New Information on K3III

Bill Claff has some information on his site - see https://www.photonstophotos.net/Charts/PDR.htm

Bill's graphs do not normalize ISO - he uses the vendors ISO. So its hard to compare DR across brands. However within brands its not a stretch to say that the ISO calibration is similar model to model. So the comparison of the K3III to a K1II or KP would be fairly representative of performance differential.

The DR points before the Accelerator adds NR looks similar (not identical) to the K3 I, II and KP. It appears that the Accelerator kicks in sooner than in the KP and is stronger in its application.  It appears 8 stops of DR (or more) is available all the way up to ISO 3200 (4000)

I would opine that the the DPReview Image Comparison tool would be somewhat similar to what we see in the KP but more so and a bit sooner in the ISO range. In fact the line looks very similar to the K1 II after ISO 800. That is not a bad thing at all!

For JPeg shooters it will be great. For raw shooters with access to state of the art NR tools like DXO, maybe not so much.  Still the performance in DR and noise is unlikely to disappoint!

https://www.dpreview.com/forums/thread/4571492

^ There was a previous thread about this.

The Squirrel Mafia wrote:

https://www.dpreview.com/forums/thread/4571492

^ There was a previous thread about this.

So the question is, how did they manage a score that's better than perfect?  They have some kind of super-secret NR going on, they won't describe it and they won't let you turn it off or bypass it.

Mark Ransom wrote:

The Squirrel Mafia wrote:

https://www.dpreview.com/forums/thread/4571492

^ There was a previous thread about this.

So the question is, how did they manage a score that's better than perfect? They have some kind of super-secret NR going on, they won't describe it and they won't let you turn it off or bypass it.

It's signal processing that they have going on which happens to include NR as well, but there's more to it than just straight NR. Color correction at high ISO is also part of the signal processing. I just did a test comparing ISO from 100 - 1.6 million. (I skipped 12,800 by accident and went from 10,000 -25,600) between jpeg at default camera NR settings and DNG raw. The amount of detail the raw maintains is quite amazing and doesn't smear at high ISO how the jpegs do. The jpegs also seem to drop the red/magenta tone higher up in the ISO chain  (25,600). Loss of detail was much less evident with the raw files in comparison.

I'll upload them to my Flickr account tomorrow so you can all have a look.

I asked the question to Bill, how he comes to that perfect DR curve, and he explained everything about the slope (which is absolutely clear) and that shot noise DR is poisson distributed resulting in a SNR of Sum(e)/Sqrt(Sum(e)) = SQRT(Sum(e)).

So the totplolgy of the curve is well explained and quite clear. ( i am also developping cameras for driver montioring and traffic sign recognition systems in automtive).

But I have no clue about the offset of the curve, why DR is limited to 11.57EV @ ISO 100. Is this based on a total electron flux? For me , that induces that there is an upper limit of FWC depending on sensor size, as with no read noise, DR translates to SQRT(FWC).

By the way, an equivalent full frame lens has ~2,25times the entrance pupil of an APSC-lens, thus 2,25x higher light flux. But this is projected on a 2,25x higher area, so the luminance denisty is the same. Per pixel you are integratin 2,25x the light assuming the same resolution, so that's why the DR of FF is ld(2,25) above that of APS-C. Everything clear, but the explanation of the absolute values is somehow missing. Unfortunately, the thread did not allow furhter posts, otherwise I had asked bill for that.

BR

Markus

Kobie M-C wrote:

It's signal processing that they have going on which happens to include NR as well, but there's more to it than just straight NR. Color correction at high ISO is also part of the signal processing. I just did a test comparing ISO from 100 - 1.6 million. (I skipped 12,800 by accident and went from 10,000 -25,600) between jpeg at default camera NR settings and DNG raw. The amount of detail the raw maintains is quite amazing and doesn't smear at high ISO how the jpegs do. The jpegs also seem to drop the red/magenta tone higher up in the ISO chain (25,600). Loss of detail was much less evident with the raw files in comparison.

Colour-correction is incredibly valuable! I can deal with noise als long as it's sort-of uniform, but I highly dislike colour-noise when I want to make a colour-pic.

Let's hope that this tech will also be put into the K-1 sucessor.

Lotz wrote:

I asked the question to Bill, how he comes to that perfect DR curve, and he explained everything about the slope (which is absolutely clear) and that shot noise DR is poisson distributed resulting in a SNR of Sum(e)/Sqrt(Sum(e)) = SQRT(Sum(e)).

So the totplolgy of the curve is well explained and quite clear. ( i am also developping cameras for driver montioring and traffic sign recognition systems in automtive).

But I have no clue about the offset of the curve, why DR is limited to 11.57EV @ ISO 100. Is this based on a total electron flux? For me , that induces that there is an upper limit of FWC depending on sensor size, as with no read noise, DR translates to SQRT(FWC).

I think this page from Bill's site is what you're looking after:

https://www.photonstophotos.net/GeneralTopics/Sensors_&_Raw/Sensor_Analysis_Primer/Engineering_and_Photographic_Dynamic_Range.htm

By the way, an equivalent full frame lens has ~2,25times the entrance pupil of an APSC-lens, thus 2,25x higher light flux. But this is projected on a 2,25x higher area, so the luminance denisty is the same. Per pixel you are integratin 2,25x the light assuming the same resolution, so that's why the DR of FF is ld(2,25) above that of APS-C. Everything clear, but the explanation of the absolute values is somehow missing. Unfortunately, the thread did not allow furhter posts, otherwise I had asked bill for that.

BR

Markus

Sjak wrote:

Kobie M-C wrote:

It's signal processing that they have going on which happens to include NR as well, but there's more to it than just straight NR. Color correction at high ISO is also part of the signal processing. I just did a test comparing ISO from 100 - 1.6 million. (I skipped 12,800 by accident and went from 10,000 -25,600) between jpeg at default camera NR settings and DNG raw. The amount of detail the raw maintains is quite amazing and doesn't smear at high ISO how the jpegs do. The jpegs also seem to drop the red/magenta tone higher up in the ISO chain (25,600). Loss of detail was much less evident with the raw files in comparison.

Colour-correction is incredibly valuable! I can deal with noise als long as it's sort-of uniform, but I highly dislike colour-noise when I want to make a colour-pic.

Let's hope that this tech will also be put into the K-1 sucessor.

Yes .. let's hope it is in the K-1 III, however, with the possibility to control the process. You know, give the user the option to use the camera creatively, choose what functions are on and off .. like with so many other features on Pentax cameras! That is why we love them ... it's the flexibility!

MarBa wrote:

Yes .. let's hope it is in the K-1 III, however, with the possibility to control the process. You know, give the user the option to use the camera creatively, choose what functions are on and off .. like with so many other features on Pentax cameras! That is why we love them ... it's the flexibility!

I'm not quite sure what kind of feature you refer to, possibly due to my ignorance on the subject.

Do you have examples of camera's with features that allow tuning of the proces of the signal-to-raw-file?

As for using the camera creatively, I believe that's the user's responsibility, not the manufacturer's

I am not so sure, as this is all based on the Nikon D300 photon transfer curve, which for sure does not represent an ideal form of a sensor.

He is stating something of 14EV as an endpoint, which might result from 14bit AD-values, but take into account, that for a white level, you have 3 stimuli adding to the total value, not just one! So maybe, there is a value of ld(3) missing. Green is not white!

Also in display metrology, white is way brighter than one of the three primaries, which sum up to white, in the end.

Color matrix calculations always take into account all three primaries for the total integral flux, that's why color matrices are 3x3 matrices.

BR

Markus

I couldn't upload the DNG's to Flickr so here's a google drive link to the folder of all 30 images jpeg and DNG at default NR levels right out of the camera.

https://drive.google.com/drive/folders/1JAUo-csVGKrYh6s7_6gMI-gUre793v0P?usp=sharing

Sjak wrote:

As for using the camera creatively, I believe that's the user's responsibility, not the manufacturer's

Yes, but that's limited by the aspects of the camera that can be controlled by the user.

Imagine if Pentax decided they had perfected auto exposure and removed the ability to set your own aperture or shutter speed.  Do you think anybody would accept that?

Hi Bill,

thanks for your pateint replies. That can take me a step further

Do you have a soruce, where I could find that formular correlating ISO to total flux? that would be very interesting for me.

Thanks in advance

Markus

Most of the cameras have noise reduction turned off, but some Pentax cameras have forced noise reduction. Why they are listed as equals in that table? At least put an asterisk mark next to them or something.

The thing about noise reduction is that people have different preferences and algorithms evolve all the time. Those Pentax cameras do not allow you to enjoy from that flexibility.

Thanks, Bill.

bclaff wrote:

Fotoni wrote:

Most of the cameras have noise reduction turned off, but some Pentax cameras have forced noise reduction. Why they are listed as equals in that table? At least put an asterisk mark next to them or something.

...

Interesting idea.
On the charts there are symbols to indicate things like noise reduction but the tables doesn't take this into account.
I'll think about this.

When do you qualify it as noise reduction?

Pentax has a history/reputation of getting better results out of a sensor (e.g., on K-5, K-1) than other camera manufacturers (e.g., Nikon, Sony).  This well pre-dates the accelerator chip. So they are doing better signal processing than others.  Is this noise reduction?

In the past this was seen as a selling point, not a concern.

Have they simply upped their processing and with the accelerator chip are simply applying it to varying degrees at different ISOs?

Measurements might suggest that X is going on, but maybe it is Y...

My thoughts,

Doug