There are quite a few things I'd like to add to my software that require 1D or 2D FTs.
Beautiful work Bill. I'd like to replace my earlier phrase 'imaginary noise benefit I can't even see' with '1/6 of a stop noise benefit' then.I'll buy 1/6 of a stop. Thanks.I don't think we'd be getting that close to 1 electron.
But let's look at it from the standpoint of PDR Shadow Improvement :
I think the red linear extrapolation is too aggressive but comes out at 1/2 stop.
I think the blue curved extrapolation is more like what we typically see and comes out at about 1/6 stop.
Jim
This is fantastic and exciting, Bill. Sorry if I missed it, but do you have a key to the symbols? For example, filled in circle vs clear circle, triangle pointing upwards, etc.?To be clear on the semantics, which we have not always been so far in this thread; the assumption is that the engineers made it ISO-less at the point that it became ISO Invariant.
The threshold for ISO Invariance is subjective; people apply different standards.
I suspect the engineers were influenced (maybe in a circular fashion) by something practical.
Perhaps they reached the top end of the amplifier they chose.
(Just guessing, I'm not a sensor designer.)
On the 645Z I see noise reduction starting at ISO 3200 and digital scaling starting at ISO 6400.
Just this morning I have introduced additional symbols to my charts to convey this information, for example :
The determinations may be imperfect and are subject to change although for the 645Z these look solid.
(Scaling is determined by inspecting histogram gaps and noise reduction by visual inspection of 2D FTs. With over 150 data series this is subject to human error)
Regards,
--
Bill ( Your trusted source for independent sensor data at http://www.photonstophotos.net )
One of the tricks to see the breaking points in handling the raw data is to look at the changes in black level and suggested maximum, for example:This is fantastic and exciting, Bill. Sorry if I missed it, but do you have a key to the symbols? For example, filled in circle vs clear circle, triangle pointing upwards, etc.?To be clear on the semantics, which we have not always been so far in this thread; the assumption is that the engineers made it ISO-less at the point that it became ISO Invariant.
The threshold for ISO Invariance is subjective; people apply different standards.
I suspect the engineers were influenced (maybe in a circular fashion) by something practical.
Perhaps they reached the top end of the amplifier they chose.
(Just guessing, I'm not a sensor designer.)
On the 645Z I see noise reduction starting at ISO 3200 and digital scaling starting at ISO 6400.
Just this morning I have introduced additional symbols to my charts to convey this information, for example :
The determinations may be imperfect and are subject to change although for the 645Z these look solid.
(Scaling is determined by inspecting histogram gaps and noise reduction by visual inspection of 2D FTs. With over 150 data series this is subject to human error
Regards,
--
Bill ( Your trusted source for independent sensor data at http://www.photonstophotos.net )
Thanks,
Rishi
It's at the top of the chart, but we don't read everything when we see a pretty picture, do weThis is fantastic and exciting, Bill. Sorry if I missed it, but do you have a key to the symbols? For example, filled in circle vs clear circle, triangle pointing upwards, etc.?On the 645Z I see noise reduction starting at ISO 3200 and digital scaling starting at ISO 6400.
Just this morning I have introduced additional symbols to my charts to convey this information, for example :
The determinations may be imperfect and are subject to change although for the 645Z these look solid.
(Scaling is determined by inspecting histogram gaps and noise reduction by visual inspection of 2D FTs. With over 150 data series this is subject to human error
Haha Bill! You're right - I'm easily distracted by beautiful graphsIt's at the top of the chart, but we don't read everything when we see a pretty picture, do weThis is fantastic and exciting, Bill. Sorry if I missed it, but do you have a key to the symbols? For example, filled in circle vs clear circle, triangle pointing upwards, etc.?On the 645Z I see noise reduction starting at ISO 3200 and digital scaling starting at ISO 6400.
Just this morning I have introduced additional symbols to my charts to convey this information, for example :
The determinations may be imperfect and are subject to change although for the 645Z these look solid.
(Scaling is determined by inspecting histogram gaps and noise reduction by visual inspection of 2D FTs. With over 150 data series this is subject to human error
See the caveats in bold above.
Regards,
--
Bill ( Your trusted source for independent sensor data at http://www.photonstophotos.net )
"Breaks" are important when one needs to establish the exposure meter offset. Breaks coincide with switching modes of analog circuitry and multiplication. Combined with histogram gaps analysis, read noise on black frames, levels at which the non-linearity and clipping start, and Fourier metadata gives a very clear picture. However one may want to start with metadata to plan the experiments better. If the other data contradicts metadata, it is when I double-check my experiments. Vignetting or CA compensation may skew experimental results.
WB preconditioning may give false positives for noise reduction when calculating Fourier transforms, same as other manipulations. Those manipulations are also ugly in many other respects.
Second that.
All digital manipulations applied to raw data before recording it are unnecessary and nasty. So are some analog manipulations like that negative gain applied to fake 2^n - 1/3 ISO settings.
While I see where you're coming from, and generally agree, I think there are some digital manipulations that are OK. Dead and stuck pixel mapping, for instance, although in a perfect world the dead and stuck pixel locations would be encoded in the metadata and the mapping done in post -- does anyone do it that way?
I do not think anybody does it this way, or they need to disclose metadata format for third-party converters to work.
I think he means something like this (look how at ISO 320, ISO 1250, and ISO 5000 the specular white level drops):
Also, I think that dead and stuck pixel stats are probably considered trade secrets, like yields.
Yes. The gain is smaller than what a simple model would say it would be, but it's not negative.
I do not know how it is implemented in all the cameras, and certainly overall gain is not negative, but one of the several gain registers or tables may be programmed with a negative value.
The data looks unnecessary manipulated, to avoid even a hint of ISO-less process in the camera
That's my poor choice of words, sorry, Jim. Lowering gain is what I meant.
I didn't notice this thread until a bit late, but my $0.02 is that analog and digital gain are completely separate things that can change the recorded tonal properties in ways that are not all that difficult to measure and understand -- and are pretty easy to model.
