CHDKPTP - PC Remote Control Performance Analysis - page 7 - RAW Shooting and Processing

reyalp

14079

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #60 on: 02 / September / 2012, 15:04:11 »

Advertisements

Quote from: SticK on 02 / September / 2012, 14:42:45

that I have noticed when I change the "market" value say by 1 f-stop in the shoot command, the "real" value (the one you are referring to in ISO given by CHDK) remains unchanged. That's what puzzles me. I recall talking about this before.

If the exif doesn't update, this is probably a separate issue. You might want to examine the canon Maker Note.

Quote

Useful for me then would a table then of "real" ISO values vs "market" values

You can easily generate such a table yourself. There are propcases which reflect both, so if you set the ISO to each available value in the Canon UI, and then record the value of the corresponding propcases, you will have table.

Why haven't we already done this for you ? Because the relationship between "real" and "market" appears to vary between cameras. I've attached one I made for SD990.

Quote

Are you saying then that shooting with Canon ISO 100 results in different intensity than shooting Nikon ISO 100? Do you see my position?
...
For example, what is the "real" value of Canon ISO 80 for the S90? What is the "real" value of Canon ISO 80 for the SX110? Are they different? What is the "real" value of Canon ISO 1600? Are the "real" values linear, ie +1 f-stop is twice the previous value, whatever it is?

Your mission, should you choose to accept it...

I would expect that the "real" values are such that the APEX equation works. However, I haven't made any effort to rigorously verify it.

edit:
updated csv to correct mislabeled columns, add *96 to ISO conversions

sd990iso.csv (0.51 kB - downloaded 45 times.)

« Last Edit: 02 / September / 2012, 19:13:57 by reyalp »

Logged

Don't forget what the H stands for.

reyalp

14079

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #61 on: 02 / September / 2012, 15:31:34 »

Quote from: SticK on 02 / September / 2012, 14:59:37

Re: the "corrupted DNG" I just sent you ... RAWTherapee 4 chokes on it, but, DCRAW -4 -T does appear to convert the file correctly. I tried two files, so this could be a RAWTherapee problem, not yours, but please check.

Adobe dng_validate reports no errors. irfanview (which generally doesn't have trouble with CHDK DNGs) shows an image with a black stripe at the top and the rest is all white (not sure if this is expected ?). Picasa shows similar results.

using dng_validate to convert to tiff gives an all white tiff.

Logged

Don't forget what the H stands for.

SticK

779

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #62 on: 02 / September / 2012, 16:08:11 »

Thanks for checking the DNG ... agreed // I will send off a bug report to RAWTherapee.

Quote "Your mission, should you choose to accept it..."

I don't want my camera to self-destruct, but I will take the risk so this effort had better be NIST-calibrated....

SX110

Market Real
80 131
100 164
200 329
400 658
800 1316
1600 2633

k=1.646

Real Market
100 61 (camera can't go that low)
131 80 SX110 low end
200 122
400 244
800 488
1600 976
2634 1600 SX110 high end
3200 1952 (S90 high end - untested yet)

Therefore if I want to shoot a NIST-calibrated ISO value of 200, I say {shoot -sv=122}, right? I expect then if I loaded Agfa acetate ISO 200 into my Canon A-1 in theory it would have the same exposure as the digital shot. Is this right?

Now if I am on the right track, is there an equivalent of a switch in shoot that would multiply the constant on my behalf?

Logged

reyalp

14079

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #63 on: 02 / September / 2012, 16:23:09 »

Quote from: SticK on 02 / September / 2012, 16:08:11

I don't want my camera to self-destruct, but I will take the risk so this effort had better be NIST-calibrated....

SX110
Market Real
80 131
..
Real Market
100 61 (camera can't go that low)
...

I think you have your "market" and "real" labels reversed in the second list, and something wrong with your numbers in the first. What appears on the Canon display is the "market" value (i.e. the marketing department dictates we must have ISO x), and is in my experience generally higher than the "real" value (but I guess sx110 could be different here...).

Quote

Therefore if I want to shoot a NIST-calibrated ISO value of 200, I say {shoot -sv=122}, right? I expect then if I loaded Agfa acetate ISO 200 into my Canon A-1 in theory it would have the same exposure as the digital shot. Is this right?

It should be true that
TV + AV = BV + SV
but, quoting from the APEX doc I posted earlier

Quote

These brightness and illuminance value tables are based on the nominal values of the respective meter calibration constants (K and C) suggested for use when the APEX system was first defined. The actual tables would depend on a particular manufacturer’s choice of those constants, which reflects that manufacturer’s opinion of the “proper exposure” for a given scene brightness or illuminance with a given ISO sensitivity of the film or digital imager.

My emphasis.

Quote

Now if I am on the right track, is there an equivalent of a switch in shoot that would multiply the constant on my behalf?

No. Perhaps it could be added.

Logged

Don't forget what the H stands for.

SticK

779

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #64 on: 02 / September / 2012, 16:48:43 »

Quote "something wrong with your numbers in the first."

The values in the first table are purely empirical, so they neither wrong nor right. I said {shoot col_1} and EXIF reported col_2.

Quote "I think you have your "market" and "real" labels reversed in the second list"

Nope .... I idea was to find out what to say to shoot so that the camera exposes at the "real" ISO value.

Quote "No. Perhaps it could be added."

That would be very interesting: if a known standard luminance can be presented to a camera, then the exact k could easily be derived. Therefore, if you and I obtain the same NIST-traceable standard, your camera's k would be different from mine. Hence we then could independently shoot the same scene by setting exactly the same parameters and we would both have the same exposure, NIST calibrated.

However, unless Canon has hidden the NIST fudge factor somewhere in its code and you can access it, I have a feeling that if I generate the same Table 1 for the S90, I will get *exactly* the same results. I would be *very* surprised if I see any difference.

Logged

reyalp

14079

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #65 on: 02 / September / 2012, 17:18:58 »

Quote from: SticK on 02 / September / 2012, 16:48:43

Quote "something wrong with your numbers in the first."

The values in the first table are purely empirical, so they neither wrong nor right. I said {shoot col_1} and EXIF reported col_2.

Terminology matters, the "real" and "market" terms come from the Canon firmware. If you reverse them or make up your own terminology, then we will have a failure to communicate.

Also, the exif is not necessarily reliable (and multiple variants of the same value can be stored) so you shouldn't be relying on it unless you have already empirically determined that it's reliable on your particular camera with the settings you are using.

As I said before, a correct way to determine the relationship between "real" and "market" values is to use the propcases. Here's one way you could do this:
1) Find the "real" and "market" ISO propcases IDs for your camera. To do this, you need the propset number and the corresponding propsetN.h include file from http://trac.assembla.com/chdk/browser/trunk/include S90 is propset 3, sx110 is propset 2
2) For each ISO value available in the Canon UI, set the ISO using the Canon UI, with no CHDK overrides in effect. This is your "market" ISO, in standard ISO units.
3) Half press the shutter to update all the ISO related propcases
4) Using the propcase debug display, record the values of PROPCASE_SV_MARKET and PROPCASE_SV. These are your market and real ISO values in APEX*96 units.
5) When you are done, you can convert the APEX*96 values to traditional ISO values.

edit:
The propcase debug display is explained here http://chdk.wikia.com/wiki/CHDK_User_Manual#Debug_Parameters

« Last Edit: 02 / September / 2012, 17:23:48 by reyalp »

Logged

Don't forget what the H stands for.

SticK

779

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #66 on: 02 / September / 2012, 17:47:16 »

S90 ISO TABLE

Here it is // same as SX110 // I've labeled the columns in a more generic fashion. All this assumes of course that the "real" ISO values are those in the EXIF table from a CHDK image file.

Table 1 - Empirical values

At -sv= EXIF ISO
80 131
100 164
200 329
400 658
800 1316
1600 2633 these two have the nearly same brightness (makes sense)
1944 3200 (calculated // not empirical - ie you can't give the S90 -sv= a value higher than 1944)
3200 5266 these two have the nearly same brightness (makes sense)

"makes sense" because the S90 tops out at 3200 EXIF ISO.

k=1.646 (same as SX110)

Table 2 - What to set -sv in order to get the desired "real" ISO

for EXIF ISO: set sv= to:
80 47
100 61
200 122
400 244
800 488
1600 976
3200 1944 (S90 limit)

Same as SX110 except the wider range.

QED.

« Last Edit: 02 / September / 2012, 17:50:06 by SticK »

Logged

SticK

779

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #67 on: 02 / September / 2012, 18:23:42 »

Quote "S90 is propset 3, sx110 is propset 2"

The three related SV values are almost the same for both cameras, 2 units difference. Is that why the tables between the cameras are the same? Are these the "reference values?" If so, are they compile-time defaults, or not? ie Do I have to go through your procedure to set each ISO value ??

For example, as-is, according to my S90 table, if say shoot -sv=122, will I get a "real" ISO exposure of 200? ... that is, a standardized ISO?

For this I am bit at a loss still.

The S90 has digic 4 // I don't know what impact that would have.

Logged

reyalp

14079

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #68 on: 02 / September / 2012, 18:34:34 »

Quote from: SticK on 02 / September / 2012, 18:23:42

The three related SV values are almost the same for both cameras, 2 units difference.

These values are just identifiers, they tell you the "property case" number that contains the value. They are not related to the value itself. See http://chdk.wikia.com/wiki/PropertyCase

Quote

For example, as-is, according to my S90 table, if say shoot -sv=122, will I get a "real" ISO exposure of 200? ... that is, a standardized ISO?

No, the shoot command works in "real" values. According to your table above, real 122 corresponds to market 200. This is typical, as I said, the market value is usually higher than the "real" value. So to shoot what would be say ISO 80 on the display, you'd want something like -sv=50

Quote

The S90 has digic 4 // I don't know what impact that would have.

I don't think this matters. The propcase numbers change by propset, which doesn't always correspond to digic versions. Different cameras with the same digic use widely different sensors.

edit:
Here's the data from the spreadsheet I posted earlier, with a couple corrections and additional columns

Code: [Select]

ui	sv96	iso_r		apex	sv96_m	iso_m	apex_m	sv96_m – sv96	apex_m – apex	iso_m/iso_r
1600	795	972.1978879844	8.28125	864	1600	9	69		0.71875		1.6457554782
800	699	486.0989439922	7.28125	768	800	8	69		0.71875		1.6457554782
400	603	243.0494719961	6.28125	672	400	7	69		0.71875		1.6457554782
200	507	121.524735998	5.28125	576	200	6	69		0.71875		1.6457554782
100	411	60.762367999	4.28125	480	100	5	69		0.71875		1.6457554782
80	370	45.1928569196	3.85416	449	79.9452	4.677	79		0.82291		1.7701912526

UI = value entered in Canon UI
sv96 = observed PROPCASE_SV value
iso_r = 3.125*2^(sv96/96)
APEX = sv96/96
sv96_m = observed PROPCASE_SV_MARKET value
iso_m = 3.125*2^(sv96_m/96)
apex_m = sv96_m/96

From this we can say
1) ISO / APEX calculation is correct (iso_m derived from market APEX96 = UI value)
2) for most values, "market" ISO is 1.6... times the "real" value. (roughly 2/3 of a stop)
3) ISO 80 is a special case.

« Last Edit: 02 / September / 2012, 19:11:31 by reyalp »

Logged

Don't forget what the H stands for.

SticK

779

Re: CHDKPTP - PC Remote Control Performance Analysis

« Reply #69 on: 02 / September / 2012, 20:29:24 »

Thank you sincerely for your huge effort to help me understand. I will read your post several times and look at your data carefully.

For now, to see if I am on the right road, I discovered after my 1st read it could be a matter of interpreting correctly the empirical data from my TABLE 1.

This is my original empirical TABLE 1 with original labeling

At -sv= EXIF ISO
80 131
100 164
200 329
400 658
800 1316
1600 2633 these two have the nearly same brightness (makes sense)
1944 3200 (calculated // not empirical - ie you can't give the S90 -sv= a value higher than 1944)
3200 5266 these two have the nearly same brightness (makes sense)

This is the relabeled TABLE 1

In words: For me to shoot a scene at a desired calibrated ISO value I would say {shoot -sv=calibrated_sv}, which then CHDK converts for Canon to shoot at its own internal value that forces the shot to comply with the calibrated standard I gave to -sv=.

desired
calibrated_sv Internal ISO that is used by Canon firmware (the value that appears in EXIF)
47 80 (the lowest VGA gain possible in an S90)
80 131
100 164
200 329
400 658
800 1316
1600 2633
1944 3200 (the highest VGA gain possible in the S90)

Therefore if my interpretation is correct, the S90 has an "real" ISO range of 47 to 1944, according to the calibrated standard. Likewise, I am allowed an -sv= parameter range of 47 to 1944 without exceeding the gain bounds of the VGA circuit. Correct?

« Last Edit: 02 / September / 2012, 21:09:13 by SticK »

Logged