*Based on testing of my own a6500 camera. This is different than reported at Photonstophotos.net which shows ISO 320 as the second base ISO for the a6500.
**Refer to this article for some clues as to why.
If you find any of this information useful, please let me know and leave a comment.
Talk of ISO invariance has been gaining popularity with camera geeks for the past few years and hit the frenzy level a couple of months ago when Tony Northrup made his "ISO is Fake" video which stirred up a lot of controversy. Tony created a followup video on ISO Invariance that really intrigued me because he demonstrated the effects of "Dual Gain Architecture" used in many modern camera sensors including ones made by Sony. I have read about ISO invariance a few times but never took much interest in it because I didn't see a practical implication for my photography. It's a complex topic that kind of turns the exposure paradigm upside down for digital cameras; if I can't benefit from it, why spend my time trying to figure it out?
ISO invariance is the property of some cameras that allow the exposure to be adjusted in post-processing as if the ISO had been increased at the time of capture. For example, a raw image file from an ISO invariant (or ISO-less) camera will produce roughly equivalent images whether the image is captured at ISO 3200 in camera or captured at ISO 400 in camera and then brightened by +3 stops in Lightroom. (And if you don't understand why it is 3 stops, you should probably stop reading this post now.)
Tony sums up the practical benefits of ISO invariance in his video: "High ISOs will clip the highlights, low ISOs will not, and often the noise levels are identical."
Dual Gain sensor architecture essentially means that the camera sensor has more than one base ISO. Essentially, the camera has two ISO values where noise is minimized and dynamic range (ability to capture highlight detail without clipping) is maximized. By knowing if a camera is ISO invariant and the base ISO for that camera, it is possible to take advantage of that knowledge to improve image quality in certain situations.
After watching Tony's video on ISO invariance, I saw another video by Justin Majeczy that demonstrated the benefits of taking advantage of ISO invariance for night and astrophotography. Justin found that he was losing a lot of highlight detail in his night imagery by using higher ISOs in camera. With an ISO invariant camera, he could shoot at lower ISO with the exact same total exposure (e.g., 15 seconds at f/2.0), then increase the brightness in post producing an image with no increase in noise and much improved highlight detail.
I wanted to test this concept with my Sony a6500 to see if I could get the same benefits. The first step is to determine the base ISOs for the camera, so I headed over to photonstophotos.net and started looking at the charts. Several of the charts for noise for the a6500 show the dual gain ISO to occur at ISO 320. Interestingly, the same charts for the a6300 and a6400 which have very similar sensors show the dual gain ISO to occur at ISO 400. More on this later.
I set the camera up on the tripod in my office and pointed it at the corner with my espresso maker that includes very dark shadows along with very bright outdoor highlights through the French doors. Starting with ISO 320 as the base, I decided to use ISO 5000 as my default exposure because it is 4 stops brighter than ISO 320 and would be a reasonable ISO to use for astrophotography. The shutter speed for the metered exposure at ISO 5000 and f/8.0 was 1/13 second, so I lowered the ISO to 320 and captured a series of exposures at 1-stop ISO intervals (320, 640, 1250, 2500, 5000, 10000, and 20000). Then in Lightroom, I changed the brightness values to match the ISO 5000 image, i.e., +4, +3, +2, +1, 0, -1, and -2. I also set Highlights to -80 and Shadows to +80 so that blown highlights and shadow noise would be very obvious.
The results were not what I expected; the ISO 320 image was considerably noisier than the ISO 5000 image. Looking at the detail comparison below, the ISO 320 image has larger white specks as opposed to the more uniform grain in the ISO 5000 image. Also note the improved highlight detail in the LED left of the gage. The increased noise in the ISO 320 image would be expected with a non-ISO invariant camera, but I had already seen from the charts that the a6500 should be ISO invariant from ISO 320.
Going back to Photonstophotos.net, the charts for the a6300 (and a6400) indicate the dual gain ISO kicks in at ISO 400 rather than ISO 320 as shown for the a6500. These camera all share basically the same sensor-it is reportedly identical for the a6300 and a6500 while the newer a6400 may be slightly different and has a better image processor. Based on this information, I decided to try the test again using ISO 400 as the base ISO to see if the results improved.
The image below shows a side-by-side comparison of the ISO 320 exposure to the ISO 400 exposure with both brightened by 4 stops in Lightroom. I used a slightly longer focal length for the ISO 400 series, so these images are not exactly lined up at 100%. I zoomed in on the darkest part of the photo to show the substantial difference in noise levels between these two images. The ISO 400 image on the right is much cleaner without a lot of the salt noise so prevalent in the ISO 320 image. The ISO 400 image also has much cleaner colors while the ISO 320 image suffers from some major color contamination. This comparison clearly shows that, at least for my a6500, dual gain ISO occurs at ISO 400, contrary to the results published on Photonstophotos.net.
Now knowing the ISO 400 is the appropriate second base ISO to use for testing the a6500, I repeated the experiment starting at ISO 400 and captured images at ISO 400, 800, 1600, 3200, 6400, 12800, 21600, and 51200. This time the comparison of the underexposed ISO 400 exposure to the ISO 6400 exposure revealed little difference. Processed in Lightroom, I see less noise in the deep shadows in the pushed ISO 400 exposure but at the expense of some color variations.
However, looking at the highlights, there is a substantial loss of highlight detail in the ISO 6400 exposure. Looking at the comparison below, outside the window there is a white pillow with a green fern embroidered on it. The fern is clearly visible in the ISO 400 exposure, but completely lost in the ISO 6400 exposure. Similarly, all color is lost from the tan electrical panel in the upper left corner-it appears as flat gray in the ISO 6400 image.
Sometime in the middle of all of this testing, I came across a post by Thom Hogan that was very critical of ISO invariance and blamed a lot of the misinformation on use of imprecise math by Adobe in their raw converter (Thom's post is very informative and well-written). Thom holds up Raw Photo Processor as the gold standard for raw processing, but also mentions that RawTherapee uses precise floating point math, so I decided to run the images through RawTherapee (version 5.6 in Windows) to see what it can do. I also have Capture 1 (version 12.0) installed on my system, so I gave it a try as well.
I tried to use similar settings as in Lightroom. For Capture 1, I set Highlights and Shadows to 100 and Noise Reduction Luminance to 0 and Color to 10. In RawTherapee, I set Highlight Compression to 500, Shadow Compression to 50, Luminance Noise Reduction to 0, and Chromiance Noise Reduction to Automatic Global.
The Capture 1 results looks very similar to Lightroom with a little more noise in the ISO 400 exposure and a little less noise in the ISO 6400 exposure. In Raw Therapee, the ISO 6400 exposure is very clean while there is some very noticeable color contamination in the ISO 400 exposure. Just based on the shadows comparison, I would have a hard time choosing between the ISO 400 image from Lightroom and the ISO 6400 image in RawTherapee.
Although there does not appear to be much of a noise penalty for shooting at ISO 6400, there is a substantial loss of highlight detail. Looking at the series of images, the ISO 3200 exposure shows some loss of detail, but I didn't notice any loss in the images shot at ISO 1600 and below. In the shadows, I don't see much difference at all between ISO 1600 and ISO 6400. Based on this comparison, I think ISO 1600 provides a good tradeoff for maintaining dynamic range and highlight detail while providing a usable preview image and letting the camera do some of the heavy lifting of boosting the exposure from the base ISO. I will capture some exposures at ISO 400, 1600, and 6400 to confirm this choice the next time I get to shoot at night.
*Update: I shot images for a Milky Way panorama recently and captured exposures at ISO 1600 and 3200. That's only 1 stop difference in exposure, so not a great test, but honestly I can't tell any difference in the images.
I have added this final section just to show some comparisons to demonstrate that high ISO by itself does not cause noise, but rather, it is the lack of light that necessitates shooting at high ISO that leads to noisy images. Martin Bailey recently talked about this on his podcast and recommended increasing ISO to reduce grain.
I wanted to test his recommendation, so I captured these images as part of my ISO invariance test to explore using high ISO and found these results to be rather interesting. The first comparison shows the ISO 6400 exposure (the proper metered exposure), and an ISO 25600 exposure lowered 2 stops in post. The ISO 25600 exposure does show substantially more blown highlights outside the window, but in other areas of the image it actually looks much better. The image details are slightly better, and the shadows have much lower noise.
The final comparison below is intended to demonstrate that the amount of light captured has more influence on image noise than ISO. For this comparison, I set the ISO to 51200. The image on the left was captured as part of the test, so it is 3 stops overexposed from the metered correct exposure at ISO 6400. The image on the right was captured at the same ISO but using the proper metered exposure setting. It is clearly evident that reducing the amount of light captured by 3 stops drastically increases the noise in the image. Although the image is fairly clean of shadow noise, the loss of highlight and upper midtone detail at 3 stops overexposed at high ISO is unacceptable.
I think Martin is correct in his assessment of lowering noise by increasing the ISO, but I'm not sure I would even intentionally shoot this way (Martin does explain in detail how to protect the highlights using this method). Given the very good noise reduction capabilities now available in almost every raw processor, I think I would rather save the highlights and use software to reduce grain rather than risk capturing an unusable image.