According to Photons to Photos, my ICLE-6700 is mostly ISO invariant after ISO 320. This made me wonder: could I just consistently shoot at ISO 320 and brighten the shots to the desired brightness in Darktable afterwards?
The following is a shot of a silly little constructed and dark scene, once at ISO 320, and once at ISO 12800 (i.e. +5.3EV relative to the ISO 320 shot). Both photos have been taken at F1.4 with a shutter speed of 1/4 and a consistent white balance. In my processing of both photos, the white balance and the highlight reconstruction modules are enabled using the defaults, while the color calibration module is set to the same color temperature (3425K) in both photos. The ISO 320 shot has been brightened to +5.3EV using the exposure module.
The brightened ISO 320 shot is considerably brighter, the colors are flatter and the noise looks more extreme in places. The histograms look very different. Surprisingly, the RAW file of the ISO 320 shot is also significantly smaller.
Am I misunderstanding something fundamental? Is my camera not actually ISO invariant? Is something strange happening in Darktable?
I’m really curious to learn more. Thanks in advance for all replies!
I am interested in this idea of ISO invariant. I see it as a great improvement to sensors. The range you have used here is extreme and maybe too extreme to be a perfect test, but your results are good. If I tried this with my Canon R7 the results would be crap from 320 ISO if 12800 was required. I will test this later to make sure I am correct in my presumption. Maybe try comparing ISO 320 and ISO 3200 as that is still a huge forgiveness range in real world terms.
I just processed your image and your camera is ISO invariant. Considering how black the 320 ISO image was it came up amazingly well. I set target exposure and target color to the 12800 image and then used the eyedroppers in the exposure module and color calibration module to set the 320 ISO image to the same values. In my case the 320 ISO is a little duller but this is still good is in a real world shoot you had underexposed by 5 stops.
Here is my test with a Canon R7 12800 ISO (left) and 320 ISO (right)
I will take your Sony sensor any day over my sensor. But the Canon performed better than I anticipated for both detail and noise.
But interestingly (to me at least ) if I only worry about the painting itself and not the dark wood I can tweak the contrast in the AgX module and get very similar result from the 320 ISO (bottom image)
The ISO320 image is definitely brighter. If you use the “area exposure mapping” feature in the Exposure module, you should be able to match the exposures exactly, and then the noise comparison should be “fairer”.
This is not surprising, as a6700 mandates compression. Your ISO 12800 shot might use all 14 bits after analog to digital conversion, whereas the ISO 320 one will use 9 or 10 as it is underexposed, so it will compress much better (the difference should be even more striking in lossy mode).
Re shadow noise, you might now be seeing effects of quantization noise after digital gain in post as you’re underexposed and using only 9-10b of the ADC, most likely w/ compression side effects as well.
Re exposure difference, ISO markings (esp. the intermediate ones) are not “strict” in a sense, so the 5.3 EV might not be exact.
So, “ISO invariant” doesn’t mean you can set a6700 to ISO 320 and forget. You always want to expose properly and utilize the full range of the ADC. It just means that you have (almost) nothing further to gain in shadows by increasing ISO (the input-referred read noise tells the same story) should you choose to sacrifice some highlights (cf. earlier tech of a6000). Otherwise, for the given aperture and shutter speed, just choose the highest ISO possible where you don’t blow the highlights.
Thanks for all the great answers, everyone! I am now less confused.
I see. So if I am understanding correctly, then (by underexposing) I have significantly reduced the resolution of the output of the ADC, which is presumably the cause of the color shifts in the underexposed shot? Is this also why the vignetting is less pronounced in the underexposed shot?
Thanks, I wasn’t aware of this! After properly normalizing the exposure to +4.8EV using @europlatus’ area exposure mapping suggestion, the noise levels of both shots seem to be comparable in areas of similar brightness (though their colors still differ and the vignetting remains less pronounced in the underexposed shot), which is what I would expect given the Photons to Photos graph.
Just to double-check my understanding: I assume this is true even in situations where I will have to decrease the exposure in post for the shot after “exposing to the right” by increasing ISO as you suggest, so that I can make use of the full resolution of the output of the ADC, correct?
Right. Although, if the subject allows, it would be preferable to reduce shutter speed (or open up the aperture) than raise ISO (in this “ISO invariant” case only of course). More photons captured is always better.
Raw files from my R6. ISO 1600 vs ISO 6400 vs ISO 51200. It seems like there is some kind of hot pixel suppression from ISO 12800 with this camera model.
Cameras often report a slightly different ISO on the dial than tests come back with, but the graphs I’ve seen show 1 stop difference in output for each stop difference in requested ISO. There was just an offset.
For example. ISO 100 may have measured as ISO 83 on an independent test, but ISO 200 was still one stop higher than ISO100 and ISO400 one stop higher than that.
Ah. Then I am once again curious about what might be causing the difference in brightness
I can consistently reproduce the difference in brightness between ISO 320 at +5.3EV in Darktable and ISO 12800 across several different scenes and even notice a more subtle shift in brightness at ISO 1600 at +2EV compared to ISO 6400. I briefly thought that the illumination of the camera display might be causing it, but turning it off does not change the outcome.
OTOH, boosting @Peter’s ISO 1600 shot to +5EV and comparing it to the ISO 51200 shot looks practically identical in terms of exposure to me (though the colors are obviously also a bit flatter in the boosted ISO 1600 shot).
) if I only worry about the painting itself and not the dark wood I can tweak the contrast in the AgX module and get very similar result from the 320 ISO (bottom image)
