-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Iliah Borg
Forum Pro
Why do you say that? The issue is with missing levels in the shadows of the red channel, right in the raw.
Iliah Borg
Forum Pro
It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
--
About LibRaw | LibRaw
No comment on the delta E implication of this 'combing?It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
--
http://www.libraw.org/
I don't see how such a small (1 unit - 3 at worst) difference in a single channel can have such a huge delta E difference? Thoughts?
It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
--
http://www.libraw.org/
I don't have access to the original RAW file, but I bet I could make a good looking JPEG from the original file, assuming I didn't go from RAW to ProPhoto to Adobe or sRGB.
Iliah Borg
Forum Pro
Oh yes, with a little painting you can.I don't have access to the original RAW file, but I bet I could make a good looking JPEG from the original file, assuming I didn't go from RAW to ProPhoto to Adobe or sRGB.It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
--
About LibRaw | LibRaw
46 levels sounds bad, until you reaslise you would only expect 92 levels in this range with a 14-bit lossless camera. Ergo, you lose only 1 bit with the Sony, which I'm assuming is mostly noise anyway. It would be trivial to randomise the last bit to fill out the histogram.
So if the this is really the issue (I'm not convinced, you shouldn't need any red data to represent deep blue water), then it will affect any digital camera, even the D810.
Iliah Borg
Forum Pro
But deltaE is not defined for raw data. On a side note, deltaE is an approximation of what we know of human perception.No comment on the delta E implication of this 'combing?It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
--
About LibRaw | LibRaw
Iliah Borg
Forum Pro
Yes, that's too few, especially combined with white balance that stretches the range by 2.4 times (but 2x on D810, which helps here), making it 105-331, with essentially only each fifth level populated and the "compensation" gaps now stretched from 4 pixel values to nearly 10.
Comparing to other channels, with twice the amount of levels over the same region of the image, and taking demosaicking into account, we have what we have.
All the details are in the "unwanted" weak channel. Blue sky, blue water, details are in red channel. Open any image containing the sky, void the red channel, and see how it flattens the colour variation.
--
http://www.libraw.org/
Last edited:
I was curious, so I did a little digging.
The D810 also has red as a weaker channel, which makes sense given that it's a Sony-made sensor.
http://www.dxomark.com/Cameras/Nikon/D810---Measurements (click into color response)
Then I saw some people having RAW conversion issues with various camera profiles on the D810 when it first came out:
https://www.flickr.com/groups/1567431@N22/discuss/72157646074582954/
See top image here:
However, if this is an issue of combination of Sony RAW, dropped red channel, and color profiles, I am not seeing it in my own photos.
I have shot a lot of photos with very bright reds, including red gradients in skies, and not seen the severe posterization effects people are saying exist.
Here are a few:
Do note that these RAW files were processed with the Adobe Standard profile.
If this is a condition that I need to look out for, I would like to know, because I want to avoid the problems if possible. Is it the Camera Standard/Vivid profile that is causing the issue again?
--
http://www.lightfinity.net
Last edited:
Rather than just looking at one photo, from one guy...has anybody else seen it this bad on their photos or on other webpages? The A7rii is starting to get out there and there are tons of pictures posted. I personally haven't seen this issue, but 1 doesn't make a population. So what is everybody else seeing?
it isn't, no, but it is when transformed into something that looks like a photo. All we need to know is a ballpark. As for Delta E only being an approximation of what we know of human perception, well of course it is. Most things are - however the latest versions of delta E are fairly accurate and more than suitable for ballparks. i.e If delta E 2000 is <0.5 it's unlikely we can see it obviously. If it's over 3 the we can probably all see it. Many printers work to delta E of 3 so that's your 'ballpark' accurate color.But deltaE is not defined for raw data. On a side note, deltaE is an approximation of what we know of human perception.No comment on the delta E implication of this 'combing?It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
--
http://www.libraw.org/
My point is if we can get a value of delta E 2000 to an order of magnitude then it will be indicative of the errors we would expect. Would it likely be more accurate than guessing based on looking at a combed histogram? I don't know - worth a shot though..
Tim
Do you have an idea why they provide their raw files in this form and lossy compressed only? Looking at it from the outside it doesn't make much sense.
If that's the case then it sounds like the problem would still be there even if the RAW file was 14-bit (lossless).
I'd love to get the original RAW to see what effect adding a couple of bits of noise would have to the result. If done pre-demosaicing it could be a lot.
Iliah Borg
Forum Pro
I hear it differently, stretching a fully populated region by 2x is much less prone to problems.
I wonder is it a valuable solution and not a last resort for a 43-megapixel 14-bit camera with very low noise.
Iliah Borg
Forum Pro
But than we need to agree on what transform to use, and what demosaicking; another can of worms.it isn't, no, but it is when transformed into something that looks like a photoBut deltaE is not defined for raw data. On a side note, deltaE is an approximation of what we know of human perception.No comment on the delta E implication of this 'combing?It is posterization issue too all right. When you have as few as 46 levels in the 44-138 range in the red channel - and that is over 1.8 million red pixels - it is too few:
It is simpler to know the camera limitations and not to use it when it does not work well enough for the purpose.
--
About LibRaw | LibRaw
Assuming similar FWC to the A7R you're capturing 135 photons to record a level of 44 at ISO100. I haven't done the calculation but I'd be surprised if there wasn't at least 1 bit of shot noise. Hence you can replace the missing bit with noise and you're effectively back where a a lossless 14-bit sensor would be.
It's a solution. Having lossless RAW files is a better solution. But we don't have that, so I'll make do with what we do have.
billzilla2000
Veteran Member
Iliah Borg
Forum Pro
I'm afraid you have a wrong model in mind, I never saw this type of solution working well, preserving the resolution the sensor supposed to deliver.
Keyboard shortcuts
- f
- Forum