Pages: 1 ... 3 4 [5]   Go Down

Author Topic: DxO marks  (Read 40742 times)

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311

I do not need to imagine, as I do use RAW+JPEG with the E-M5, and do not see any problem with either raw or JPEG files: the in-camera JPEGs have appropriate levels (not to bright or too dark), and there is no detectable problem with the raw files

open in rawdigger and see the raw histogram... but I believe that you will call that "non detectable" :-)

, which various raw convertors give similar JPEGs with default settings.

you mean converters like ACR/LR that apply hidden exposure correction (convert to DNG and see baseline exposure tag value)

And as I have tried to explain, the lower raw level placement does no measurable harm to noise levels.

that is not the issue... the effect of difference in gain itself provided that you deliver the same amount of light to the sensor is not that big (provided that you are not clipping by applying a bigger gain for example, etc) - but the issue is the one that I indicated above... if you JPG is "normal" you typically have a lot of room in all raw channels to expose more, but you can't (if you need in camera JPG and/or useable preview)... but I understand that you have no issue to leave a stop or more on the table even if you can have more light to the sensor, it is a personal call.


But perhaps by "undersaturated raw" you mean a raw file

I mean a raw file with >> 1/3 EV to clipping in all raw channels for non spec. highlights... camera meters to produce JPG and if "nominal" ISO is 1 stop ahead of "measured" then manufacturer sacrifices possibly better SNR for extra safety (highlights) for JPG shooters... not a problem if you shoot raw, if you understand what camera does, what raw converters like LR/ACR do (hidden expocorrection) and do not care about previews... then you can meter properly, dial in positive correction and pull back in raw conversion... but otherwise again you leave a lot of light on the table in the most situations... and that was happening with sensors before EM5 where sensors performance was worse...

Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
maximizing raw levels is nothing to do with maximizing sensor exposure
« Reply #81 on: January 10, 2013, 04:49:11 pm »

open in rawdigger and see the raw histogram... but I believe that you will call that "non detectable" :-)
I only care about differences that are detectable in the final displayed prints, possibly measured by SNR measurements when the sensors receives equal exposure level. I do not care about a mere moving to the left or right of the numerical levels in the raw histogram since my end goal in to look at pictures, not histograms.

Quote
you mean converters like ACR/LR that apply hidden exposure correction (convert to DNG and see baseline exposure tag value)
Again: I do not care about how the converter gets there, I only care about how the results look. I think that baseline exposure tag is there exactly to tell the raw converter where the midtones have been placed, because (despite some misunderstanding of the intent of the descriptive ISO standards) there is no prescriptive industry standard for where levels _should_ be placed in raw files, and camera makers are free to make their own choices. So this tag probably helps raw converters to produce a good default conversion.

By the way, almost any raw conversion involves an "exposure correction", since raw files almost always place midtones half a stop or more below 18%, so that a default raw conversion (flat tone curve of contrast adjustment) must be scaled up by a half stop or more. My recurring question is when and why different levels of scaling up are better or worse. And you seem to agree that it does not matter much:
Quote
that is not the issue... the effect of difference in gain itself provided that you deliver the same amount of light to the sensor is not that big (provided that you are not clipping by applying a bigger gain for example, etc)
Agreed: that has been my main point all along!

Quote
I understand that you have no issue to leave a stop or more on the table even if you can have more light to the sensor, it is a personal call.
You completely misunderstand. Please read the end of my previous post again about the difference between "ETTR" (which is what you are talking about when you mention "more light to the sensor") and the subsequent degree of amplification of that raw signal reflected in ADC output:
None of these differences in raw levels are about giving more or less exposure to the sensor;
they are about how much amplification is applied to that output when the sensor has been given equal exposure level,
and thus where the numerical levels fall in the raw files.


To repeat, the SNR curves for the E-M5 and E-PL5 show that the same ISO settings on those cameras produce the same exposure levels and the same photo-electron counts in the photosites (as shown by equality of the SNR 18% values). The differences of up to 10% in their DxO "saturation" ISO values is not a difference in "light to the sensor": it is in "ADU levels per photo-electron".
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311
Re: maximizing raw levels is nothing to do with maximizing sensor exposure
« Reply #82 on: January 10, 2013, 05:32:25 pm »

I only care about differences that are detectable in the final displayed prints, possibly measured by SNR measurements when the sensors receives equal exposure level.

we are talking about your camera (not 2 different cameras) typically receiving less light (less lux-seconds) than possible if you try to keep in camera JPG and/or preview normal (camera metering is tuned for in camera JPGs, so you can't just dial expocorrection to get more light for better raw data w/o risk to get unusable JPG and/or preview) in a situation when "nominal" ISO is significantly higher (1 stop) than "measured" (DxO) ISO... not about just changing gain while keeping aperture and exposure time constant...

I do not care about a mere moving to the left or right of the numerical levels in the raw histogram since my end goal in to look at pictures, not histograms.

and I am not talking about  that - it is you trying to turn the table to the gain game, instead of considering inconvenience of "nominal" ISO being significantly higher (1 stop) than "measured" (DxO) ISO for a raw shooter who wants for whatever reason raw metering tailored to the best raw data, not extra safe guaring of highlights for JPG shooters (albeit they are THE market)
« Last Edit: January 10, 2013, 05:35:38 pm by Vladimirovich »
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311
Re: maximizing raw levels is nothing to do with maximizing sensor exposure
« Reply #83 on: January 10, 2013, 05:39:17 pm »

and camera makers are free to make their own choices.

they are, nobody is stating that it is against any standards, just that "measured" (DxO) ISO (and "nominal" = "measured") is better for a raw shooter.

So this tag probably helps raw converters to produce a good default conversion.

except you think that receiving a stop less exposure (not gain mind you - but actual light) is somehow does not make any difference...
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311
Re: maximizing raw levels is nothing to do with maximizing sensor exposure
« Reply #84 on: January 10, 2013, 05:45:25 pm »

By the way, almost any raw conversion involves an "exposure correction", since raw files almost always place midtones half a stop or more below 18%, so that a default raw conversion (flat tone curve of contrast adjustment) must be scaled up by a half stop or more.

it depends on how do you meter to expose and which your raw converter you use (and how)... for example (in Adobe's realm, default settings, what Adobe DNG converter will write to DNG during raw -> DNG conversion, that is what ACR/LR are using w/ raws /unless overwritten in a custom camera profile/ - conversion just the way to check) baseline exposure for GH2 = -0.5, baseline exposure for GH3 = 0, baseline exposure for EM5 = +0.5...  
« Last Edit: January 10, 2013, 05:48:15 pm by Vladimirovich »
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311
Re: maximizing raw levels is nothing to do with maximizing sensor exposure
« Reply #85 on: January 10, 2013, 06:00:36 pm »

My recurring question is when and why different levels of scaling up are better or worse. And you seem to agree that it does not matter much....that has been my main point all along!
indeed that I agree - if you receive the same amount of light (aperture and exposure time) and you do not incur any clipping in raw or any ill effects (alleged non linearities, blooming, whatever) near well saturation point then changing just the gain may or may not be beneficial depending on your sensor and if there are benefits they might not be serious (depending on sensor)... for Canon-like sensors there might be some gains in shadows (while gain is applied to a signal, not to digital data), for Sony-like sensors there might be no practical sense, for Panasonic-like sensors (they are closer to Sony than to Canon) - you decide... and for cameras where gain is neither analog nor digital but merely a tag value in raw files it is even not a question... but I was trying to turn the tables  8) myself towards other aspects of the situation...
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10385
Re: DxO marks
« Reply #86 on: January 10, 2013, 08:51:32 pm »

BJL,
I don't believe I've ever read so much complication over what is for me an essentially simple matter. I imagine that novice photographers reading your posts, who are thinking about getting more serious about photography, would not know whether they are coming or going.

I could be wrong, but my impression is that it has been established several years ago that anyone who wants to capture the most detail, and the widest dynamic range with the lowest noise that their camera is capable of, is advised to shoot in RAW mode.

Furthermore, when shooting in RAW mode, one need not be concerned about 'placement of midtones' which you mention so frequently.The concern is, at least my concern is, that I give the maximum exposure to the scene whilst still retaining full detail in any highlight areas that I consider are important to the composition.

The process is known as Expose To The Right (ETTR) which I know you are quite familiar with, so I think it would help in such duscussions if you were to state from the outset that you are proposing an alternative method to ETTR.

My impression is, that you are proposing a method which involves as little post-processing as possible of scenes that generally do not have a high contrast ratio. In fact, I recall in a previous post, you made the comment that you believe most photographs taken by the public at large are not of high-contrast scenes and do not require cameras with a high DR capability, which is no doubt true.

The problem here seems to be that we are both involved in quite different approaches to the art (or craft) of photography. I'll tell you what mine is, then you can tell me what yours is.

For me, when I take a photograph, I generally consider it as a form of extremely detailed note-taking of a scene which I find interesting for any reason. Whether or not I can reproduce that initial interest, or emotional impact, or even enhance it, during the post-processing of the image or print, is the challenge.

Sometimes, first attempts fall flat, and I put the image aside and perhaps return to it months or even years later, with the benefit of greater experience, greater skill (hopefully) and improved processing programs.

The essential point I'm making is that my style of photography is more of a peripatetic, opportunistic and unplanned style, as opposed to the photographer who largely creates the scene he's about to photograph, using various props, and/or controls the lighting, has a client, a time frame and a schedule he has to meet for business purposes.

I can understand that certain professional photographers may find that the jpegs out-of-the-camera are sufficient for their purposes when time constraints are important and getting the image to the client as quickly as possible is a priority.

This is not my situation. Maybe it's yours. So please tell us, BJL, what your situation is, regarding your photographic style.
Logged

Charles Johnson

  • Newbie
  • *
  • Offline Offline
  • Posts: 8
Re: DxO marks
« Reply #87 on: January 11, 2013, 11:00:08 am »

As far as I can tell, the concept of "camera equivalence" by Falk Lumo is identical to my "equivalent image" that was published on luminous-landscape in 2007.  My essay had the title, "Why Is My 50mm Lens Equivalent to 80mm on a 35mm Camera and Why Is There More Depth-of-Field?"  It appeared in my book, SCP, as Chapter 11 with the title, "What is an Equivalent Image?"  I explained what parameters have to be scaled when the sensor size changed in order to get an image that appears to be the same in every way. However, I did not consider the pixel count or noise floor.  SCP = "Science for the Curious Photographer." 
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
DxO marks in the low light case (sensor saturation not possible)
« Reply #88 on: January 11, 2013, 03:53:49 pm »

Ray and Vladimirovich,
    I suspect that we are talking at cross purposes, due to thinking about different situations. My comments are addressed at the low light case where it is not possible to get enough exposure to saturate the sensor: no photosite is close to full.

Then the exposure strategy is to still to expose as far to the right as possible (maximum exposure level, minimum exposure index), but then there is no fixed relationship between how much signal the photosites get and what the raw file numerical levels are: the relationship between the two depends on factors like the analog gain applied, which varies both with the camera's ISO sensitivity setting and with the manufacturer's decisions.

Also I agree with Ray that it is simpler to think only about raw files, not in-camera JPEGs (when worrying about optimum exposure, I bracket, so one exposure might give the best raw while another exposure level gives the best in-camera JPEG.)

As I said before, the DxO SSat is a useful measure at minimum ISO setting, for ETTR purposes.
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
different sensors can react very differently to analog gain
« Reply #89 on: January 12, 2013, 11:51:20 am »

My thanks to Jack Hogan in this post for pointing to Bill Claff's excellent Photographic Dynamic Rage comparison chart tool, which provides a nice way to see how different sensors react differently to extra analog gain in low light situations, when no photosite is getting close to full-well exposure. In particular, consider some quite different sensors, the Canon 5D Mk 3, Olympus E-M5, and Nikon D800e.

As far as I can tell from these graphs and other sources:
  • The 5D3 sensor applies gain on-chip during transfer from photosite to the sensor's edge, and then transports the signal along the sensor edge and to an off-board ADC. As a result, at base ISO speed the noise introduced "downstream" (after amplifier gain) is three or four stops greater than the dark noise . Thus extra amplification by as much as three or four stops can reduce the effects of this downstream noise, but of course the amount must be adjusted to avoid clipping highlights, so it must vary with exposure level. (It might be ideal to vary gain according to the maximum exposure level of highlight photosites.)
  • The E-M5 and D800e do column-parallel ADC right next to the amplifier, so there is no significant transport noise and the only downstream noise is from the ADC ("quantization noise"?), and once the signal is amplified about one stop above base level, the amplifier output has noise floor above the level of this downstream noise, and further amplification does not improve signal quality: the signal and noise at each pixel get equally amplified, so the SNR at each pixel is unchanged. The only IQ effect of extra amplification is possible clipping of highlights. The amplification is therefore done mainly for convenience in raw conversion, with the side benefit of avoiding a raw output histogram that is hard to read due to being pushed far to the left.
(Note: with low sensor exposure levels, meaning high exposure index, the histogram of actual photosite signal levels as a fraction of well capacity is far to the left, but the closest we get to seeing that is by setting the camera at its minimum ISO speed setting.)
« Last Edit: January 12, 2013, 03:48:13 pm by BJL »
Logged

Peter van den Hamer

  • Newbie
  • *
  • Offline Offline
  • Posts: 43
Re: DxO marks
« Reply #90 on: January 22, 2013, 02:29:59 am »

As far as I can tell, the concept of "camera equivalence" by Falk Lumo is identical to my "equivalent image" that was published on luminous-landscape in 2007.

I agree. I actually ran into both articles over time. Due to their similarity, I actually got confused at some point whether they were written by the same author. So when I decided I needed a reference for the topic, I simply found Falk's version first.

Strictly speaking I could have referenced both, or Falk could have referenced you as related prior work (assuming you were first and he knew about your article). I can still add a note to the Luminous Landscape version of the article. Let me know if you want this. There is also a version of the article at DxO's site, but this is harder for me to change (and gets less attention).
Logged

fike

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1413
  • Hiker Photographer
    • trailpixie.net
Re: DxO marks
« Reply #91 on: January 22, 2013, 11:08:04 am »

Shifting gears here:

How can it be that an APS-C from Nikon (the D5200) outscores the full-frame 5DIII--84 to 81?  Is this score a true reflection of IQ reality?  Does this undersell the Canon camera, or is Canon THAT FAR behind?

I am really beginning to pine for sites that just show you the reference images and let you draw your own conclusions.  This putting a number to it seems to inflate pretty minor differences and obscure the things that might matter to you...those things you see with your own eyes.
Logged
Fike, Trailpixie, or Marc Shaffer

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: DxO marks
« Reply #92 on: January 22, 2013, 02:24:22 pm »

Shifting gears here:

How can it be that an APS-C from Nikon (the D5200) outscores the full-frame 5DIII--84 to 81?  Is this score a true reflection of IQ reality?  Does this undersell the Canon camera, or is Canon THAT FAR behind?

I am really beginning to pine for sites that just show you the reference images and let you draw your own conclusions.  This putting a number to it seems to inflate pretty minor differences and obscure the things that might matter to you...those things you see with your own eyes.

I would take the overall score with a grain of salt. Afterall, the Nikon D800e outscores the Phase One IQ180 back. The weighting of the various tests might not be how you would weight them. In the case of the D5200 and Canon 5DIII, the dynamic range at low ISO is what gives the Nikon a higher score. The SNR at 18% and other measures are better for the Canon. The Canon sensors are limited by old technology with a high read noise at base ISO. However, the larger sensor gives it an advantage when shot noise comes into play. The Nikon and Canon are evenly matched in terms of resolution, but the APS sensor places more demands on the lens which is not taken into account in the DXO testing.

Posting reference images would be a good idea, but then the lens and demosaicing software or hardware also comes into play.

Regards,

Bill
Logged

Vladimirovich

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1311
Re: DxO marks
« Reply #93 on: January 22, 2013, 07:21:57 pm »

but the APS sensor places more demands on the lens which is not taken into account in the DXO testing.
but then you shall use lenses designed for smaller circle, smaller sized optical elements (diameter of elements) shall have lesser issues when produced even vs FF lenses with only center of the frame used on those,  isn't it so ?
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: DxO marks
« Reply #94 on: January 23, 2013, 12:51:03 am »

Hi,

That is much about DR at minimum ISO playing a major role in the DxO weighting.

On the other hand, I have a very high ranking camera the Alpha 99 and also a camera that ranks close to the 5DII the Alpha 900. The Alpha 99 has about two stops more DR, but it took me several months to find an image that would demonstrate it, that was duping a high contrast Velvia slide in a dark room eliminating all stray light. Im real world situation I'd say the DR advantage may be masked by lens flare.

I don't have Canons so I cannot do a direct comparison.

Best regards
Erik


Shifting gears here:

How can it be that an APS-C from Nikon (the D5200) outscores the full-frame 5DIII--84 to 81?  Is this score a true reflection of IQ reality?  Does this undersell the Canon camera, or is Canon THAT FAR behind?

I am really beginning to pine for sites that just show you the reference images and let you draw your own conclusions.  This putting a number to it seems to inflate pretty minor differences and obscure the things that might matter to you...those things you see with your own eyes.
Logged
Erik Kaffehr
 

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Re: DxO marks
« Reply #95 on: January 23, 2013, 08:48:05 am »

but then you shall use lenses designed for smaller circle, smaller sized optical elements (diameter of elements) shall have lesser issues when produced even vs FF lenses with only center of the frame used on those,  isn't it so ?

Yes and no. I have personal experience only with Nikon. Their crop frame lenses are termed DX and these are represented by midrange zooms and a few primes, all consumer grade. Professional grade lenses are found in the FX line and these are often used on DX cameras when no equivalent DX lens is available. Examples are the 70-200 f/2.8 VR1 and VR2 and the entire Zeiss line.

Your comment about using only the center of the image applies to the 70-200 f/2.8 VR1. That lens had a very good reputation when Nikon had only crop frame sensors, but with the advent of full frame sensors it became apparent that edge resolution was wanting.

Regards,

Bill
Logged

risedal

  • Newbie
  • *
  • Offline Offline
  • Posts: 29
Re: DxO marks
« Reply #96 on: January 29, 2013, 05:36:09 pm »

Shifting gears here:

How can it be that an APS-C from Nikon (the D5200) outscores the full-frame 5DIII--84 to 81?  Is this score a true reflection of IQ reality?  Does this undersell the Canon camera, or is Canon THAT FAR behind?

I am really beginning to pine for sites that just show you the reference images and let you draw your own conclusions.  This putting a number to it seems to inflate pretty minor differences and obscure the things that might matter to you...those things you see with your own eyes.

Hello. this will be my first answer here  (you may have read answers from me at dpreview, Canon Rumors, etc.)

The APS sensor from Toshiba/Nikon and Sony/Nikon with raw-vise   ADC on the sensor edge  has both a high QE  and very low read  out noise regarding electrons which  Canon can not compete with  today because of theirs older  sensor technology= long analog signal path way and theirs late amplifications stages.
This means that Canons DR will be inferior at base iso compare to Nikons sensors. Regarding picture quality and especially above 550 iso the Canon 24x36mm sensor will  be better due the  larger sensor size and the demands of the lenses are smaller with a 24x36mm sensor compared to the pixel density in a APS size sensor and area (contrast, resolution) What we can hope for is that Canon can reduce the  high read out noise, banding  and increase the QE

(my choice of sentence structure and word choice can be a little Swe / english)

 
« Last Edit: January 29, 2013, 06:31:04 pm by risedal »
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: DxO marks
« Reply #97 on: January 29, 2013, 06:41:02 pm »

Hi,

Canon cannot achieve the high DR at low ISO the latest generation of Nikon, Pentax and Sony can. DxO mark is only about DR and noise and handling of color, it doesn't take resolution into account.

My personal opinion is that DR may be overvalued, but it is certainly nice to have.

At high ISO the Canons catch up.

Best regards
Erik

Hello. this will be my first answer here  (you may have read answers from me at dpreview, Canon Rumors, etc.)

The APS sensor from Toshiba/Nikon and Sony/Nikon with raw-vise   ADC on the sensor edge  has both a high QE  and very low read  out noise regarding electrons which  Canon can not compete with  today because of theirs older  sensor technology= long analog signal path way and theirs late amplifications stages.
This means that Canons DR will be inferior at base iso compare to Nikons sensors. Regarding picture quality and especially above 550 iso the Canon 24x36mm sensor will  be better due the  larger sensor size and the demands of the lenses are smaller with a 24x36mm sensor compared to the pixel density in a APS size sensor and area (contrast, resolution) What we can hope for is that Canon can reduce the  high read out noise, banding  and increase the QE

(my choice of sentence structure and word choice can be a little Swe / english)

 
Logged
Erik Kaffehr
 

risedal

  • Newbie
  • *
  • Offline Offline
  • Posts: 29
Re: DxO marks
« Reply #98 on: January 29, 2013, 08:54:30 pm »

My personal opinion is that a large DR=14 stop , mainly because of low read out noise gives us a freedom to under expose a contrasty motive with several stops, bring in high lights who are fare above 4 stops  of middle grey and then adjust the rest of motive which means that you got a larger visual DR with the  Nikons sensors compared to for example  Canon who has 12 times higher read noise  and  also banding in theirs 11.5 stops of DR ,how much banding  is depending on the model.6D has today the best sensor data of all Canon including 1dx regarding QE and less banding than 5dmk2 mk3
« Last Edit: January 29, 2013, 09:18:22 pm by risedal »
Logged
Pages: 1 ... 3 4 [5]   Go Up