From DXOMark: More pixels offsets noise

Started May 21, 2009 | Discussions
photo nuts Senior Member • Posts: 1,364
From DXOMark: More pixels offsets noise

Just noticed this:

http://www.dxomark.com/index.php/eng/Insights/More-pixels-offsets-noise !

"Contrary to conventional wisdom, higher resolution actually compensates for noise"

Will this persuade DPReview to change their stance on high pixel density sensors since they seem to hold certain amount of respect for DXOMark.

Phil Askey Veteran Member • Posts: 9,821
Re: From DXOMark: More pixels offsets noise

"Read it in the Newspaper, must be true"

photo nuts wrote:

Just noticed this:

http://www.dxomark.com/index.php/eng/Insights/More-pixels-offsets-noise !

"Contrary to conventional wisdom, higher resolution actually
compensates for noise"

Will this persuade DPReview to change their stance on high pixel
density sensors since they seem to hold certain amount of respect for
DXOMark.

-- hide signature --

Phil Askey
Editor, dpreview.com

 Phil Askey's gear list:Phil Askey's gear list
Leica M9 Canon EOS 5D Mark III Olympus OM-D E-M5 Leica M8 Panasonic Lumix DMC-GH4 +11 more
dw1337 Regular Member • Posts: 117
Re: From DXOMark: More pixels offsets noise

"Read it in dpreview.com, must be true"

Phil Askey wrote:
"Read it in the Newspaper, must be true"

photo nuts wrote:

Just noticed this:

http://www.dxomark.com/index.php/eng/Insights/More-pixels-offsets-noise !

"Contrary to conventional wisdom, higher resolution actually
compensates for noise"

Will this persuade DPReview to change their stance on high pixel
density sensors since they seem to hold certain amount of respect for
DXOMark.

Oly Canikon
Oly Canikon Senior Member • Posts: 1,206
Re: From DXOMark: More pixels offsets noise

Its time that dpreview came forward with a proper analysis of this issue. Lets see the evidence that you base your views on and lets have some informed discussion. Not just sniper attacks.

Makinations
Makinations Veteran Member • Posts: 5,688
Re: From DXOMark: More pixels offsets noise

I think this is the dpr take on the subject ...

http://blog.dpreview.com/editorial/2008/11/downsampling-to.html

I believe it has been debunked by a few people but as comments are not enabled on that it is hard to tell.

 Makinations's gear list:Makinations's gear list
Olympus XZ-1 Canon PowerShot G9 Panasonic Lumix DMC-G3 Canon EOS 40D Olympus OM-D E-M5 +13 more
kwik Forum Member • Posts: 84
Zing!

dw1337 wrote:

"Read it in dpreview.com, must be true"

Nice one! I read this on dpreview.com though:

http://forums.dpreview.com/forums/read.asp?forum=1018&message=30211624

and it looks pretty convincing. Comments are even allowed. ; )

Richard Butler
Richard Butler dpreview Admin • Posts: 2,350
Re: From DXOMark: More pixels offsets noise

Ultimately the main dispute is between whether you measure noise at the 1:1 level or at an arbitrary image size. dpreview has always worked at the pixel level because it doesn't make any assumptions for what you're going to do with the images. As soon as we start deciding for you what size and what method you would use for outputting your image, the more quickly the reviews lose relevance - our results are consistent with each other and we provide full-size samples for you to download and process, so that you can see how things work in your workflow.

Richard - dpreview.com

Oly Canikon
Oly Canikon Senior Member • Posts: 1,206
Re: From DXOMark: More pixels offsets noise

But by choosing to stay with 1:1 you are introducing a sliding scale. You give a huge disadvantage to the higher MP camera. When what is relevant is how both cameras would do for a set (arbitrary but the same for both) output size.

Working at pixel level is misleading if the two sensors are a different size. What you are doing is effectively printing an 8x10 from one camera and a 16x20 from the other and saying " oh look this one is noisier."

By the way, I do appreciate your taking the time to respond and having this dialog. I'll keep an open mind as much as I can. Please try to convince me.

cheers

R Butler wrote:

Ultimately the main dispute is between whether you measure noise at
the 1:1 level or at an arbitrary image size. dpreview has always
worked at the pixel level because it doesn't make any assumptions for
what you're going to do with the images. As soon as we start deciding
for you what size and what method you would use for outputting your
image, the more quickly the reviews lose relevance - our results are
consistent with each other and we provide full-size samples for you
to download and process, so that you can see how things work in your
workflow.

Richard - dpreview.com

kwik Forum Member • Posts: 84
Re: From DXOMark: More pixels offsets noise

R Butler wrote:

Ultimately the main dispute is between whether you measure noise at
the 1:1 level or at an arbitrary image size. dpreview has always
worked at the pixel level because it doesn't make any assumptions for
what you're going to do with the images.

I think it's safe to say that you're going to be printing the images. Or, maybe not. Maybe just downsizing them to display on the web.

As soon as we start deciding
for you what size and what method you would use for outputting your
image, the more quickly the reviews lose relevance - our results are
consistent with each other and we provide full-size samples for you
to download and process, so that you can see how things work in your
workflow.

There's no need for "you" to "start deciding" what method anyone will use for outputting their images. DPR consistenly blasts camera manufacturers for raising the pixel count on the basis of increased noise, when, in fact, the overwhelming evidence is that increasing the pixel count is the most direct path to greater IQ.

ck3
ck3 Senior Member • Posts: 2,925
Re: From DXOMark: More pixels offsets noise

R Butler wrote:

Ultimately the main dispute is between whether you measure noise at
the 1:1 level or at an arbitrary image size. dpreview has always
worked at the pixel level because it doesn't make any assumptions for
what you're going to do with the images.

Sorry, but that's not true. While there is nothing wrong with looking at the pixel level there is a lot wrong with looking exclusively at the pixel level. A statement like "High ISO performance worse than 40D" (from the 50D conclusions) does indeed make assumptions about what the user wants to do with the high ISO images - namely compare them at 100% on his computer screen.

tkpatric Senior Member • Posts: 1,544
Re: Zing!

Noise factor is old-fashion. Take the dimension of the picture when you start to see noise as reference.
--
http://users.telenet.be/patric/

 tkpatric's gear list:tkpatric's gear list
Canon PowerShot G15 Sony a6000
BertIverson Veteran Member • Posts: 3,612
Re: From DXOMark: More pixels offsets noise

kwik wrote:

DPR consistenly blasts camera
manufacturers for raising the pixel count on the basis of increased
noise, when, in fact, the overwhelming evidence is that increasing
the pixel count is the most direct path to greater IQ.

Especially since the vast majority of digital photos are shot at low ISO . Indeed others have said that it is mainly the sensor size not pixel density that determines IQ.
Bert

 BertIverson's gear list:BertIverson's gear list
Sony RX100 Sony Cyber-shot DSC-RX10 +4 more
igb Senior Member • Posts: 2,637
Re: From DXOMark: More pixels offsets noise

R Butler wrote:

Ultimately the main dispute is between whether you measure noise at
the 1:1 level or at an arbitrary image size.

I take you mean "at an other arbitrary image size", no?

dpreview has always
worked at the pixel level because it doesn't make any assumptions for
what you're going to do with the images.

Is that the real reason? Or the fact that in 1999 digicams ouput 0.3 Mp and conclusions derived from pixel level analysis were more meaningful coupled with hard wired resistance to change?

As soon as we start deciding
for you what size and what method you would use for outputting your
image, the more quickly the reviews lose relevance

How's looking at an 15Mp image at 100% in a > 2Mp display even remotely relevant?

  • our results are

consistent with each other and

we provide full-size samples for you
to download and process, so that you can see how things work in your
workflow.

Which is good, and totally uncorrelated with whether or not you noise comparisons are useful and/or your conclusions misleading.

Imagine for a moment that you drop pixel level noise analysis, how much would your productivity increase? 20% more reviews?
--
-------------------------------------------------------
My Galleries: http://webs.ono.com/igonzalezbordes/index.html

Lee Jay Forum Pro • Posts: 51,895
Re: From DXOMark: More pixels offsets noise

R Butler wrote:

Ultimately the main dispute is between whether you measure noise at
the 1:1 level or at an arbitrary image size. dpreview has always
worked at the pixel level because it doesn't make any assumptions for
what you're going to do with the images.

Sure you are. You are assuming that the final printed or displayed image size will always be proportional to the original image pixel dimensions.

The other approach is to assume that the final printed or displayed image size will be a constant regardless of original image pixel dimensions.

You can't assume nothing, you have to assume something. Those two above are the most reasonable. You chose the former, virtually every user of their cameras chooses the later. I've never once had a client chose a print size based on the pixel dimensions of the camera with which I shot their image.

-- hide signature --

Lee Jay
(see profile for equipment)

 Lee Jay's gear list:Lee Jay's gear list
Canon IXUS 310 HS Canon PowerShot SX260 HS Canon EOS 5D Canon EOS 20D Canon EOS 550D +22 more
Henrik Andersson Senior Member • Posts: 2,454
Re: From DXOMark: More pixels offsets noise

ljfinger wrote:

Sure you are. You are assuming that the final printed or displayed
image size will always be proportional to the original image pixel
dimensions.

The other approach is to assume that the final printed or displayed
image size will be a constant regardless of original image pixel
dimensions.

You can't assume nothing, you have to assume something. Those two
above are the most reasonable. You chose the former, virtually every
user of their cameras chooses the later. I've never once had a
client chose a print size based on the pixel dimensions of the camera
with which I shot their image.

Eventually DPReview will change their mind, I guess. It's so obvious. For instance when they compare high-ISO performance of cameras.

-- hide signature --

Henrik

GaborSch Veteran Member • Posts: 7,203
Worthless results, not only DxO

DxO creates comparison results without any practical use; for example downsizing a 21Mpix and a 65Mpix image to 8 or 12 Mpix.

On the other hand, DPReview's dynamic range measurements are of great entertainment value, like when the DR is increasing with higher ISO. In fact, many if not most of their DR measurements are jokes.

GaborSch Veteran Member • Posts: 7,203
The difference

ljfinger wrote:

You can't assume nothing, you have to assume something. Those two
above are the most reasonable. You chose the former, virtually every
user of their cameras chooses the later. I've never once had a
client chose a print size based on the pixel dimensions of the camera
with which I shot their image

The difference between taking a given print size for the comparison and using the pixel level measurements is, that the latter represents an "ideal" measurement, which can be transformed in other measurement, while the former is good for nothing but the given size.

igb Senior Member • Posts: 2,637
Please walk me through

GaborSch wrote:

The difference between taking a given print size for the comparison
and using the pixel level measurements is, that the latter represents
an "ideal" measurement, which can be transformed in other
measurement, while the former is good for nothing but the given size.

the transformation of pixel level measurement into "other measurement"
--
-------------------------------------------------------
My Galleries: http://webs.ono.com/igonzalezbordes/index.html

Steen Bay Veteran Member • Posts: 6,974
Noise and resolution..

R Butler wrote:

Ultimately the main dispute is between whether you measure noise at
the 1:1 level or at an arbitrary image size. dpreview has always
worked at the pixel level because it doesn't make any assumptions for
what you're going to do with the images. As soon as we start deciding
for you what size and what method you would use for outputting your
image, the more quickly the reviews lose relevance - our results are
consistent with each other and we provide full-size samples for you
to download and process, so that you can see how things work in your
workflow.

Richard - dpreview.com

When you (DPR) measure/compare the resolution, you assume the same output-size, and compare at 'image level' (Lines Per Picture Height). Wouldn't it be reasonable to do the same, when you measure/compare the noise?

Barry Fitzgerald Forum Pro • Posts: 29,888
Re: From DXOMark: More pixels offsets noise

I don't really trust anything DxO say being blunt.

Near 11 stop compacts for DR, yet nobody manages to get that in the real world!

Keyboard shortcuts:
FForum MMy threads