Fujifilm X100 CLEVER ISOs above ISO1600

[Guillermo Luijk]

Looking at DxOMark results for the new brand Fujifilm X100, one gets this:

Effective ISOs above ISO1600 remain the same as ISO1600. This, contrarily to what many could think, is GOOD news and an indication of the clever guys from Fujifilm (or we'd rather say the silly guys from other brands like Canon or Nikon).

It's not a matter of digital vs real (analogue amplification) ISO. ISOs above ISO1600 are digital in most cameras (in some above ISO3200), and for high ISO settings the discusion digital vs real is irrelevant since having real ISOs above ISO1600 doesn't provide any advantage on any camera for the RAW shooter.

The good thing here is that unlike in most cameras, the Fuji X100 shooting above ISO1600 doesn't care of digitally mutiplying the RAW data by the corresponding factor of 2 before saving the RAW file. In this way, ISO3200 on this camera is just ISO1600 plus some metadata telling the RAW developer 'hey! this is supposed to be ISO3200, so display it to the uninformed user 1 stop brighter than it really is'.

What is the good news? the X100 RAW shooter above ISO1600 gets a correct exposure in the camera display and in-camera JPEG, but keeps in the RAW file all the highlights information that other cameras stupidly ruin just for multiplying the RAW data to achieve the illusion of high ISO.

A collateral effect of this X100 behaviour will be many users claiming their X100 has an incredibly good highlight headroom when shooting ISO3200 and specially ISO6400. This is due to the fact that ISO3200 and ISO6400 in the X100 will mean a 1 and 2 stops respectively RAW data underexposure, where other cameras insist on digitally pushing their high ISO RAW data, blowing all that highlight information. We should add to this the fact that measured ISOs for the X100 are nearly 1 stop below the standard expected ISO.

This is actually no advantage to the X100 if you really know how your Canon/Nikon camera works (simply never use ISOs above ISO1600/ISO3200 when shooting RAW and expose according to your camera's effective ISO), but I always wondered why those manufacturers were so stubborn to digitally multiply high ISO RAW data.

Regards

[ErikKaffehr]

Thumbs up for Guillermo and Fujifilm!

Good thinking!
Erik

Looking at DxOMark results for the new brand Fujifilm X100, one gets this:

Effective ISOs above ISO1600 remain the same as ISO1600. This, contrarily to what many could think, is GOOD news and an indication of the clever guys from Fujifilm (or we'd rather say the silly guys from other brands like Canon or Nikon).

It's not a matter of digital vs real (analogue amplification) ISO. ISOs above ISO1600 are digital in most cameras (in some above ISO3200), and for high ISO settings the discusion digital vs real is irrelevant since having real ISOs above ISO1600 doesn't provide any advantage on any camera for the RAW shooter.

The good thing here is that unlike in most cameras, the Fuji X100 shooting above ISO1600 doesn't care of digitally mutiplying the RAW data by the corresponding factor of 2 before saving the RAW file. In this way, ISO3200 on this camera is just ISO1600 plus some metadata telling the RAW developer 'hey! this is supposed to be ISO3200, so display it to the uninformed user 1 stop brighter than it really is'.

What is the good news? the X100 RAW shooter above ISO1600 gets a correct exposure in the camera display and in-camera JPEG, but keeps in the RAW file all the highlights information that other cameras stupidly ruin just for multiplying the RAW data to achieve the illusion of high ISO.

A collateral effect of this X100 behaviour will be many users claiming their X100 has an incredibly good highlight headroom when shooting ISO3200 and specially ISO6400. This is due to the fact that ISO3200 and ISO6400 in the X100 will mean a 1 and 2 stops respectively RAW data underexposure, where other cameras insist on digitally pushing their high ISO RAW data, blowing all that highlight information. We should add to this the fact that measured ISOs for the X100 are nearly 1 stop below the standard expected ISO.

This is actually no advantage to the X100 if you really know how your Canon/Nikon camera works (simply never use ISOs above ISO1600/ISO3200 when shooting RAW and expose according to your camera's effective ISO), but I always wondered why those manufacturers were so stubborn to digitally multiply high ISO RAW data.

Regards

[kencameron]

How can I find out the effective ISO for my other cameras (5dmkii and 40d)?

[Guillermo Luijk]

How can I find out the effective ISO for my other cameras (5dmkii and 40d)?

Using them, and seeing how much highlight headroom they provide with respect to the their metering.

[meehee dee]

Hi Guillermo,
thank you for you interesting post. I have seen dpreview saying the same thing and I wasn't aware most digital cameras are using this technique. I still have a question and I hope it is not to stupid. If the X100 uses underexposed ISO1600s for ISO3200 and 6400, it has to "stretch it's Bit depth out of the "shadow areas" of the file to get an correctly exposed image, or not?? So you would loose Bit depth (especially in the darker areas)??
Thanks,
Michael

[Guillermo Luijk]

So you would loose Bit depth (especially in the darker areas)??

Correct, you loose bitdepth, exactly the same as any other camera that builds its high ISOs from lower ISOs. The difference is that in the X100 this happens in the RAW development, not in the camera, so in the X100 you don't waste highlights information that should be easily usable by the RAW developer.

E.g. these histograms are ISO3200 and ISO6400 on a Canon 5D2. ISO6400 is ISO3200 expanded by 1EV before saving the RAW data.


The loss of bitdepth is not a problem here (noise dithers any possible posterization), the problem is most cameras are needlessly loosing highlights information in their high ISOs just because they are expanding the lower ISO histogram in the camera.

Regards

[meehee dee]

Thanks a lot! This was very helpful!
Best regards,
Michael

[meehee dee]

I have another, related question, maybe you can help me here, too. It is about to the DR-Settings of the X100.
What dpreview tells me and what my own experience using Lightroom is, is, that once you put DR200 (DR400) on, the camera will underexpose the RAW by one (two) stop(s) and (in JPEG-MODE) uses it's software and RAW processing engine to correct the exposure, pushing details from the "shadow area" of the picture. To get the underexposed images, the camera fools itself and does not apply any analogue amplification to the sensors (ISO increase) but measures the exposure time and aperture as if it had been increased. In other words it still uses ISO200 but measures as if it would use ISO400 (ISO800). If I understood this correctly so far, this should also lead to a loss of bit depth (as did the ISO3200 and ISO6400 boosts). This should be most prominent in the shadow area of the picture since in the end the picture was "exposed to the left" to rescue highlight information (sacrificing shadow details and bit depth, I would say). Can everybody agree or am I missing something??
Thanks and best regards,
Michael

[Guillermo Luijk]

If I understood this correctly so far, this should also lead to a loss of bit depth (as did the ISO3200 and ISO6400 boosts). This should be most prominent in the shadow area of the picture since in the end the picture was "exposed to the left" to rescue highlight information (sacrificing shadow details and bit depth, I would say). Can everybody agree or am I missing something??

You perfectly understood how the DR modes work (in fact it's the same as Canon's HTP). But your conclusion about bitdepth loss is again wrong: bitdepth loss in the RAW file because of these ISO tricks is NEVER a problem in practice because noise will always dither any possible posterization. If you can obtain a visibly posterized RAW (forget the histograms), no matter with which camera, no matter which ISO setting, I will buy you a new camera of any brand and any price  

Left image is 8-bit, right image is 5-bit. The bitdepth loss didn't mean any visible quality loss because of posterization:



Regards

[meehee dee]

Hi Guillermo,
thanks again for your answer. So I am not necessarily wrong, but it doesn't matter in real life is what you are saying, right? But in a way bit depth only isn't a problem because you got bigger one's to deal with if you are applying the DR (ISO) functions. So you basically introduce two negative effects, of which one isn't so obvious but apparent. I attached a picture which was taken in RAW and with and without DR400 (the DR400 was processed in Lightroom with my best skills to match the correctly exposed image) and both magnified on the dark area (door). So we do have lots of noise but possibly also posterization at the door parts, don't you think. Also theoretically with around 7 stops of dynamic range, the first stop has 2048 "brightness levels but the last one would only have 32 (and that would be without any contrast added). So shouldn't it easily end up as a dark mash that is obviously also very noisy??
Thanks a lot and best regards,
Michael

[Guillermo Luijk]

the first stop has 2048 "brightness levels but the last one would only have 32 (and that would be without any contrast added). So shouldn't it easily end up as a dark mash that is obviously also very noisy??

In practice you'd only see the noise, so in practice the lack of levels will never be a problem, no matter if you add contrast (you would be also contrasting noise).

Paradoxically, losing bitdepth is usually a problem only in absence of noise, so it is not likely to happen in the shadows. When a large surface of uniform colour has a high SNR, conversion to 8-bit (JPEG) may end in visible posterization. This is typical in skies: they usually have a good SNR for being the brightest parts of the capture. If there are no clouds providing texture, conversion to JPEG can produce banding, and the solution is to introduce some dithering mechanism (e.g. adding noise).

Look at this example: wall of plain colour, in the first version noise dithers the bands:


But if we eliminate noise (something that sounds a good idea in principle), bands become visible:


If the images are too dark to see the bands, open them in PS and push exposure equally on both.

Regards

[meehee dee]

Ok, so one could also say that we just found another problem of underexposing 
Once you push the exposure, you might end up having noise in your shadows and posterization in you highlights..
So better be careful with your DR-Settings and only use it if the scene really has a big dynamic range, so that the highlights would still be there and have lots of details.
So in real world this should be a problem with the ISO boosting we talked about yesterday (in terms of noise and posterization)..
Best regards,
Michael

[aman74]

Just wanted to say thanks for the info and wanted to subscribe, but I see no option to do that on this forum...I'm new though and maybe missing it somewhere.