of more detail. I've seen this effect many times. Just add a tiny bit of noise in post to your images, and they'll look noisier but also more detailed.

Also, to remove all the bad discolouration in the ISO 100 shot, some kind of processing is necessary. It remains to be demonstrated that there are options available today that are superior to what Ricoh appears to be doing with their processing. It is a bit difficult to see this thread grow and grow when simple solutions are available (direct comparison to K-1 and apples-to-apples comparisons).

I fully agree with you that the effect of the accelerator chip should be optional. Even if there is no better approach to NR today, in the future perhaps even better methods can be devised that would require the unprocessed RAW data. AFAIC, the accelerator chip should be optional at all ISO levels.

[BigMackCam]

You could use dcraw to create an undemosaiced TIFF file from a RAW file. You wouldn't be able to visually inspect it in a meaningful way, but you could run analyses on it.

That's an interesting thought... Honestly, I have no experience with dcraw - something to add to my learning list

[mohb]

AFAIC, the accelerator chip should be optional at all ISO levels.

yet another niche camera idea

[Kunzite]

We should not be comparing one image that hasn't received noise reduction (ISO 100) with one that has (ISO 800). Just the presence of noise alone in the ISO 100 shot could create the illusion of more detail. I've seen this effect many times. Just add a tiny bit of noise in post to your images, and they'll look noisier but also more detailed.

Right. And IMHO this is exactly what the OP did. Not pointing out fingers - it took me a few tries to properly match sharpening between the pictures.
When comparing I'm trying to make sure I'm looking for actual threads, and not noise. It's very difficult, because all the noise here is made about areas where those threads cannot be resolved.

[Rondec]

You could use dcraw to create an undemosaiced TIFF file from a RAW file. You wouldn't be able to visually inspect it in a meaningful way, but you could run analyses on it.


We should not be comparing one image that hasn't received noise reduction (ISO 100) with one that has (ISO 800). Just the presence of noise alone in the ISO 100 shot could create the illusion of more detail. I've seen this effect many times. Just add a tiny bit of noise in post to your images, and they'll look noisier but also more detailed.

Also, to remove all the bad discolouration in the ISO 100 shot, some kind of processing is necessary. It remains to be demonstrated that there are options available today that are superior to what Ricoh appears to be doing with their processing. It is a bit difficult to see this thread grow and grow when simple solutions are available (direct comparison to K-1 and apples-to-apples comparisons).

I fully agree with you that the effect of the accelerator chip should be optional. Even if there is no better approach to NR today, in the future perhaps even better methods can be devised that would require the unprocessed RAW data. AFAIC, the accelerator chip should be optional at all ISO levels.

I guess I'm not totally convinced that there is any negative effect going on here. Certainly it is quite subtle and people are really grasping to try to make a big deal about it. Certainly this is no "star eater" situation. There may be some smoothing done to out of focus areas, but even there it isn't a huge effect and one that is only seen by looking at images at 100 percent.

But what is the point of spending an extra couple of hundred dollars on a K1 II if not for the accelerator chip? If people are that concerned then they should probably just get a K-1 and be done with it.

I will say that I am more looking forward to comparisons between the K-1 and K-1 II than to iso 100 files that have been pushed three stops and iso 800 files. I suppose it helps to figure out what the accelerator chip is actually doing, but in the end, photography is about making images and it seems unlikely that this effect (scarcely visible at the pixel level) is going to detract from the image's quality.

[Andrea K]

No one else has the m2 yet? Seems like a rather easy test to do, and a lot of people would be genuinely interested in the results, I bet a lot of people are weighing their decisions to upgrade to m2 (doesn't matter which way, new body or upgrade).

You mean comparison tests? To do the right choice? This is funny, here,for many, the voice of the ambassador was enough ....

[beholder3]

That's an interesting thought... Honestly, I have no experience with dcraw - something to add to my learning list

It seems dcraw has been discontinued.
The modern successor seems to be libraw.

[BigMackCam]

You mean like this: Menu item "Noise Reduction" (see K-1 Mk II Manual pp 50, 55)?

It permits fine tuning (even turning off) the in-camera noise reduction feature for various ISO levels.

Is that for RAW too, or just JPEG?

[stevebrot]

You could use dcraw to create an undemosaiced TIFF file from a RAW file. You wouldn't be able to visually inspect it in a meaningful way, but you could run analyses on it.

Or one could simply use dcraw to create a linear TIFF with zero sharpening and one's choice of options for interpolation. Such is readily viewable, though I have found that one must take care with viewing software. Lightroom, for example, will remap tones from a linear TIFF to even out and center the histogram. How deep does this rabbit hole go?


Steve

[aikaarska]

Is that for RAW too, or just JPEG?

Yes, it's for jpg files. RAW should be just RAW, not cooked stuff.