Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: ETTR and LR 4 (and DxO)  (Read 6514 times)
Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« on: April 22, 2012, 08:44:03 AM »
ReplyReply

In the past I felt there was a reasonable consistency between what my camera showed as overexposed by blinkies and what I saw in Lightroom. Now with PV 2012 i LR4 this consistency has totally gone! I see in a number of pictures that a full stop higher exposure will show sometimes less blinkies than the one stop lpwer exposure on the LCD screen on the camera. My camera is a Canon 1Ds mkIII and I assume that a blinky is generated if at least one channel is blown out. I also assume that an area shown as overexposed in Lightroom (by pushing the J-key) also is one where at least one channel is blown out. In other words all other areas in the photo not shown as overexposed in LR do not have any channel blown out. So now the question: How can an area that is shown is partially blown out in LR3 or PV2010 be shown as not blown out at all in PV2012?

So if what PV2012 shows as not blown out at all is correct (which I assume) then there is a really new situation with ETTR! I have started bracketing all my landscape shots and now choose the one that is ETTR in Lightroom. Quite often this gives me one stop extra.

It could seem that since I understand that the DxO measurements are based on the DxO raw converter, that the dynamic range of a given camera is not necessarily what the DxO number says.... (assuming that the DxO rawconverter most likely is more like PV2010 than PV2012 in this respect).
Logged

luxborealis
Sr. Member
****
Offline Offline

Posts: 1169



WWW
« Reply #1 on: April 22, 2012, 09:22:56 AM »
ReplyReply

Interesting... you seem to have identified 3 different definitions of "clipped highlights": (1) the camera's definition (as shown by blinking highlights in playback - which may be the least accurate if the playback image has been processed in-camera as they often are), (2) the PV2010 definition and (3) the PC2012 definition (as shown by highlight clipping in LR). Of course, DxO will have their own definition and, no doubt, Capture One and Aperture may have different definitions again.

Rather than having to assume, it would be helpful to know what the exact definition is in each case (98%+, 99.9%+ or over 100% in one, two or three channels, or...?) Of course, the threshold value will vary with each camera raw version. This just might bring renewed requests for the option of setting our own clipping threshold value within LR which, over the years, photographers have been asking for.

Great question, Hans!
Logged

Terry McDonald
Revealing the art inherent in nature
- visit luxBorealis.com.
Have a read of my PhotoBlog and subscribe!
Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« Reply #2 on: April 22, 2012, 09:35:02 AM »
ReplyReply

Well, actually part of it is really a question for Schewe Smiley
Logged

BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3771


« Reply #3 on: April 22, 2012, 10:23:48 AM »
ReplyReply

In the past I felt there was a reasonable consistency between what my camera showed as overexposed by blinkies and what I saw in Lightroom. Now with PV 2012 i LR4 this consistency has totally gone! I see in a number of pictures that a full stop higher exposure will show sometimes less blinkies than the one stop lpwer exposure on the LCD screen on the camera. My camera is a Canon 1Ds mkIII and I assume that a blinky is generated if at least one channel is blown out. I also assume that an area shown as overexposed in Lightroom (by pushing the J-key) also is one where at least one channel is blown out. In other words all other areas in the photo not shown as overexposed in LR do not have any channel blown out. So now the question: How can an area that is shown is partially blown out in LR3 or PV2010 be shown as not blown out at all in PV2012?

Hi Hans,

All three clipping indications are based on the resulting image after rendering the Raw data. The camera bases it's blinkies on the JPEG data (so after applying your camera presets for colorspace, contrast, sharpening, WB, etc.). Lightroom apparently does so as well, because you can make the clipping warnings go away with adjustments. The main difference is between PV2010 and PV2012. PV2010 was probably quite close to the truth when using 'as-shot' settings and the other controls at default for the camera. PV2012 on the other hand will treat partially clipped data (which you can only determine with a tool like 'Rawdigger' which shows the actual Raw data) as somewhat repairable. It's unfortunately not possible to set that threshold ourselves, so we'll have to follow what Adobe determines good enough.

From their point of view it's somewhat understandable to only warn for clipping if it cannot be resolved easily, and PV2012 can do a convincing job recovering some of the really lost info, but I prefer more control and understanding of the real data that's available to work from. That's why I use bracketing in critical situations, use ETTR, and inspect the Raw data with Rawdigger. That also allows me to better interpret the camera blinkies, and make better exposures.

Quote
So if what PV2012 shows as not blown out at all is correct (which I assume) then there is a really new situation with ETTR! I have started bracketing all my landscape shots and now choose the one that is ETTR in Lightroom. Quite often this gives me one stop extra.

PV2012 can only make assumptions when it recovers lost/clipped data. That may lead to duller and less accurate highlight rendering than from a proper exposure. I would suggest to first learn to interpret the PV2012 indicators, based on Rawdigger's real assessment, rather than blindly following the PV2012 indications. You may get better conversions based on real data, even if that is a bit before PV2012 starts giving warnings.

We're all going through a learning curve here with the new PV2012.

Quote
It could seem that since I understand that the DxO measurements are based on the DxO raw converter, that the dynamic range of a given camera is not necessarily what the DxO number says.... (assuming that the DxO rawconverter most likely is more like PV2010 than PV2012 in this respect).

No, I think DxO generally gives reliable info, consistent with the underlying Raw data. I've had a similar issue with Capture One, where it's 'filmcurve' tonality suggested highlight clipping where the real Raw data still has headroom, but when switched to a linear tone curve the clipping indication became reliable, and consistent with the real data. Fortunately C1 does allow to set the threshold for the clipping indications.

Cheers,
Bart
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #4 on: April 22, 2012, 01:45:18 PM »
ReplyReply

In the past I felt there was a reasonable consistency between what my camera showed as overexposed by blinkies and what I saw in Lightroom. Now with PV 2012 i LR4 this consistency has totally gone! I see in a number of pictures that a full stop higher exposure will show sometimes less blinkies than the one stop lpwer exposure on the LCD screen on the camera. My camera is a Canon 1Ds mkIII and I assume that a blinky is generated if at least one channel is blown out. I also assume that an area shown as overexposed in Lightroom (by pushing the J-key) also is one where at least one channel is blown out. In other words all other areas in the photo not shown as overexposed in LR do not have any channel blown out. So now the question: How can an area that is shown is partially blown out in LR3 or PV2010 be shown as not blown out at all in PV2012?

I agree with all of what Bart posted, but would like to add a few points. According to Eric Chan, PV2012 automatically performs highlight recovery and Jeff Schewe appears to be ambiguous on this topic. As usual, tests with your own camera are in order. As an example, here are a few tests I did with the Nikon D3 by exposing a Stouffer wedge and comparing the actual raw histogram as shown by Rawnalize and Rawdigger with the camera histogram and blinking highlights as well as the ACR histogram and highlight indicator. I used the Nikon standard picture control (normal contrast).

This image demonstrates that the Nikon histogram is slightly conservative, but close enough for practical work.


The blinking highlights are a bit more conservative.


The ACR histograms in PV2010 are complicated by the +0.5 EV exposure offset used by ACR for this camera. To compensate for this offset, it is necessary to decrease the ACR exposure by 0.5 EV.

Here is PV2010 with no exposure compensation for image 05 which is slightly below clipping as shown by the Rawnalize histogram. It appears blown in ACR.


Using the -0.5 EV compensation, the image still appears slightly overexposed.


PV2012 shows a more accurate histogram for the properly exposed image. The baseline offset seems not to be in effect.


Image 03 has clipped green channels. The blue channel is at clipping and the red channel is just short of clipping (this is with a 5000K light box).


The PV2012 histogram exhibits minimal clipping, indicating the effect of highlight recovery.



I don't know what baseline offset PV2010 uses for your camera (this can be determined by converting to DNG and using an exif reader), but for my own camera, PV2012 shows improvement.


Regards,

Bill
Logged
Schewe
Sr. Member
****
Online Online

Posts: 5500


WWW
« Reply #5 on: April 22, 2012, 02:10:02 PM »
ReplyReply

According to Eric Chan, PV2012 automatically performs highlight recovery and Jeff Schewe appears to be ambiguous on this topic.

I guess it all depends on what you consider to be "highlight recovery"...the new image adaptive PV 2012 has an auto-ranging capability which can appear as "recovery" but I don't think it really is in a classic sense. What I think is going on is that the way the highlights are rendered in PV 2012 has a different roll off which extracts more highlight detail. Note that in a linear capture, the brightest area of the capture has a lot of data even if there is no clipping. Recurving the tone mapping to extract more usable image data is certainly what PV 2012 is doing. That's what I think people are seeing...and I don't call that "highlight recovery" which to me means extracting image detail when one or more channels are clipped. I'm pretty sure there is a small degree of classic highlight recovery going on even when all Basic panel controls are at zero which I think is what Eric is talking about and I think it's the new tone mapping that is making the biggest differences. But a minus Exposure or Highlights is clearly when the classic "highlight recovery" really kicks in and that has been substantially improved in PV 2012.

I'm still coming to grips with PV 2012 and I've been working with it in various forms since late last summer...what I think is going on is that this new tone mapping logic is really, really leading edge and traditional descriptions and terms fail to really describe what going on under the hood. I've read the post on Lightroom Journal titled Magic or Local Laplacian Filters? and that stuff is too deep for me to understand...but what I will say it that PV 2012 has made a major improvement in the ability to tone map raw captures. I can't say it's "perfect" (there are a couple of types of images where tone bleeding can occur) but compared to PV 2010 it's clear that one can get a lot more image detail out of highlights and shadows. I hesitate to call this a change in the dynamic range of the image but is sure does make it easier to deal with higher contrast scenes...I'm going to be in the process of writing about this stuff in the very near future...the way I usually do that is ask some questions of Eric and Thomas, swirl it around and spit it out and see if they agree with me. I'll let ya all know how that goes :~)
Logged
Schewe
Sr. Member
****
Online Online

Posts: 5500


WWW
« Reply #6 on: April 22, 2012, 03:37:31 PM »
ReplyReply

In the past I felt there was a reasonable consistency between what my camera showed as overexposed by blinkies and what I saw in Lightroom.

If by that you mean that most all cameras' display an sRGB (or Adobe RGB) ripped preview and then apply a very, very conservative indicator for clipping (for Canon about 1 stop) then that would be generally correct. The camera LCD and histogram are 8 bit renderings based on the camera company's raw to JPEG SDK rendering. For the 1DS MII (which I also shoot) it's an easy 1 stop conservative clipping warning. If you are doing ETTR of a scene whose contrast range is about stop or 2 lower than the sensor's dynamic range it can be very tricky to nail the optimal exposure. I tend to bracket +- 2/3 stop around an exposure compensated meter reading...usually the plus 2/3 is fine.

And yes, PV 2012 CAN make far better use of highlight detail in a raw capture. The new tone mapping extracts a lot more highlight textural detail and can go into the shadows a lot better without artifacts. That will have an impact on how you think about doing ETTR and HDR...and with today's sensors the need for ETTR has, I think been reduced but not eliminated. It's gotten to the point where rather than worry over much about ETTR I'm more concerned about nailing a "correct" exposure. For high contrast scenes that's deciding whether or not highlight texture is visually important, if it is, I'll cut exposure to preserve highlights or do a bracket so I can blend 2 or more layers to retain texture. For a lower contrast scene I'll still push exposure to the right as long as I can maintain the require F-stop & shutter speed to get the technical capture optimal. Sometimes even pushing the ISO a bit (depending on the camera's noise signature).
Logged
deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 966



« Reply #7 on: April 22, 2012, 09:15:11 PM »
ReplyReply

It could seem that since I understand that the DxO measurements are based on the DxO raw converter

absolutely not - they sell a separate software = http://www.dxo.com/intl/image_quality/dxo_analyzer
Logged
deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 966



« Reply #8 on: April 22, 2012, 09:20:29 PM »
ReplyReply

Interesting... you seem to have identified 3 different definitions of "clipped highlights": (1) the camera's definition (as shown by blinking highlights in playback - which may be the least accurate if the playback image has been processed in-camera as they often are), (2) the PV2010 definition and (3) the PC2012 definition (as shown by highlight clipping in LR). Of course, DxO will have their own definition and, no doubt, Capture One and Aperture may have different definitions again.

some raw converters will show the real raw histogram, for example rawtherapee, rpp - which is the real situation with the data and you can see what is really clipped (which channel(s)), unlike ACR/LR where you have only approximation (in some cases good enough for you though).

Logged
elied
Sr. Member
****
Offline Offline

Posts: 269


« Reply #9 on: April 23, 2012, 06:27:29 AM »
ReplyReply

The LR histogram is far from an approximation since it represents image data after the application of WB, which changes the data set fundamentally and can in itself cause clipping that was not present prior to the application. Daylight WB, for instance, can multiply red channel values by more than x2. Tungsten WB will increase the blue channel by even more. WB is the primary reason why both camera and LR histograms are remote from Raw reality.
Logged

Roll over Ed Weston,
Tell Ansel Adams the news
bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #10 on: April 23, 2012, 07:23:47 AM »
ReplyReply

The LR histogram is far from an approximation since it represents image data after the application of WB, which changes the data set fundamentally and can in itself cause clipping that was not present prior to the application. Daylight WB, for instance, can multiply red channel values by more than x2. Tungsten WB will increase the blue channel by even more. WB is the primary reason why both camera and LR histograms are remote from Raw reality.

One main difference between the monochrome camera histogram and the ACR/LR histogram is that the former is a luminosity histogram, whereas the ACR/LR histogram is a composite RGB histogram (see Cambridge in Color). In Photoshop, one can select the type of histogram. The luminosity histogram is largely determined by the green component. The more advanced cameras have color histograms showing the RGB channels separately after white balance. To avoid the white balance multipliers, one can load a UniWB profile into the camera, but saturation clipping can still occur, since the widest color space available with most cameras is Adobe RGB. It is time that the camera makers should offer a true raw histogram with a log base 2 scale (f/stops) on the x-axis. A log scale is often better for the y-axis as well.

Regards,

Bill
Logged
Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« Reply #11 on: April 23, 2012, 07:29:35 AM »
ReplyReply

Thanks for all the answers and comments. I have used Rawdigger to analyze some of my pictures and compared to both LR PV2010 and PV2012 clipping indications using a completely neutral setting in the develop module. One thing is the fairly big difference between PV2010 and PV2012 which bothers me as to how to choose a bracketed shot that is optimal and without any channel clipped. I knew, of course, that the histogram changes and also the clipping indicator when I change e.g. the exposure slider in LR (as well as many other sliders of course), but I somehow had the belief that in a neutral setting the clipping indicator would be a good indicator for any clipped RAW data and wrong I was!

So not only can we not expose to the right when using our cameras as Michael has written about here http://www.luminous-landscape.com/tutorials/optimizing_exposure.shtml but we are also stuck in post processing unless we check the RAW histogram using a tool like Rawdigger for each picture since we cannot rely on the clipping indicators in any way in LR. This may sound a bit overstated, but from my analysis of some test shots done today it is clear that there is no safe way to know that a channel is clipped just by looking at LR histogram and clipping indicators (which I understand is purely a function of the histogram).

So what do we do? Well if the picture looks good is there a problem? Maybe we don't see the problem on our screens when doing the pp. Maybe we see it when a large print is made and we can see in the sky that areas are not rendered right due to clipping. One thought come to mind that I would be very desirable to have an option in LR to show clipping based on RAW data and not just the rendered data.
Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #12 on: April 23, 2012, 07:36:12 AM »
ReplyReply

Thanks for all the answers and comments. I have used Rawdigger to analyze some of my pictures and compared to both LR PV2010 and PV2012 clipping indications using a completely neutral setting in the develop module. One thing is the fairly big difference between PV2010 and PV2012 which bothers me as to how to choose a bracketed shot that is optimal and without any channel clipped. I knew, of course, that the histogram changes and also the clipping indicator when I change e.g. the exposure slider in LR (as well as many other sliders of course), but I somehow had the belief that in a neutral setting the clipping indicator would be a good indicator for any clipped RAW data and wrong I was!

Hans,

My experience with PV2012 is similar to yours. With respect to PV2010, I have found that a linear tone curve (sliders on main tab zeroed and point curve to linear) gives an accurate indication of clipping if one compensates for the BaselilneExposure offset that ACR/LR uses (+0.5 EV for the Nikon D3, so one must use -0.5 EV exposure in ACR).

Regards,

Bill
Logged
Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« Reply #13 on: April 23, 2012, 08:11:26 AM »
ReplyReply

As an example of the test shots I did and where Rawdigger shows clipping and what the clipping indicators were in LR using PV2010 and PV2012 with neutral settings and WB as AWB from the camera (close to daylight). The metering in the camera (Canon 1Ds mkIII) using matrix metering (aka. evaluative) would naturally underexpose this scene, so I bracketed with 1/3 stop between between EV 0 to EV +2. Only at EV+2 were there clipping in the RAW data (_MG_4747) and LR PV2012 showed no clipping and PV2010 a very large clipped area.  Screen shots from LR for _MG_4746 are not attached but LR PV2010 showed clipping and PV2012 not (of course).
Logged

Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« Reply #14 on: April 23, 2012, 08:22:16 AM »
ReplyReply

If by that you mean that most all cameras' display an sRGB (or Adobe RGB) ripped preview and then apply a very, very conservative indicator for clipping (for Canon about 1 stop) then that would be generally correct. The camera LCD and histogram are 8 bit renderings based on the camera company's raw to JPEG SDK rendering. For the 1DS MII (which I also shoot) it's an easy 1 stop conservative clipping warning. If you are doing ETTR of a scene whose contrast range is about stop or 2 lower than the sensor's dynamic range it can be very tricky to nail the optimal exposure. I tend to bracket +- 2/3 stop around an exposure compensated meter reading...usually the plus 2/3 is fine.

And yes, PV 2012 CAN make far better use of highlight detail in a raw capture. The new tone mapping extracts a lot more highlight textural detail and can go into the shadows a lot better without artifacts. That will have an impact on how you think about doing ETTR and HDR...and with today's sensors the need for ETTR has, I think been reduced but not eliminated. It's gotten to the point where rather than worry over much about ETTR I'm more concerned about nailing a "correct" exposure. For high contrast scenes that's deciding whether or not highlight texture is visually important, if it is, I'll cut exposure to preserve highlights or do a bracket so I can blend 2 or more layers to retain texture. For a lower contrast scene I'll still push exposure to the right as long as I can maintain the require F-stop & shutter speed to get the technical capture optimal. Sometimes even pushing the ISO a bit (depending on the camera's noise signature).

Thanks! I'm looking forward to your writeup. I can imagine that your Adobe friends (and you as well, maybe?) would be against adding a RAW histogram option to an already fairly rich interface. Maybe a button or short cut (of the many still available Smiley) to display the RAW histogram and the clipping for the RAW data. For Canon sensors and a high DR scene that one would prefer to pp in LR as a single image one really need to have the optimum exposure not to loose some of that precious DR that the Canon does not have too much of. For Sony Exmor sensors it's a lesser need, but still optimizing noise performance and avoid blending can be quite desirable. Especially with the great working of the highlight and shadows sliders in both the detail panel and in the grad filter and brush.

So a reliable (RAW) clipping indicator would be really good to have in LR.
Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #15 on: April 23, 2012, 09:04:02 AM »
ReplyReply

As an example of the test shots I did and where Rawdigger shows clipping and what the clipping indicators were in LR using PV2010 and PV2012 with neutral settings and WB as AWB from the camera (close to daylight). The metering in the camera (Canon 1Ds mkIII) using matrix metering (aka. evaluative) would naturally underexpose this scene, so I bracketed with 1/3 stop between between EV 0 to EV +2. Only at EV+2 were there clipping in the RAW data (_MG_4747) and LR PV2012 showed no clipping and PV2010 a very large clipped area.  Screen shots from LR for _MG_4746 are not attached but LR PV2010 showed clipping and PV2012 not (of course).

Hans,

Did you use the proper baseline offset for the 1DsMIII for the PV2010 rendering? It is +0.35 EV, so you would have to dial in -0.35EV exposure. This can be determined with the Rawdigger Exif utility with a DNG file.

Regards,

Bill
Logged
John R Smith
Sr. Member
****
Offline Offline

Posts: 1357


Still crazy, after all these years


« Reply #16 on: April 23, 2012, 09:07:08 AM »
ReplyReply

How do you find out what the LR baseline offset is for a Hasselblad 3FR file?

John
Logged

Hasselblad 500 C/M, SWC and CFV-39 DB
and a case full of (very old) lenses and other bits
Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« Reply #17 on: April 23, 2012, 09:15:35 AM »
ReplyReply

Hans,

Did you use the proper baseline offset for the 1DsMIII for the PV2010 rendering? It is +0.35 EV, so you would have to dial in -0.35EV exposure. This can be determined with the Rawdigger Exif utility with a DNG file.

No, I didn't. The large clipping area goes away on _MG_4747 with -0.65EV and on 4746 it goes away with -0.35EV. I'll check later what Rawdigger says as I'm going out shooting now Smiley

However I still would prefer an easy to use clipping indicator that says what RAW data is clipped. Sometimes one could choose to ignore it or crop the area away.
Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2825



« Reply #18 on: April 23, 2012, 10:01:31 AM »
ReplyReply

How do you find out what the LR baseline offset is for a Hasselblad 3FR file?

John,

You would convert the raw file to DNG (if it isn't already in DNG) and use an exif reader to check the baseline offset. Rawdigger can read the exif.

Regards,

Bill
Logged
Hans Kruse
Sr. Member
****
Offline Offline

Posts: 760



WWW
« Reply #19 on: April 23, 2012, 02:50:33 PM »
ReplyReply

Hans,

Did you use the proper baseline offset for the 1DsMIII for the PV2010 rendering? It is +0.35 EV, so you would have to dial in -0.35EV exposure. This can be determined with the Rawdigger Exif utility with a DNG file.

Regards,

Bill

I checked and it is correct (of course) that the baseline offset is 0.35EV for the 1Ds mkIII.
Logged

Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad