worth of unique values. Anyone can repeat this exercise for themselves, and question my method.nearly 14 bits
. Parallel samples of Nikon D600 and D3 files (I found these on imaging resource), show data show less than 11 bits of actual unique sample
RX1 and A99 raw files both seem to Summary:
Samples I tested on my RX1 showed no more than 1,530 unique pixel values, though they covered a 12 bit range (values 0-4095), and presumably are results of a 14 bit conversion ,which implies 0-16383 range. (Though I could accept it, if they really meant to say a 14 bit conversion that produces 12 bit output.)
I was experiencing banding artifacts in a B&W conversion that I was doing (RX1 raw file). I will admit to all kinds of possible reasons for this, related to graphics card/display issues, exposure (I purposely overexposed to get as much info in the sky as possible), in camera vignetting correction, color space choices, noise reduction, overaggressive tone tweaking, etc. But still, even when backing off of edits, and exercising care, etc., the banding didn't seem justified to me, especially since I believed I would be getting 14 bit conversions.Why I started digging:
I went into Raw Digger, to examine actual histogram data. I looked for my best exposed file (in terms of number of unique values produced) and started comparing it against a variety of raw files from A99, D600, D3.What I did:
What I found in the RX1 file (A99 is similar, but the particular file was a bit worse):
What this means and doesn't mean:
Though the file sizes are nearly the same, and this type of output should compress pretty well, the Sony raw files are close to the same size as the Nikon files, so smaller files doesn't seem to be the motivation.
I don't have proof that the Nikon files contain genuinely unique data, rather than faking the data for people like me to be impressed.
I'm aware that visually, I could not possibly comprehend all these different values, but my point is I want maximum headroom in my files for things like highlight recovery.
It is entirely possible that with exhaustive empirical visual testing of a variety of raw files that I would draw very different conclusions than what the hard numbers seem to be telling me.
Conclusion: for someone in search of the best continuous tonal range available in affordable, compact digital, I would really like a definitive explanation of what the heck is going on!
(I'm posting a virtual copy in the Alpha forum, pardon me those who visit both forums.)
- There are 1530 distinct values in my best raw file out of the sample set of files I tried.
- The range of values is 0-4093 (which is twelve bits per channel output, not 14, alas)
- The values in certain ranges are consolidated, in other words:
- 0-800 contains 801 unique values (i.e. is continuous)
- 801-1424 contains 320 unique values (skips 1 out of every 2)
- 1424-1427 contains 1 unique value (skips 2 of 3)
- 1428-2023 contains 149 unique values (skips 3 of 4)
- 2024-2029 contains 1 unique value (skips 5 of 6)
- 2030-4093 contains 258 unique values (skips of 8)