Auto ETTR match camera settings?

[chadat23]

Does auto ettr limit itself only to settings that you can input through the camera's UI, or will it also use in between values and then just set the camera's display and metadata to the nearest value, say, taking a 1/22 second exposure while recording it as 1/20?

I've written yet another deflicker tool. It reads the iso, fstop, and shutter speed info out of the dng files to calculate relative brightnesses in order to set the exposure in Lightroom. It generally works well when I set my camera to manual and make adjustments as needed, but I get a lot of flickering when I use auto ettr. Thanks!

[Walter Schulz]

First: You are aware shutter speed settings shown in Canon menu are - most of the time - not true but matter of convention?
You can test it by setting a 30 seconds shutter speed -> Cam shutter will be open for 32 (thirty-two) seconds.
See http://www.scantips.com/lights/fstop2.html
Take special attention to the fact 1/20 actually differs when set cam to 1/2 stop or 1/3 stop.


And LR process engine 2012 (default for LR) may cause additional trouble because it will do some kind of autocorrection which may add to flicker. Cross check with other RAW converters. May I suggest to run RAWtherapee and look what you get there.

[a1ex]

The internal representation of exposure values in Canon firmware is in 1/8 EV increments.

Some shutter speed measurements from movie mode: https://www.magiclantern.fm/forum/index.php?topic=16814.msg164062#msg164062

However, timings in photo mode are very likely to be different.

On most models, setting shutter speed in 1/8 EV increments from software should give either exact value or a difference of +/- 1/8 EV (api_test.lua checks for that). However, this is only what Canon firmware reports back.

How this translates to mechanical shutter timings is unknown. This table contains some sensor timings, but sensor exposure is a little longer than mechanical shutter opening.

If you have a light source that does not flicker and does not change in intensity, you can try to take pictures of a blank wall at various shutter settings, and record the median (not average) value (e.g. with deflick.mo) so we can cross-check the table referenced above.

BTW, ISO is likely to change (slightly) the sensor response (both gain and linearity). Changing aperture does not always give an ideal change in brightness (lens-dependent) and will also affect the vignetting (also lens-dependent and affected by focus distance). All these are going to affect an EXIF-based deflicker algorithm.

[chadat23]

Accounting for nominal vs. actual shutter speeds did previously make things a bit better but it's still distinctly flickery.

I took some pictures with a constant fstop of 4 and a constant iso of 100. From playing around I knew that ettr would settle on 5s shutter, f/4, & iso 100. All of the shutter speeds were manually entered other than the "ettr 5"s which ettr calculated. Here's what I got. I'm now wishing I did more replicates for more statistical significance but regardless, there's a definite pattern and it looks like auto ettr is settling on values other than what's showing up as the nominal values. It looks like ettr is controlling the shutter or maybe some combination of knobs to the nearest 1/6 stop but again, I didn't do enough replicates with tight enough experimental control to be confident in that. Looks like unless I find some buried dng metadata or come up with something really clever, my algorithms won't play nice with auto ettr.

And sorry about the bad formatting.

Shutter    crs:Exposure2012                Theoretical EV from 5s        Actual EV from mean of                    Dev between theoretical
                from deflicker.mo                  based on Walter's link        manually entered 5s (-1.898365)      and actual EV deviations
ettr 5       -2.06436                                 0                                        -0.165995                                      -0.165995
6              -2.24175                               -0.34                                   -0.343385                                      -0.003385
5              -1.89454                                0                                          0.003825                                       0.003825
ettr 5       -2.06394                                0                                         -0.165575                                      -0.165575
4              -1.55574                                0.33                                     0.342625                                       0.012625
5              -1.90219                                0                                         -0.003825                                      -0.003825
ettr 5       -2.06266                                0                                         -0.164295                                      -0.164295

[a1ex]

Looks like ETTR picks a shutter speed rounded to 1/2 EV, but your selection in Canon menu is 1/3 EV. Canon firmware accepts both sets of values, regardless of their menu setting, but looks like it's taking it into account when writing the EXIF.

If you print camera.shutter.raw with a Lua script (see api_test.lua), that value should be a bit more useful for deflicker purposes.

The Tv value from ML menu can also be useful, but it's rounded to one decimal place.