Want to Increase The Dynamic Range of Your EOS 5D Mark IV Raw Files?

[rrcphoto]

Hey, thanks for posting those real results. It looks like there's really something to it, if you use the right software, and you spend a gob of time on it.

it's really not that much work really. and even though it doubles the size of the RAW if you are shooting a wedding,etc or something with alot of white, it just may save you by giving you that additional exposure.

 

[privatebydesign]

I'm not seeing anything groundbreaking.

 

[aceflibble]

What I'm seeing from y'all testing is a lot of overestimating what 1 stop difference in dynamic range looks like.

And heads up, if you're comparing it to the recovery sliders in Lightroom, bear in mind that the LR sliders at 100% typically work out to be 2-2.5 stops. (Exactly how much depends on the file in question; neither slider can be measured in hard stops and they don't scale in a linear fashion). So if you can match that with this methodology, that means you're able to push this further than the initial 1-stop claim. (Which isn't surprising because usually combining two images which are 1 stop apart in exposure does usually work out to equal about 1.5 stops of added range, as the dynamic range of each exposure doesn't simply scale up/down linearly, either.)

tl;dr version: This appears to work, parallax aside. I'd agree with anybody who says they don't see the point because it's more effort than simply using recovery sliders for basically the same end result, but I don't agree with anybody who isn't impressed that this can work, or suggests it doesn't work. Getting 1 stop or more recoverable data isn't to be sniffed at.

 

[privatebydesign]

(

I also uploaded 2 pics of a church nearby using DPRSplit.
https://goo.gl/UWVrwh

Neither exposure is blowing highlights so the differences are not realizable.

Here is both frames adjusted in LR to optimal, they are essentially the same. The second screen shot is the before after of the image with the higher exposure, nothing is blown.

To be sure the only way you can realize any benefit from this Dual Pixel ability is if you ETTR and the 'normal' exposure blows highlights and the 'secondary' exposure doesn't, then you can realize the 1 stop more range in those non-blown highlights from the secondary frame.

 

[cgc]

I think that the current DPRAW implementation by Canon has not the properties required for a job requiring critical quality. Beyond than blending issues there are other artifacts which can be noticeable:

http://www.fredmiranda.com/forum/topic/1453555/0

For example, take a photo of a evenly illuminated white wall and look at the -1EV exposed frame: its lighting levels won't be uniform (a side will be brighter than the other):

http://www.fredmiranda.com/forum/topic/1447974/3&year=2016#13705105

 

[Pax2You]

I originally tried combining the posted files (thank you by the way) using an hdr merge in Affinity but I had ghosting with the automatic alignment and artifacts when I tried the "remove ghosting" option. So I took a different route. I did an image stack and chose "Mean". This created a more realistic exposure merge. I then manually aligned the image before exporting as a Tiff then tweaking in LR 5.7. You can definitely see the offset between the images. I didn't sharpen at all and didn't do more than basic processing. This method gave me no apparent haloing or artifacts. Interesting....

 

[mclaren777]

Quick test...

Thank you for testing this out.

 

[EzDingo]

NP... As the original article mentions [IF] any part of the image is actually clipped then there is about 1 stop of extra highlight detail to be had from the dual pixel alternate frame. Trying to process these 2 frames in a HDR method does not seem logical to me compared with direct masking back in some extra highlights if required.

 

[bludragon]

Can anyone explain the math behind this one? I mean, if the normal image is made up by summing the two values from each half of a dual pixel, which each collect half the light, how is it better than just halving brightness in the original image?

The only way I could see it helping is if the file format runs out of bits to store the max value from combining the dual pixels. That is the case for jpegs which drop some of the highlight detail, but I'd be surprised if it is the case for regular RAW files.

 

[SecureGSM]

https://www.rawdigger.com/howtouse/canon-dual-pixel-mode-highlights-are-there

Can anyone explain the math behind this one? I mean, if the normal image is made up by summing the two values from each half of a dual pixel, which each collect half the light, how is it better than just halving brightness in the original image? ...

 

[hne]

Can anyone explain the math behind this one? I mean, if the normal image is made up by summing the two values from each half of a dual pixel, which each collect half the light, how is it better than just halving brightness in the original image?

The only way I could see it helping is if the file format runs out of bits to store the max value from combining the dual pixels. That is the case for jpegs which drop some of the highlight detail, but I'd be surprised if it is the case for regular RAW files.

Each half-pixel is going through an analog-to-digital converter and ends up in a 14 bit memory slot as a number between 0 and 16383, for example 8375 being 10 0000 1011 0111 and 8402 being 10 0000 1101 0010. The two half-pixels are then added together into a new 14 bit memory slot, 8375+8402=16777 or 100 0001 1000 1001 which wouldn't fit in 14 bits. The value is clamped to largest number you can encode with 14 bits: 16383=11 1111 1111 1111. Blown higlight.

If you have dual pixel raw turned on, you'd have two pictures in your .CR2, one with blown hilights being clamped to 11 1111 1111 1111 and one with 10 0000 1101 0010. The number of extra stops of hilights is 0.034285: (ln(16777)-ln(16383))/ln(2)

The largest gain from this is when you almost blow out both half-pixels: 11 1111 1111 1110+11 1111 1111 1110=111 1111 1111 1100 (3276 in decimal) which gives .9999 stops extra hilight detail which means very close to double the brightness recorded. The math behind the number of stops: (ln(32764)-ln(16383))/ln(2) ≃ .999912

In short, as long as your hilights are at most blown by just a hair under a stop, this would make them recoverable, albeit with a minor parallax shift. Anything over 1 stop would still give an extra stop of hilights in the gradient towards the blown-out region.

Canon could implement this hilight recovery directly in camera, if they were to add a 15th bit to the .CR2 format.
But then, the .CR2 format also wastes a few bits on a non-zero black point (which is really good for shadow denoise) so we wouldn't get 15 bits of DR just like that.

 

[Sharlin]

So the (sub)photosites have some headroom in well capacity, being able to represent higher values before saturating than the ADC and post-ADC digital electronics. Basically the sensor has a ”native” ISO lower than 100.

 

[privatebydesign]

So the (sub)photosites have some headroom in well capacity, being able to represent higher values before saturating than the ADC and post-ADC digital electronics. Basically the sensor has a ”native” ISO lower than 100.

No, they have half the well capacity, it is a maths issue.

If a 14 bit ADC had a range from 0-10, 10 being clipped, then if each half is below 5 the highlight will have detail in the main frame. If both sub pixels register 8 then in the main frame the value would be >10 so blown, but the value from the sub frame is 8, so it retains highlight detail.

The point is the main frame has the FWC of two sub pixels added together to make the output value limited at 14 bits, this means if the value of the sub frames pixel is less than 1/2 the 14 bit ceiling you will get additional highlight detail out of the sub frame. As we all know 1/2 or x2 is equal to 1 stop, so in theory there is a possible 1 stop improvement in highlight detail using this technique.

 

[Iliah Borg]

Gentlemen:

We made some changes,
http://updates.fastrawviewer.com/data/DPRSplit/DPRSplit-0.8.2.10-x64-Setup.exe
http://updates.fastrawviewer.com/data/DPRSplit/DPRSplit-0.8.2.10.dmg

- font size on the first run (and margins for Windows) are set based on the screen size. If the base font size setting is changed to smaller, the application window is also changed to smaller, accordingly. Works OK with 1200 by 800 pixels screens. The control is in Preferences.

To ease the pain for those who want to use both frames to get less noisy shadows from the "main" (composite, A+B) frame:
• Adjust EXIF shutter speed by -1EV for second frame (for HDR merge) – simulates reduction of exposure by 1 stop by setting the shutter speed in the EXIF of the auxiliary frame 1 stop faster. This may be useful (and even necessary) while merging first and second frame in certain HDR stacking applications that analyze frames for exposure parameters.

For the first beta, we were careful not to confuse the raw converters that do not process the extreme highlights correctly. Now we offer a finer control in Preferences:
• Data Maximum: controls the content of the DNG White Level tag:
o Normal White –«white» will be set to the Normal White value found in makernotes of CR2 file.
o Specular White (default) – «white» will be set to the SPecular White value found in makernotes of CR2 file.
o Data Max (or specular) – «white» will be set to the actual maximum found in raw data of the CR2 file or to Specular White, if Specular White is higher.
o Data format max – 16383 (minus the black level if the processing mode is set to subtracting black). This option may result in color artifacts in the highlights (such as pink clouds) if the raw converter incorrectly clips the highlights after applying white balance.

 

[Mt Spokane Photography]

I tried the updated app. I used lightroom, helicon, and photomatix on the same image. I liked it for the test because of the very dark bird feeder and the almost blown out white post.

Once again, Lightroom = Fail, no need to show another repeat.

Helicon, excellent output, it will need a little tweaking to boost the shadows.

Photomatix, I liked the "Realistic" output because it is warmer and boosts the shadows.

Top = Helicon, Bottom = Photomatix..

 

[Iliah Borg]

Thank you for testing.

Here is the "main" frame, magenta highlighting indicates the green channels of raw data being blown out, red highlighting is where both green and blue channels are clipped, black spots are indicating clipping in all the channels. From "Statistics" you can see how many pixels / % of each of the colour channels are clipped.

Now, the "auxiliary" frame, clipping is very tolerable:

Settings:

 

[zim]

Sounds like a really interesting ‘development’ but FWIW here’s the file processed through Affinity Photo develop module. I’m a big fan of AP but not of the develop module, it has IMHO many issues I much prefer DXO Pro for RAW development but my version doesn’t support the 5D4.
Anyway my take on this image is that it’s no test for the 5D4 sensor which with the most basic of adjustments (highlight adjustment and a little tone curve) in AP can recover those highlights. I didn’t really pay much attention to shadows and I didn’t use Overlays which would have given me much more control of the image.

Don’t you need something a bit more extreme to show if there are any benifits to dual pixel RAW?

 

[Mt Spokane Photography]

Thank you for testing.

Here is the "main" frame, magenta highlighting indicates the green channels of raw data being blown out, red highlighting is where both green and blue channels are clipped, black spots are indicating clipping in all the channels. From "Statistics" you can see how many pixels / % of each of the colour channels are clipped.


Now, the "auxiliary" frame, clipping is very tolerable:


Settings:

So it looks like I was able to recover almost all of the 18% clipped pixels - Impressive! I focused on the black portion of the feeder, so the camera did boost the exposure because of the dark area. I expected that. It was set to evaluative metering, but still gave some priority to the center of the frame.

 

[Iliah Borg]

So it looks like I was able to recover almost all of the 18% clipped pixels - Impressive! I focused on the black portion of the feeder, so the camera did boost the exposure because of the dark area. I expected that. It was set to evaluative metering, but still gave some priority to the center of the frame.

I think yes, the pole looks natural, without dull grey spots resulting from reconstructing highlights through interpolation.

Some folks prefer Lumariver HDR ( http://www.lumariver.com ) because it results in DNG file with virtually no ghosting.

 

[Mt Spokane Photography]

Some folks prefer Lumariver HDR ( http://www.lumariver.com ) because it results in DNG file with virtually no ghosting.

Unfortunately, its Mac only. I was really unhappy in 1992 when my company standardized on pc's, but I've become so adept at using one that a Mac might seem strange 25+ years later.