K20D vs D300 shadow noise and DR

Started Feb 27, 2008 | Discussions
*isteve Veteran Member • Posts: 9,509
K20D vs D300 shadow noise and DR

I managed to blag both cams for about 5 mins each this afternoon. After all the hoo-ha about the D300 supposedly having a noise floor thats 2 stops lower than the K20D, I cant say it makes much difference in practice if its true.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

-- hide signature --

Steve
Pixel peepers miss the big picture.
http://www.pbase.com/steve_jacob

Lance B
Lance B Forum Pro • Posts: 33,881
Re: K20D vs D300 shadow noise and DR

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

If that's 2 stops difference in noise then I would have to agree. In practice it makes FA difference.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

-- hide signature --

Steve
Pixel peepers miss the big picture.
http://www.pbase.com/steve_jacob

-- hide signature --

Lance B

Originator of the term 'LBA'.

http://www.pbase.com/lance_b
GMT +11hours

 Lance B's gear list:Lance B's gear list
Nikon Z7 II Nikon Z9 Nikon AF-S Teleconverter TC-17E II Nikon AF-S Teleconverter TC-20E III Nikon AF-S Nikkor 80-400mm f/4.5-5.6G ED VR +19 more
OP *isteve Veteran Member • Posts: 9,509
Re: K20D vs D300 shadow noise and DR

Lance B wrote:

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

If that's 2 stops difference in noise then I would have to agree. In
practice it makes FA difference.

Is that FA* or just regular FA? Or DA for that matter.

First Pentax photographer....
"Ive got FA lenses"

Second Pentax photographer
"Ive got DA lenses"

Canon photographer
"Whats wrong with you guys, Ive got hundreds of them.... "

But yeah it wasnt a great test but I cant see a whole bunch of difference.

-- hide signature --

Steve
Pixel peepers miss the big picture.
http://www.pbase.com/steve_jacob

SFT007 Veteran Member • Posts: 5,874
Re: K20D vs D300 shadow noise and DR

Thanks for the test Steve, I'm not interested in a new camera yet but I am interested to see how this CMOS sensor performs as I will no doubt eventually upgrade my K10D when the time comes. I agree about looking at the results and not just concentrating on numbers and doing just that - well putting aside the better AWB performance on the D300 I like that it basically eliminated color (chroma) noise very very well and the luminance noise seems to have a tighter grain which I also like. Right now we are comparing 12MP vs 14MP and to put it simply - the results are as one would expect but maybe the K20D sensor indeed does deliver noise closer to the level of a 12MP sensor which is nice! The WB difference is something else - I don't doubt the D300 is overall generally a slightly superior camera to the K20D but the K20D when one knows what they are doing will yield excellent results I am sure so in the end thanks to the unique sensor design here with Samsung/Pentax, having 14.6MP's on APS-C isn't as bad as most would have believed just only a year ago! That's progress.

-- hide signature --
 SFT007's gear list:SFT007's gear list
Olympus TG-6 Olympus OM-D E-M5 Panasonic Lumix G Vario 7-14mm F4 ASPH Panasonic Lumix G Vario 100-300mm F4-5.6 OIS Panasonic Leica Summilux DG 25mm F1.4 +4 more
adrian4765 Regular Member • Posts: 211
Re: K20D vs D300 shadow noise and DR

I think he's using a different f.a. to what you're thinking of...

 adrian4765's gear list:adrian4765's gear list
Pentax K-3
SFT007 Veteran Member • Posts: 5,874
FA - F*ck All! Come on Steve - I thought you were British! :) (nt)

adrian4765 wrote:

I think he's using a different f.a. to what you're thinking of...

-- hide signature --
 SFT007's gear list:SFT007's gear list
Olympus TG-6 Olympus OM-D E-M5 Panasonic Lumix G Vario 7-14mm F4 ASPH Panasonic Lumix G Vario 100-300mm F4-5.6 OIS Panasonic Leica Summilux DG 25mm F1.4 +4 more
OP *isteve Veteran Member • Posts: 9,509
Re: K20D vs D300 shadow noise and DR

adrian4765 wrote:

I think he's using a different f.a. to what you're thinking of...

No, I dont think he is...;)

-- hide signature --

Steve
Pixel peepers miss the big picture.
http://www.pbase.com/steve_jacob

OP *isteve Veteran Member • Posts: 9,509
Read my post again....

I do know what f*ck all and d*mn all means....
--
Steve
Pixel peepers miss the big picture.
http://www.pbase.com/steve_jacob

SFT007 Veteran Member • Posts: 5,874
Haha. Touche I hear ya say :) (nt)

*isteve wrote:

I do know what f*ck all and d*mn all means....
--
Steve
Pixel peepers miss the big picture.
http://www.pbase.com/steve_jacob

-- hide signature --
 SFT007's gear list:SFT007's gear list
Olympus TG-6 Olympus OM-D E-M5 Panasonic Lumix G Vario 7-14mm F4 ASPH Panasonic Lumix G Vario 100-300mm F4-5.6 OIS Panasonic Leica Summilux DG 25mm F1.4 +4 more
adrian4765 Regular Member • Posts: 211
Re: Read my post again....

oh very clever. i completely missed it!

 adrian4765's gear list:adrian4765's gear list
Pentax K-3
GordonBGood Veteran Member • Posts: 6,312
Re: K20D vs D300 shadow noise and DR

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

Steve, I conceed that the difference in noise isn't very much for most people's use of the K20D including the use of SilkyPix, but I think it would be interesting to run those same two images through my software and the results posted side by side with your images as per the above link. From this we would not only learn about any potential limitations of the K20D as compared to its major competitors, but would also learn something more about the processing of SilkyPix as to application of noise reduction.

I think this is particularily relevent with Herr_Bob's observation that the noise pattern looks grainier and less pleasing in your K20D patches than the D300 patches, which may indicate that more NR was automatically applied by SilkyPix.

As per the request on your referenced thread, could you please send the images to me via a service such as YouSendIt?

Since the other thread is likely to be full by the time this gets done, I would post the results here with all due credit to you.

Regards, GordonBGood

LuzArt Veteran Member • Posts: 6,643
Re: K20D vs D300 shadow noise and DR

Sinan Tarlan wrote:

Thanks for the test Steve, I'm not interested in a new camera yet but
I am interested to see how this CMOS sensor performs as I will no
doubt eventually upgrade my K10D when the time comes. I agree about
looking at the results and not just concentrating on numbers and
doing just that -

I'll check back in 3 months...

Ben
--

http://www.luzartphotography.com
http://luzart.blogspot.com

 LuzArt's gear list:LuzArt's gear list
Nikon D700 Nikon AF-S Nikkor 70-200mm f/2.8G ED VR II Nikon AF-S Teleconverter TC-20E III Tamron AF 28-75mm F/2.8 XR Di LD Aspherical (IF) +2 more
GordonBGood Veteran Member • Posts: 6,312
Re: K20D vs D300 shadow noise and DR

Steve, as anticipated, the other thread is full, so I'll take the liberty of copying your message across to this one, as follows:

GordonBGood wrote:

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

Copied across as follows:

"Gordon I have no objection to anyone finding limits, but a lot of people take information out of context and assume something to be dire when it clearly is not. A real example helps put it into context. I hope more people provide more examples, but all I can say is using an of the shelf RAW converter does not seem to show up a massive difference.

From the original file, you should be able to tell if the -2 patch is 9 stops under perhaps? If there any way I could tell?

I cant explain the shift from +2 to +3. I am quite happy to put the raw files up if you want to try them yourself but I wont be able to do so till this time tomorrow. Hope its OK."

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

True, people do take things out of context and providing images is good, which is why I would like to take a comparative processing stab at the same data, with output as images.

I'm thinking that even with NR off, the SilkyPix demosiacing engine tends to do a lot of work in trying to "connect the dots" in noisy images (as in producing quite a lot of "mazing artifacts") and this may be acting as a noise reduction/detail resolution limit in noisy images such as these EV boosted ones.

Yes, from the original file I would do a linear conversion (no tone curve applied) and would be able to tell the exact relative luminance levels of your dark areas in each of the images. You should be able to do the same with available tools, but the problem is that neutral or flat tone response curves sometimes are just relative to another curve that isn't entirely linear. dcraw has options to do it if you know how to use that.

When I receive the files, I will report on these true relative luminance levels for the patches as well as do my alternative processing and give the luminance and chrominance standard deviations for the dark patches. Sending the files tomorrow is fine, as I really should do some other work in the meantime

Regards, GordonBGood

Steve, I conceed that the difference in noise isn't very much for
most people's use of the K20D including the use of SilkyPix, but I
think it would be interesting to run those same two images through my
software and the results posted side by side with your images as per
the above link. From this we would not only learn about any
potential limitations of the K20D as compared to its major
competitors, but would also learn something more about the processing
of SilkyPix as to application of noise reduction.

I think this is particularily relevent with Herr_Bob's observation
that the noise pattern looks grainier and less pleasing in your K20D
patches than the D300 patches, which may indicate that more NR
was automatically applied by SilkyPix.

As per the request on your referenced thread, could you please send
the images to me via a service such as YouSendIt?

Since the other thread is likely to be full by the time this gets
done, I would post the results here with all due credit to you.

Regards, GordonBGood

GordonBGood Veteran Member • Posts: 6,312
Re: K20D vs D300 shadow noise and DR

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

Steve, I just downloaded and put the above referenced image into Photoshop (should have done this before) to confirm your numbers. I see that your Standard Deviation (SD) numbers have a small problem in that they are calculated over all three RGB channels, which are thus then subject to the scatter of the three channels relative to each other as seen by substantial changes when you change WB.

Better analyical numbers would have been obtained by using a SD based on luminance, in which case the K20D has consistently about half again the noise as the Nikon D300 at whatever light levels these patches represent. They appear to be somewhat brighter than the ISO 100 crops I provided earlier, which likely explains the less dramatic difference than as I showed as we have discussed.

Regards, GordonBGood

TheLastMan Regular Member • Posts: 436
Re: K20D vs D300 shadow noise and DR

Thanks for the pictures Steve. Yes, to all intents and purposes, the two are identical - and certainly when printed there will be no noticeable difference.

As I state on the other thread I think Gordon (who is clearly leagues ahead of most of us in his in-depth technical knowledge and expertise) is possibly over-estimating the concern that the vast majority of photographers might have over a very minor difference in DR.

Because of the limited gradation in the lower stops of a 12bit RAW file, only the top 6 stops (0 to -5ev) are really useable anyway, regardless of what is happening at -10ev.

When in-camera processor and data storage speeds are advanced enough to cope routinely with 20MB+ image files, we might more generally see 14bit RAW files available on DSLRs which will allow another 1-2 stops of useable data to be extracted from an image file.

If Moores law continues at its current pace (big if!), we will eventually see 16 bit RAW files coming from DSLRs. I would guess these to be 50MB+. A 16 bit RAW file would have as many levels of gradation at -5ev as a 12 bit file has at-1ev. "Pushing the shadows" would be a much more feasible prospect. When that is possible, the SNR of the sensor at its lower stops will really start to matter. Until that glorious day, the K20D sensor has more than enough dynamic range to produce 100iso 12 bit RAW files comparable with its major competitors.

That is enough from me on that subject. Off to Kew Gardens on Sunday to take some snaps of a large collection of Moore's sculptures (Henry's not Gordon's!).

I doubt the shallow DR of my K100D will be the main factor limiting my photographic creativity!

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

 TheLastMan's gear list:TheLastMan's gear list
Pentax K-3 Mark III Sigma 17-50mm F2.8 EX DC OS HSM HD Pentax DA 21mm F3.2 AL Limited HD Pentax DA 35mm F2.8 Macro Limited HD Pentax DA 70mm F2.4 AL Limited +7 more
Richard Smals
Richard Smals Senior Member • Posts: 1,929
D300 RAW NO NR...5 examples.

I shot today in a store with the D300 in RAW to finally get some first hand experience in the high ISO regions. We all probably now that Nikon is very smart in the NR department so i wanted to know the real noise..therefore shot in RAW.

5 100% examples included: Developed in ACR with NR and sharpening off. Exif in files. Store lights mixed with sunny outside.

ISO 400

ISO 800

ISO 1600

ISO 3200

ISO 6400

This is thus the real sensor noise the D300 produce. In relative normal inside light. I don't know how the noise will look light in darker conditions.

-- hide signature --

Bye4now

http://www.indots.nl
http://www.indots.nl/foto

I have the deepest respect for all those people who like me.

RPulley Senior Member • Posts: 1,175
Re: K20D vs D300 shadow noise and DR

GordonBGood wrote:

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

Steve, I just downloaded and put the above referenced image into
Photoshop (should have done this before) to confirm your numbers. I
see that your Standard Deviation (SD) numbers have a small problem in
that they are calculated over all three RGB channels, which are thus
then subject to the scatter of the three channels relative to each
other as seen by substantial changes when you change WB.

Better analyical numbers would have been obtained by using a SD based
on luminance, in which case the K20D has consistently about half
again the noise as the Nikon D300 at whatever light levels these
patches represent. They appear to be somewhat brighter than the ISO
100 crops I provided earlier, which likely explains the less dramatic
difference than as I showed as we have discussed.

Regards, GordonBGood

With all due respect, Gordon, wasn't it you that has told us many times that the Nikon files are noise reduced, but now believe that they are not?

The more I read of all of this, the more I trust my own eyes and my own results (that would be images and prints, not estimated or extrapolated noise counts) over pseudo scientific analysis of the performance.

In fact, looking at the D300 ISO sequence by Richard elsewhere in this thread tells me all I need to know about the D300 High ISO performance. That does not make it a bad camera, or even worse or better than camera A or B, but the proof is in the pudding, as the saying goes, and the pudding is images in this case.

Ray

TheLastMan Regular Member • Posts: 436
Re: K20D vs D300 shadow noise and DR

RPulley wrote:

With all due respect, Gordon, wasn't it you that has told us many
times that the Nikon files are noise reduced, but now believe that
they are not?

Everyone makes mistakes. At least Gordon makes the effort to try and find out. He even writes his own RAW development software for that purpose for goodness sake!

The more I read of all of this, the more I trust my own eyes and my
own results (that would be images and prints, not estimated or
extrapolated noise counts) over pseudo scientific analysis of the
performance.

Less of the "pseudo", Gordon's tests are very definitely scientific.

In fact, looking at the D300 ISO sequence by Richard elsewhere in
this thread tells me all I need to know about the D300 High ISO
performance. That does not make it a bad camera, or even worse or
better than camera A or B, but the proof is in the pudding, as the
saying goes, and the pudding is images in this case.

Actually Ray, the phrase is "the proof of the pudding is in the eating". In other words you have to eat the pudding to find out what it tastes like - and you are right. All that ultimately matters is the performance of the camera in the hands of a photographer.

Even though I have an issue with Gordon's sense of perspective over all this, his tests are valid and his posts informative, interesting and educational.

In this instance I think his findings are insignificant in a practical sense and probably not worth spending more time over. However I am glad somebody else was doing the tests and not me!

 TheLastMan's gear list:TheLastMan's gear list
Pentax K-3 Mark III Sigma 17-50mm F2.8 EX DC OS HSM HD Pentax DA 21mm F3.2 AL Limited HD Pentax DA 35mm F2.8 Macro Limited HD Pentax DA 70mm F2.4 AL Limited +7 more
herr_bob Contributing Member • Posts: 607
Re: K20D vs D300 shadow noise and DR

You are correct. I found the same problem and replied in the original link.

If I inspect three channels separately, these numbers vary wildly. The reason behind this is simple: the sample patch is not pure gray or black, so each channel has different readings. The green channel has large numbers while the blue channel has small numbers. If I calculate all channels together, I mess large ones with small ones and get an erroneous SD reading.

I could check three channels. But here I use the way this website use. This means I convert the image to Lab color mode and treat luminosity and chroma channels separately. Now even if the sample patch is not gray or black, as long as it is uniform in color, both a and b channels are close to constants and therefore the "unexpected" part of the standard deviation disappears.

Here is what I found(I use a 80x80 sample size on the bottom right corner):

In the luminosity channel, D300 SD=3.42 while K20D SD=4.69. After WB correction, K20D SD=4.50. After +3ev, D300 SD=5.49 and K20D SD=7.34. The ratio is around 1.31-1.37 regardless of digital exposure compensation.

In the a channel (green-magenta), D300 SD=4.77 and K20D SD=8.02 or 7.82 after WB correction. After +3ev, D300 SD=7.96 and K20D SD=12.87. The ratio is around 1.62-1.68.

In the b channel (blue-yellow), D300 SD=3.87 and K20D SD=5.30 or 5.81 after WB correction. After +3ev, D300 SD=7.82 and K20D SD=10.87. The ratio is around 1.37-1.50.

How can we interpret these numbers? From the latest E-3 review
http://www.dpreview.com/reviews/olympuse3/page19.asp

a change of SD by 50% could be interpreted as 1 stop advantage/disadvantage in high ISO or 2 stop advantage/disadvantage in low ISO. This is not unexpected since a digital exposure compensation (means multiply by 2 for 1 stop) corresponds to a multiplication of SD number by square root of 2=1.4142 if no NR algorithm is implemented. We can see this roughly at the tail of all SD curves above, where ISOs are boosted. However the hardware ISO implementation is better than this. This is why the slope of the curve is far less than 1.4 at ISO lowered than 1600.

Let's come back to our comparison. This means the D300 has at least 2/3 stop of advantage in luminosity noise and more than 1 stop of advantage in chroma noise. This is by no means the conclusion because it could vary among different RAW converters/algorithms and samples. Moreover, this is only a ISO200 test.

By the way, I also want to mention another feature of noise where the SD number doesn't and cannot cover--the grain pattern. Finer grains introduce pleasing visual feelings while coarse grains normally do the opposite. This is also what we see in the D300/K20D comparison. However, this is partly offset by the fact that K20D is 14MP and D300 is 12MP. So I cannot comment on this on papers. The photographers should have their own "acceptable ISO" based on final products, whether web albums or prints.

GordonBGood wrote:

*isteve wrote:

I managed to blag both cams for about 5 mins each this afternoon.
After all the hoo-ha about the D300 supposedly having a noise floor
thats 2 stops lower than the K20D, I cant say it makes much
difference in practice if its true.

http://forums.dpreview.com/forums/read.asp?forum=1036&message=26957784

Steve, I just downloaded and put the above referenced image into
Photoshop (should have done this before) to confirm your numbers. I
see that your Standard Deviation (SD) numbers have a small problem in
that they are calculated over all three RGB channels, which are thus
then subject to the scatter of the three channels relative to each
other as seen by substantial changes when you change WB.

Better analyical numbers would have been obtained by using a SD based
on luminance, in which case the K20D has consistently about half
again the noise as the Nikon D300 at whatever light levels these
patches represent. They appear to be somewhat brighter than the ISO
100 crops I provided earlier, which likely explains the less dramatic
difference than as I showed as we have discussed.

Regards, GordonBGood

RPulley Senior Member • Posts: 1,175
Re: K20D vs D300 shadow noise and DR

TheLastMan wrote:

RPulley wrote:

With all due respect, Gordon, wasn't it you that has told us many
times that the Nikon files are noise reduced, but now believe that
they are not?

Everyone makes mistakes. At least Gordon makes the effort to try and
find out. He even writes his own RAW development software for that
purpose for goodness sake!

Agreed, and no slight or insult to Gordon was intended (he is also not the only one that performs these pseudo-scientific tests, btw).

The more I read of all of this, the more I trust my own eyes and my
own results (that would be images and prints, not estimated or
extrapolated noise counts) over pseudo scientific analysis of the
performance.

Less of the "pseudo", Gordon's tests are very definitely scientific.

I must disagree.

He is evaluating shots taken under uncontrolled conditions, as is everyone else.

Is the light source color calibrated and it's power source tightly regulated so as to make sure that each shot receives exactly the same exposure and color temp?

Is the target in the image a calibrated target with known color and reflective properties?

Are the cameras set to exactly the same white balance or is the PP software set to produce exactly the same tone curve for each?

Are the cameras setup in exactly the same location with all possible variables related to the camera position made as identical as possible?

Are the temperatures of the sensors and camera bodies the same for each camera?

I could go on into the variables of the processing tools for different cameras and the lenses and so on, but please, these are not scientific evaluations by any stretch of the imagination. In fact, I believe that some of the discussion above is related to exactly my point, in that no one seems to know if the tone curve in Silkypix is linear or if it applies any NR even if it is turned off and so on....

I find the discussion as interesting as the next person, perhaps even more so than many, but I think what some others are having problems with is the absolute authoritative tone that these discussions take on, and how others who might not be so technically inclined view the discussions.

To be fair, Gordon has mentioned many times that most will never see this or care about it, but that is like the retraction printed on page 99. No one will get that far.

This is the reason I even mentioned the Nikon noise reduction issue. That little "mistake" has been widely accepted as gospel by many, and apparently, it simply is not true, or maybe it is....

In fact, looking at the D300 ISO sequence by Richard elsewhere in
this thread tells me all I need to know about the D300 High ISO
performance. That does not make it a bad camera, or even worse or
better than camera A or B, but the proof is in the pudding, as the
saying goes, and the pudding is images in this case.

Actually Ray, the phrase is "the proof of the pudding is in the
eating". In other words you have to eat the pudding to find out what
it tastes like - and you are right. All that ultimately matters is
the performance of the camera in the hands of a photographer.

Looking once again at Richard's Nikon D300 samples, or pudding if you will, I certainly do not see any ground-breaking noise performance there, nor anything about that performance that would lead me to buy a D300 if noise were my main concern, which seems a bit contrary to all of the measurebating and things we have heard about the D300 in these threads. Is it a fine camera that will take great images? Yes, it seems to be, but if one were to read all of the "scientific" discussion over this and not have Richard's images to look at, one could come to the conclusion that the D300 is 4x better than the K20D and some kind of wonder camera in terms of noise performance.

Worse, sans any image to peruse, you might also draw the conclusion that the K20D would produce pretty shabby images in comparison, as after all, 4x worse is a lot, right?

Even though I have an issue with Gordon's sense of perspective over
all this, his tests are valid and his posts informative, interesting
and educational.

I agree and have had several discussions with him over the last year os so, and he is always helpful and very polite.

In this instance I think his findings are insignificant in a
practical sense and probably not worth spending more time over.
However I am glad somebody else was doing the tests and not me!

In one sense, he is probably completely correct: assuming that the K20D has more low ISO noise than Camera X,Y and Z, the reviewers will pounce on that and make much ado over it, insignificant or otherwise.

Ray

Keyboard shortcuts:
FForum MMy threads