Do have iso160-multiples have more dr & less noise?

[chmee]

i'm not quite sure if it fits into this thread or in the adtg-investigation. on this part i am only half-knowlegded and you're quite beyond my horizon. the worldwide community (at least the german part) cant really cope with this new finding. and i have some basic questions as well.

* if we're just optimizing the sensorgain - we dont change the base noise in the shadows, just the noise appearance in the middles/highlights, isnt it?
* if we change the multiplier for the sensorgain per iso, we change the idea of REI (recommended exposure index), isnt it?
* shouldnt do ETTR basically the same? we lose potential highlight-data, but improve the appearance in the lights..
* are the base iso's really native? according to the REI it isnt.
* are these quantifization glitches in sub-isos academic or real problems?

the main finding is (for me), that the t-stop-f-stop cheat was proven.

ah btw. is this the "simple" schematic look of the datapath right behind the sensor?


regards chmee

[a1ex]

1) noise is not changed, you get more details that otherwise would be clipped to white
2) I think so (the image will get darker, so you need to use a slower shutter speed or a wider aperture)
3) no
4) no (because ISO can go lower than 100 while still keeping the ADC saturated). However, "native ISO" is not really well defined (see this discussion).
5) I don't know

For schematic, I believe there are at least two analog amplification stages (CMOS and ADTG). Besides digital amplification, there might be some nonlinear corrections too (not sure). The rest matches my understanding.

[chmee]

a1ex, thx a lot. if there's a useable build, i'd be interested, how the professional raw-converters will react on these files, fi lightroom.

[PhotoCat]

I am a bit confused...

So do we have any conclusion?

Is my following understanding right?

1) ISO 160 multiples up to 1250 are good (low noise) for jpgs only, knowing that some highlight will be clipped (less DR)
2) ISO 100 multiples are good for raw shooters, as the raw file is not manipulated digitally in camera. ETTR will give as good SNR as 160 multiples??
3) Avoid ISO 125 multiples

Please help! Tks!

[a1ex]

My current opinion (hope you don't mind the LaTeX formatting; I'm writing a detailed article with my findings):

\subsection{So, does ISO 160 have less noise and more DR than ISO 100 or 200?}

\begin{itemize}
\item \textbf{Photo RAW (CR2)}:
\begin{itemize}
\item On recent cameras (550D and newer), these ISOs do have around 0.1 stops of more dynamic range, compared to their full-stop counterparts.
\item On older cameras (5D2/50D), there's no benefit in using ISO 160 multiples; they are actually a bit worse because of round-off errors.
\end{itemize}
\item \textbf{Movie RAW}: not at all. ISO 160 = 200 = 250, and the only difference is in the preview image, not in the recorded RAW/MLV.
\item \textbf{JPEG / H.264}: yes, because Canon's JPEG engine is cliping things to white quite early. However, you may get slightly better results (nicer highlight rolloff) from ISO 200 pulled to ISO 160 from Magic~Lantern controls (ML digital ISO).
\end{itemize}

\subsection{Does ISO 125 have more noise and less DR than ISO 100 or 200?}

\begin{itemize}
\item \textbf{Photo RAW (CR2)}: yes, because they are implemented by increasing the ADTG gain, which does~not~improve the noise floor.
\item \textbf{Movie RAW}: not at all (see previous question).
\item \textbf{JPEG / H.264}: yes.
\end{itemize}

[PhotoCat]

Wow, thanks Alex for the quick reply!

So for older cameras like 5D2, we should be shooting at ISO 100 multiples to have the highest quality raw file,
if I understand correctly.

Just wondering if it is possible, when shooting at say ISO 1600, to meter/evaluate (false color etc) at EC + 1/3 stop
such that we are shooting ETTR by +1/3 stop and have ML generate a pulled version (by -1/3EV) of the jpg file.
This way, we will hopefully have *Both* a good raw file and a "cleaner" jpg file at the same time in camera,
without any further post processing.

Please keep up with the great work!

[a1ex]

The digital gain is also burned in the CR2 file (couldn't separate them yet).

[PhotoCat]

not sure if I follow u Alex but I guess my assumption of zero digital gain at ISO 1600 or at multiple of 100 is wrong then...

[PhotoCat]

Oh... I think I may have got it now. We need to somehow program the digital multiplier into the raw file in question
so that the Canon engine can pick it up and convert the raw file into a jpg file in camera.