Adjust RAW Exposure Measurement Area

[RenatoPhoto]

The raw exposure calculations done by ETTR give erroneous numbers when using a Fisheye lens.  I have a 5D3 and an 8 mm fisheye which produce a circle image and a large black area around the image which ETTR is analyzing as if the black border was part of the image.  As a result the  Midtone SNR and the Shadow SNR will boost the exposure of the image above the required and correct levels.



Analyzing the image at the Canon crop of 3:2 5760x3840 we can inscribe a 3:2 rectangle inside the circle of 3170x2112

The crop factor would be 5760/3170 = 1.82

The same situation occurs with the fisheye and the 7D



Analyzing the image at the Canon crop of 3:2 5184x3456 we can inscribe a 3:2 rectangle inside the circle of 4506x3004

The crop factor would be 5184/4506 = 1.15

Also for my EFs 10-22 Canon on 5D3



Analyzing the image at the Canon crop of 3:2 5760x3840 we can inscribe a 3:2 rectangle inside the circle of 4062x

The crop factor would be 5760/4062 = 1.41

And yet another Tamron 17-50 on my 5D3



The crop factor would be 5760/3920 = 1.46

I assume that there will be other lenses so the option would be to have a user selectable crop factor or just pick some that could benefit all like:
1.2, 1.5, 2, 3

I put the 3 so that we could also measure the correct exposure for 5X cropped mode (equivalent to 3X crop factor).

[Audionut]

Also for my EFs 10-22 Canon on 5D3
And yet another Tamron 17-50 on my 5D3

Are you cropping these in post?
Why the APS-C lenses on a FF?

[RenatoPhoto]

Are you cropping these in post?

Of course.

Why the APS-C lenses on a FF?

Why not?  Not necessary to duplicate lenses, and I don't have money to waste.

[a1ex]

You even get a wider FOV if you mount the APS-C Samyang 8mm on FF.

http://www.360pano.de/en/samyang_5D_1D.html

[Audionut]

Of course.

Try using the zoom on the lens instead.  Unless you're going for extreme aspect ratio in the center of the lens, the end result will be the same.

[RenatoPhoto]

Try using the zoom on the lens instead.  Unless you're going for extreme aspect ratio in the center of the lens, the end result will be the same.

Sometimes I use these in timelapses and want the full FOV so I can pan into the extra pixels area.

[Greg]

You even get a wider FOV if you mount the APS-C Samyang 8mm on FF.

http://www.360pano.de/en/samyang_5D_1D.html

This lens is about 9.5mm 

[RenatoPhoto]

Hi Alex,

I would like to spend some time trying to learn more about ML and this change is something that I am interested in digging a little further.

I was looking in ETTR.c and found some stuff but I am not sure if I am in the right track.

From the variable names it would seem that the exposure value comes from raw_to_ev but this seem to be a function that I could not even find the definition...

Then I found in raw.c some other stuff where you set width, height, skip_left, skip_right, skip_top which are pixels.  Here I am wondering if this could be used to redefine this borders so that the exposure calculations could be adapted to a specific crop size.

Anyway this is as far as I got and was wondering if you had any other suggested areas where I could explore the code to achieve these change.

[a1ex]

raw_to_ev is in raw.c (transforms from linear raw values to log EV values). Scalar.

Width and height are read-only (image buffer size - what Canon gives us). Skip values are optical black areas (not part of the image).

What you need here is a histogram computed from a cropped area (histogram.c), since ETTR does all the work from histogram analysis. Probably the most straightforward is to add a crop window (x,y,w,h) to raw_hist_get_percentile_levels.

[RenatoPhoto]

What you need here is a histogram computed from a cropped area (histogram.c), since ETTR does all the work from histogram analysis. Probably the most straightforward is to add a crop window (x,y,w,h) to raw_hist_get_percentile_levels.

Found it!  Ouuuchh much more complicated than I imagined! 
Thanks for the information provided and I am still far away from understanding this code.