Isn’t this the chart we should all be looking at for ISO invariance? A straight line indicates an ISO invariant sensor.
https://www.photonstophotos.net/Charts/PDR_Shadow.htm#Canon%20EOS%20550D,FujiFilm%20X-T5
I am not so sure. I just loaded my Canon G16 compact into this graph (green line) and it matches the fuji on the graph, but I know it doesn’t in real life.
Well, according to my own understanding and what various other websites have said, I think this chart is supposed to be the one we check for ISO invariance, so either the G16 is in fact ISO invariant, or there’s something else going on I don’t understand.
The graph shows how much shadow noise improves as ISO increases, relative to base ISO. So, if the line is relatively flat, you’re seeing very little shadow improvement as ISO increases, in other words, ISO invariant.
FYI, in case no one caught this: “second base ISO” and “dual gain” are the same thing.
That particular graph is showing relative shadow improvement compared to base. Since it’s all relative to base ISO performance, it doesn’t show the effects of sensor size.
You’re probably thinking of the regular PDR chart which DOES show effects of sensor size:
https://www.photonstophotos.net/Charts/PDR.htm#Canon%20PowerShot%20G16,FujiFilm%20X-T5,Sony%20ILCE-6700
The G16 does appear to be ISO invariant, but also has a 3-stop penalty in PDR thanks to its smaller sensor.
In general, the effects of a dualgain system seem to be more pronounced as the photosite size increases, which also leads to the cutin point being at a higher ISO. Compare the A7S3 to A7M4 and A7R5 for example:
https://www.photonstophotos.net/Charts/PDR.htm#Sony%20ILCE-7M4,Sony%20ILCE-7RM5,Sony%20ILCE-7SM3
Also a warning: You need to be careful when comparing cameras from different manufacturers, since the ISO axis may be different, shifting their graphs left/right, as the ISO rating is defined based on the behaviors of the JPEG engine and not the raw sensor data. In general it’s best for comparison to compare the two cameras at their base, then compare them at “N stops above base”
5 Likes
Looks an awful lot like the 5D mk. II 
https://www.photonstophotos.net/Charts/RN_e.htm#Canon%20EOS%205D%20Mark%20II_14,Canon%20EOS%20R7_14
1 Like
Indeed! In the beginning there was Saturation-based Exposure Index - nicely tied to the actual sensor performance - a bit like film. Then came Standard Output Sensitivity which involved 8-bit conversion to sRGB. The most popular for manufacturers now is (of course) Recommended Exposure Index based on the manufacturer’s own testing with various scenes. Hmmm … no direct connection to the sensor performance any more …
Throw in I.S.O’s latitude tables and we’re talking almost 20% margin of error.
1 Like
So many of Bill Claff’s graphs here I’m beginning to lose track, grump.
Exactly which parameter is said to be “invariant” with ISO?
Certainly not this one, eh?
“input-referred” read noise, although as I mentioned, Claff says that measuring read noise can sometimes be difficult due to some cameras “cooking” their RAW slightly.
The best indicator seems to be if the PDR graph is a straight line in log-log space, or the “shadow improvement” graph is flat. Looks like read noise in DNs is also a straight line for such cameras, with the exception of the dualgain shift for Aptina licenseees.
2 Likes