Jared Willson
Senior Member
- Messages
- 1,714
- Solutions
- 7
- Reaction score
- 1,894
I know with the X2D, the dynamic range was virtually identical in 14-bit mode as in 16-bit mode, so I decided to see whether that is true for the new X2Dii camera as well. Per Hasselblad's specifications, the new camera has a touch more dynamic range than its predecessor, so I decided to measure it.
I use PixInsight for most of my astrophotography, and it has the ability to read .3FR raw files which is convenient for measuring things like this. It includes a utility that will allow you to calculate dynamic range (as well as gain, read noise in electrons or ADU, dark current, full well capacity, etc.) using two bias frames, two dark frames, and two flat frames.
When I ran the metrics for both 16 bit images and for 14 bit images, the numbers were very close but not identical. I performed the testing twice, with two different sets of data taken a few hours apart, and got similar but not quite identical results both times.
In both data sets, the 16-bit files measured marginally better dynamic range. By marginally, I mean 14.8 stops vs. 14.7 stops in one run, and 14.7 stops vs. 14.6 stops in the other. These are pretty dang close. In fact, they are close enough that if I ran the test three or four more times I wouldn't be surprised to run into a situation where the 14 bit file measured more dynamic range than the 16 bit file.
It's possible that there is an ever-so-slight benefit to 16-bit files, but I can't prove it with my data sets. I think I will stick with 16-bit in situations where the EVF blackout time is completely irrelevant--like landscape shots taken on a tripod--but switch to 14 bit mode for virtually anything handheld. Luckily, I won't have to remember any of this as long as I get my profiles setup correctly for my particular uses.
By the way, I don't have a way to share the specific pixel math that PixInsight is using to generate the dynamic range, gain, and read noise. The measurements are contained in a script rather than in a process, and the help information in PixInsight which generally contains all the relevant math is simply absent for this one. I would not recommend using my measurements for comparison purposes only. That is, compare data sets that were measured the same way as I just did it, but don't compare my numbers to those from the Hasselblad spec sheet or from Bill Claff's results on Photons to Photos. I know Bill, for example, uses a higher noise floor in computing dynamic range since he only wants to consider shadows that have "useful" details that aren't too noisy. That's why he tends to report a DR that is significantly lower than manufacturer's specifications. He calls it "photographic dynamic range". As long as he is consistent in his calculations, it should allow valid comparisons. Same for what I just posted--valid to compare a 14-bit set of files in PixInsight to a 16-bit set of files in PixInsight, but since I don't know the actual pixel math used I wouldn't suggest anyone say, "Hey! 14.8 stops is less Fuji's GFX100S II." I would only recommend comparisons made using the exact same tools and methodology. And even then, I'm a little worried by some of the things I noticed... Specifically, don't know what Hasselblad's calibration utility does to raw files, but it's not nothing. I know it tends to address hot pixels, for example, by building some sort of bad pixel map. Is it doing more than that? Is it trying to subtract out dark current based on temperature and exposure length? I think it might be since in two of four cases in my samples my 40s dark frames actually had a slightly lower mean ADU than my bias frames. We're talking tiny differences here, but that shouldn't ever happen unless there was some processing happening to the raw files.
In any event, dynamic range seems very similar between 14 bit mode and 16 bit mode.
I use PixInsight for most of my astrophotography, and it has the ability to read .3FR raw files which is convenient for measuring things like this. It includes a utility that will allow you to calculate dynamic range (as well as gain, read noise in electrons or ADU, dark current, full well capacity, etc.) using two bias frames, two dark frames, and two flat frames.
When I ran the metrics for both 16 bit images and for 14 bit images, the numbers were very close but not identical. I performed the testing twice, with two different sets of data taken a few hours apart, and got similar but not quite identical results both times.
In both data sets, the 16-bit files measured marginally better dynamic range. By marginally, I mean 14.8 stops vs. 14.7 stops in one run, and 14.7 stops vs. 14.6 stops in the other. These are pretty dang close. In fact, they are close enough that if I ran the test three or four more times I wouldn't be surprised to run into a situation where the 14 bit file measured more dynamic range than the 16 bit file.
It's possible that there is an ever-so-slight benefit to 16-bit files, but I can't prove it with my data sets. I think I will stick with 16-bit in situations where the EVF blackout time is completely irrelevant--like landscape shots taken on a tripod--but switch to 14 bit mode for virtually anything handheld. Luckily, I won't have to remember any of this as long as I get my profiles setup correctly for my particular uses.
By the way, I don't have a way to share the specific pixel math that PixInsight is using to generate the dynamic range, gain, and read noise. The measurements are contained in a script rather than in a process, and the help information in PixInsight which generally contains all the relevant math is simply absent for this one. I would not recommend using my measurements for comparison purposes only. That is, compare data sets that were measured the same way as I just did it, but don't compare my numbers to those from the Hasselblad spec sheet or from Bill Claff's results on Photons to Photos. I know Bill, for example, uses a higher noise floor in computing dynamic range since he only wants to consider shadows that have "useful" details that aren't too noisy. That's why he tends to report a DR that is significantly lower than manufacturer's specifications. He calls it "photographic dynamic range". As long as he is consistent in his calculations, it should allow valid comparisons. Same for what I just posted--valid to compare a 14-bit set of files in PixInsight to a 16-bit set of files in PixInsight, but since I don't know the actual pixel math used I wouldn't suggest anyone say, "Hey! 14.8 stops is less Fuji's GFX100S II." I would only recommend comparisons made using the exact same tools and methodology. And even then, I'm a little worried by some of the things I noticed... Specifically, don't know what Hasselblad's calibration utility does to raw files, but it's not nothing. I know it tends to address hot pixels, for example, by building some sort of bad pixel map. Is it doing more than that? Is it trying to subtract out dark current based on temperature and exposure length? I think it might be since in two of four cases in my samples my 40s dark frames actually had a slightly lower mean ADU than my bias frames. We're talking tiny differences here, but that shouldn't ever happen unless there was some processing happening to the raw files.
In any event, dynamic range seems very similar between 14 bit mode and 16 bit mode.
