Sony a900/Lightroom Shadow Detail Test
Back in May of this year I ran an experiment with the Sony a900 and Lightroom, looking at shadow details to see what sort of results I would get at different ISOs. There is a lot of discussion around what is the “best” ISO with these cameras, and since it seems to depend on how the photographer exposes and what RAW conversion software is used I decided to perform some tests starting with the test of shadow noise handling described in this article.
Not long after buying this camera I found some information from Iliah Borg indicating that ISO 320 offered the greatest dynamic range. Although this is well above base ISO Iliah had stated that ISO 320 was required “to unstop the shadows”. I shot thousands of pictures at ISO 320 with good results but finally decided to test for myself a bit more in-depth, as I had heard photographers whose work I respect and admire suggesting ISO 160 as a cleaner setting to use.
I shot the following set of pictures in my garage, careful to avoid blocking up shadows. The blown-out side window is not relevant for this test, later I will do some testing around midtones and highlights.
Let me emphasize that this is NOT the best-controlled experiment, there are some minor exposure inaccuracies apparently induced by the camera mechanism or variations in light level. I calculated the exposure to be as low-key as possible without clipping any blacks and I think I ended up pretty darn close. I shot at this exposure in M mode at ISO 100, then in steps I increased the ISO by 1/3 stop and reduced the shutter speed by 1/3 stop. Aperture was f/2.8 as I didn’t want to get into very long exposures and possible heat issues. To be fair, I imagine it’s possible I’ve introduced some of that as I was starting at a 2 sec. exposure. Camera was mounted to a sturdy tripod and locked down – all shots with MLU and 3 sec. settle time after lifting mirror.
I imported into LR and for all pics set to Adobe standard, 2010 process(not that process version matters to my eye with everything zeroed) linear tone curve, no sharpening or NR, all sliders including brightness/contrast to zero except where noted.
In the above composite of the full set of shots you can see the exposure variations. The variations are very small, but since the variations did not result in clipped blacks I decided this set of files would do for what I had in mind. The most important thing for this test was that I had data all the way down to the lowest values with no shadow clipping in any of the files. It’s interesting(to me at least) to see that the histogram starts pushing to the right at ISO 1250 despite the fact that I kept the EV identical, and from there every shot at the next higher ISO has the histogram pushed further to the right. I could see this by just clicking the images in the LR filmstrip in succession and watching the histogram. Here is what that looks like:
For this next set of shots it’s important to note that they are not EXACTLY what was on my screen. They are JPEG saves out of PS, where my screenshots were all pasted straight from the clipboard into individual layers. I double-checked by pulling one of these saved JPEGS in as a new layer and doing a difference blend vs the original. Waving the cursor around I only saw differences of 1 or less (out of 255), but there is a difference nevertheless. To get a more accurate look at what was on my screen you can download the PSD here(100MB).
One other thing to note is that I did most of my peeping in the evening when it was dark, some of what I describe is less obvious in the daytime even though I do use different calibration profiles for day & night.
All of these have on the left my old friend ISO 320, on the right are the other exposures. These are all thumbnails, the 100% screenshot should open when you click the thumbnail. I zoomed in on an area that has(or should have) smooth tone transitions(The cylindrical motor top center) and fine detail in the AC condenser fins.
On the lower part of the motor you can see where the tones are blocked up on the ISO 100 side, the ISO 320 side may just look noisy to you but in fact that cleans up with NR as I’ll show later.
Also take a good look at the fan bracket. That is the trapezoidal feature at the bottom, left of center, under the connector. This is a good example of the classic “Red/Blue blotchiness” that can lead to much hair-tearing.
On the lower part of the motor now the red/blue has taken on more of a blue tendency.
On the fan bracket the blotchiness is if anything worse, again tending more toward blue.
This is the surprise, for me at least. Suddenly the blotchiness is gone. There are only traces of it. Here, on the fan bracket, and, well… pretty much everywhere there are smoother transitions and tones(especially after a bit of NR, as I’ll show).
Also take a look at the bolt that is just above the connector in the photo. There is far more detail here than in the ISO 320 shot(and elsewhere as will become apparent).
Back to some blotchiness on the lower part of the motor, you kind of have to look carefully to see the broad swaths of same-tone areas, but these will begin to pop out once you begin lifting shadows.
On the fan bracket it’s not so pronounced in this case but there is plenty of blotchiness lurking throughout the image just waiting for a contrast adjustment to make itself known.
On the lower part of the motor we see some improvement. Sure there is noise but at least we don’t have the blotches anymore. The noise will clean up to show relatively smooth tones across this area — at least, more so than at ISO 200.
On the fan bracket the blotchiness is if anything worse, again tending more toward blue.
Dang, I knew I forgot something… I don’t have a comparison of ISO 320 & ISO 320.
I just wanted to include this so you can see color noise start to take over and cloud these issues at and beyond 400.
Interestingly, it seems that other multiples of ISO 160(e.g. 640 and 1250) look better than the settings immediately above and below each. You can download all of the screenshots that I did or didn’t post above from here.
From there I made some virtual copies and started to play.
It’s not at all uncommon to want to “lift the shadows”, the above shows the color noise added to the ISO 320 capture when the exposure is raised 2 stops. The ISO 160 capture is still looking pretty good.
In these I adjusted NR on both photos, a quick & dirty adjustment to just knock the noise back in each to about the same point. There is a significant subjective element here, right or wrong here are the adjustments made in LR 3.4:
ISO 320: Luminance 61 Detail 50 Contrast 0 Color 23 Detail 50
ISO 160: Luminance 26 Detail 50 Contrast 0 Color 19 Detail 50
No other adjustments were made. Looking back as I type this, this is very surprising. I don’t think I’ve used a luminance adjustment over 40 or so more than a couple of times in the past. I wasn’t really looking at the numbers, just trying to get the noise levels similar on both. This really says something about the ISO 160 result, that I had to go so far to get similar – looking results in the 320 shot. On the one hand I probably crushed a bit of detail in the process, but I was impressed at how clean and detailed the radiator fins are at 160! Also the mounting bolt above the connector – much more detailed. I just went back as I am typing this and played with the slider on the 320 capture and there really isn’t much more detail than shown – just noise.
I’m not even sure why I did this. I could have managed with less NR in this area, I just wanted to see what it would look like in the shadows with that same NR, for example if I was paying attention to NR on another, brighter part of the image and this was getting NR’ed at the same time. It’s hard to see the difference but on the ISO 160 capture there is a smoother gradation of tones on the motor and the red blotchiness that the NR added to the arm directly in front of the motor on the ISO 320 shot is absent.
I’m sure there are any number of people who can look at this and say exactly why and how everything shown here comes to pass. Or maybe not, these captures were intended as a practical look and probably lack the precision of light source and capture target needed for a proper scientific evaluation. Still, I think we would all welcome comments from anyone with knowledge gleaned from really *good* tests. :cheesy:
If anyone wants to play with the original files(ideally, preparatory to commenting and sharing findings here) I have made those available as well. Processing and comparisons with other RAW converters welcome, especially RPP if you have it… I’ve sort of made my choice already among the PC-compatible converters and I think I know what to expect from them. The JPEGs are straight LR3.4 exports with no adjustments as described above, also no resizing, quality reduction, or sharpening on export.
I’m not sure when I’ll have time to get to my follow-up tests, so don’t wait for me. Go do your own testing to determine what gives you the most satisfying results, and feel free to comment with your findings!
Note: The results of this test were first posted on page three of this Dyxum thread. There is lots of good info and links in that thread for anyone wishing to dig deeper into this subject.