ETTR and ISO values Comparison
Posted: Fri Sep 27, 2013 4:04 pmMany people recommend exposing to the right in raw mode on the BMCC.
Sometimes when doing this, one finds the screen on the camera to be too washed out to be useful for composing your shot - in other words, the scene is exposed so far to the right without clipping, that the screen is almost just solid white.
A solution to this is to lower the ISO value, as it will make the on screen image darker, but because its just meta data, the camera is still recording at ISO 800 - so no data is lost.
A few people suggested that doing this would in fact change the final out come of the image, and that lower ISO values also change the way the colour or luma data is represented - and so might not be a good process after all.
I decided to try and find out for myself if this was apparent.
I while ago I shared this image:
http://tommajerski.com/publicimages/iso_example_1.jpg
Which I claimed showed no real difference at all - but it was suggested that once I applied a LUT or grade to the image, the difference would be more apparent.
So today I created this:
(12MB jpeg - better off downloading and then you can zoom around the image better)
http://tommajerski.com/publicimages/ISO-comparison.jpg
This image shows the same shot, filmed at all the various ISO values, with the EXACT same grade applied to each one, (the only change was in the Raw tab on Resolve, where I balanced the exposure value).
In the shot, I ETTR by 1 stop.
The grade consisted of 3 serial nodes:
Node 1: slight shadow boost, slight highlight reduction
Node 2: Alexa REC 709 LUT
Node 3: Film simulation LUT
The image shows the entire frame of the video in its full 2400x1350 resolution.
From this, I cannot see any difference in the images, and unless proven otherwise - I conclude that it is incorrect to say that lowering the ISO value will change the outcome of the colour and luma or SNR when ETTR.
The only exception to this is ISO 1600 -where it has been confirmed that in this situation, more bits are allocated to the shadow areas of the curve that's encoded (essentially a slightly different LOG curve) - but this is not visible in my test due to it being a relatively bright scene. (Thanks JB for the clarification)
I hope this test is of some use to others.
Sometimes when doing this, one finds the screen on the camera to be too washed out to be useful for composing your shot - in other words, the scene is exposed so far to the right without clipping, that the screen is almost just solid white.
A solution to this is to lower the ISO value, as it will make the on screen image darker, but because its just meta data, the camera is still recording at ISO 800 - so no data is lost.
A few people suggested that doing this would in fact change the final out come of the image, and that lower ISO values also change the way the colour or luma data is represented - and so might not be a good process after all.
I decided to try and find out for myself if this was apparent.
I while ago I shared this image:
http://tommajerski.com/publicimages/iso_example_1.jpg
Which I claimed showed no real difference at all - but it was suggested that once I applied a LUT or grade to the image, the difference would be more apparent.
So today I created this:
(12MB jpeg - better off downloading and then you can zoom around the image better)
http://tommajerski.com/publicimages/ISO-comparison.jpg
This image shows the same shot, filmed at all the various ISO values, with the EXACT same grade applied to each one, (the only change was in the Raw tab on Resolve, where I balanced the exposure value).
In the shot, I ETTR by 1 stop.
The grade consisted of 3 serial nodes:
Node 1: slight shadow boost, slight highlight reduction
Node 2: Alexa REC 709 LUT
Node 3: Film simulation LUT
The image shows the entire frame of the video in its full 2400x1350 resolution.
From this, I cannot see any difference in the images, and unless proven otherwise - I conclude that it is incorrect to say that lowering the ISO value will change the outcome of the colour and luma or SNR when ETTR.
The only exception to this is ISO 1600 -where it has been confirmed that in this situation, more bits are allocated to the shadow areas of the curve that's encoded (essentially a slightly different LOG curve) - but this is not visible in my test due to it being a relatively bright scene. (Thanks JB for the clarification)
I hope this test is of some use to others.
