Log versus non-log in 10 or 12 bit

[BitFriend]

I haven't seen any comments or analysis on this topic - I'd be grateful for any insights.  Some cameras (which have standard internal recording in 8 bit) allow the user to choose between a standard and log-gamma curve when recording in 10 (or 12) bit.  I have read reams of tests and results (particularly concerning DR) when recording in log, but nothing at all regarding results when recording non-log.  For example, The Z6 can record 10 bit N-log externally (and very fine it looks too), but users can also choose non-log 10 bit (external) recording.

I can appreciate why reviewers and bloggers focus on log (given the desire to maximise dynamic range) but I'm perplexed that it seems nigh on impossible to find any tests/analysis for 10 bit non-log (ditto for the Panasonic S1 post firmware).  People reading this might be perplexed that I am perplexed...after all, isn't the whole point to maximise dynamic range?  That's certainly a major factor, but workflow and turnaround in post can also be major factors with some projects.

Shooting in 10 bit non-log versus 8 bit non-log produces four times as many gradations (2^2) in encoding the colour data.  So, where there are the issues of crushing dark greys to black, 10 versus 8 bit at least provides for four times as much data and thus much better potential for lifting shadows to taste.  Non-log 10 bit won't help highlights roll off smoothly per se, but that extra data care of 2^2 (four) times as much quanta would presumably allow the user to under expose somewhat ('expose to the left') such that those highlights one is comfortable with being on the verge of blowing are at 100% on zebras, then raise exposure of all but highlights to taste in post.

The workload in post would still be somewhat involved, but not as much as needing to tweak via all manner of inverse-mapping gamma curves (which bring higher orders of uncertainty).  I'd stress that my question is not 'why are people shooting in log rather than non-log when in 10 bit?' but that, while I appreciate that log makes more sense most of the time, I can find no coverage or opinions on non-log (in 10 bit), which might make sense some of the time (also in terms of being able to shoot at base ISO and not need high order ND filter).

Please do say if I am misunderstanding something here!

[androidlad]

Regardless of the output bit depth, they are all derived from linear 12bit readout, there’s no discernible DR difference between 10bit and 8bit on the same camera with the same picture profile, the benefit is flexibility in grading (4 times more bits per stop).

[kye]

I know what you're saying @leslie and your questions are valid.

The way I see it is that it's about how many bits are allocated per stop across the full dynamic range of the camera, and how much of the dynamic range of the camera is captured and how much of it is clipped.

Mostly this isn't discussed because in most cameras the non-log formats the full dynamic range of the sensor isn't output, so you clip highlights (and maybe crush shadows).  People generally care about getting the most DR so shoot the profile with the most DR, which used to always be the log profile.

There's an interesting thing with the Canon XC10 though - it shoots 8-bit and has various profiles including C-Log / WideDR / etc.  The interesting thing about the XC10 is that it keeps the full DR of the camera even in the 'normal' profile.  What is different is the distribution of bits across each stop of DR.  In the normal mode the number of bits in the highlights and the shadows is a lot less than the bits dedicated to the mid-tones, which are typically where we put sensitive things like skin-tones.  

This makes it interesting because you're not expanding out the mids in things like skin tones as much as you would if you have a log profile, and considering that both log and normal modes were restricted to 8-bit that makes a difference.  One of the reasons people generally avoid 8-bit log profiles is that by the time you expand out the mid-tones of the image (by adding contrast) you are exposing weaknesses of the codec in the most sensitive parts of the image (skin-tones).

To my mind the goal is to get enough DR and to get enough bits in each stop of DR so that your image is robust and can be graded easily to give a desirable result.  I really noticed this moving from the 8-bit C-Log on the XC10 to the 10-bit HLG on the GH5, the 10-bit files are so much more robust in post.

In terms of what matters for you, I'd say that you'd have to just try out the modes and compare them.  Cameras are all different and the proof is in the pudding, so they say.

[leslie]

On 8/16/2019 at 11:40 AM, kye said:

I know what you're saying @leslie and your questions are valid.

The way I see it is that it's about how many bits are allocated per stop across the full dynamic range of the camera, and how much of the dynamic range of the camera is captured and how much of it is clipped.

Mostly this isn't discussed because in most cameras the non-log formats the full dynamic range of the sensor isn't output, so you clip highlights (and maybe crush shadows).  People generally care about getting the most DR so shoot the profile with the most DR, which used to always be the log profile.

There's an interesting thing with the Canon XC10 though - it shoots 8-bit and has various profiles including C-Log / WideDR / etc.  The interesting thing about the XC10 is that it keeps the full DR of the camera even in the 'normal' profile.  What is different is the distribution of bits across each stop of DR.  In the normal mode the number of bits in the highlights and the shadows is a lot less than the bits dedicated to the mid-tones, which are typically where we put sensitive things like skin-tones.  

This makes it interesting because you're not expanding out the mids in things like skin tones as much as you would if you have a log profile, and considering that both log and normal modes were restricted to 8-bit that makes a difference.  One of the reasons people generally avoid 8-bit log profiles is that by the time you expand out the mid-tones of the image (by adding contrast) you are exposing weaknesses of the codec in the most sensitive parts of the image (skin-tones).

To my mind the goal is to get enough DR and to get enough bits in each stop of DR so that your image is robust and can be graded easily to give a desirable result.  I really noticed this moving from the 8-bit C-Log on the XC10 to the 10-bit HLG on the GH5, the 10-bit files are so much more robust in post.

In terms of what matters for you, I'd say that you'd have to just try out the modes and compare them.  Cameras are all different and the proof is in the pudding, so they say.

is there another Leslie ? or have you been partaking from the dregs of btm_ alcohol fueled lens tests again ?

because i'm confused. i'm pretty confident  i couldnt really do a 8,10,12 log discussion much justice at all. although i am pretty happy with what the p4k sends my way ?

[kye]

1 hour ago, leslie said:

is there another Leslie ? or have you been partaking from the dregs of btm_ alcohol fueled lens tests again ?

because i'm confused. i'm pretty confident  i couldnt really do a 8,10,12 log discussion much justice at all. although i am pretty happy with what the p4k sends my way ?

well, not sure why I tagged you instead of the OP.  anyway, enjoy your P4K

[Geoff CB]

On 8/15/2019 at 9:33 AM, BitFriend said:

I haven't seen any comments or analysis on this topic - I'd be grateful for any insights.  Some cameras (which have standard internal recording in 8 bit) allow the user to choose between a standard and log-gamma curve when recording in 10 (or 12) bit.  I have read reams of tests and results (particularly concerning DR) when recording in log, but nothing at all regarding results when recording non-log.  For example, The Z6 can record 10 bit N-log externally (and very fine it looks too), but users can also choose non-log 10 bit (external) recording.

I can appreciate why reviewers and bloggers focus on log (given the desire to maximise dynamic range) but I'm perplexed that it seems nigh on impossible to find any tests/analysis for 10 bit non-log (ditto for the Panasonic S1 post firmware).  People reading this might be perplexed that I am perplexed...after all, isn't the whole point to maximise dynamic range?  That's certainly a major factor, but workflow and turnaround in post can also be major factors with some projects.

Shooting in 10 bit non-log versus 8 bit non-log produces four times as many gradations (2^2) in encoding the colour data.  So, where there are the issues of crushing dark greys to black, 10 versus 8 bit at least provides for four times as much data and thus much better potential for lifting shadows to taste.  Non-log 10 bit won't help highlights roll off smoothly per se, but that extra data care of 2^2 (four) times as much quanta would presumably allow the user to under expose somewhat ('expose to the left') such that those highlights one is comfortable with being on the verge of blowing are at 100% on zebras, then raise exposure of all but highlights to taste in post.

The workload in post would still be somewhat involved, but not as much as needing to tweak via all manner of inverse-mapping gamma curves (which bring higher orders of uncertainty).  I'd stress that my question is not 'why are people shooting in log rather than non-log when in 10 bit?' but that, while I appreciate that log makes more sense most of the time, I can find no coverage or opinions on non-log (in 10 bit), which might make sense some of the time (also in terms of being able to shoot at base ISO and not need high order ND filter).

Please do say if I am misunderstanding something here!

It's something I need to test on my Z6. I've fallen in love with the non-log output from it and would also like to see the results of focusing on increased color information instead of increasing DR.