8 Bit Vs 10 Bit Test

By | 31/10/2022

Entropy512 •

Veteran Member

• Posts: 5,865

Re: You can test the limits of eight-bit vs 10-chip log yourself

Horshack wrote:

You tin can utilize Davinci Resolve to load a lossless piece of material into the color page, catechumen it to your log contour of selection, and then do sample grades to see how well it holds up. I own the paid studio version of Resolve but the free version should piece of work fine with 8-flake files – you can download it here.

I recently did an 8-chip vs 10-bit log vs rec.709 and raw comparing here:

https://www.dpreview.com/forums/post/64884289

I’ve been doing something similar by creating ICC profiles that correspond to the various Sony log colorspaces and transfer functions, applying them to shots I’ve taken, and loading them into RawTherapee, and also doing some plots of the spacing between code values.

I’ve generally institute that 8-bit S-Log2 is commonly acceptable unless yous start pushing information technology, Southward-Log3 is about immediately problematic.

For a refresher, Due south-Log2, S-Log3, and HLG transfer functions for viii-bit lawmaking values:



Here’south a plot of the number of code values per EV:



Clearly non all log formulas are created equal.

A major mistake for both of Sony’due south curves is that they don’t use the lowest 22 code values of eight-bit data – that’s nearly 10% of all code values

I accept unfortunately lost the link, but I recently read an experimental paper where they determined the human threshold for perception of quantization in gradients. Going by retention, it was roughly two-three%, with the threshold being a bit higher at depression luminance. I’ll endeavour once more to find that link this evening…)

2% corresponds to 1.0/log2(1.02) = 35 lawmaking values per EV

three% corresponds to 1.0/log2(1.03) = 23 code values per EV

S-Log2 winds up midway between these, S-Log3 doesn’t even come close. HLG is the surprise hither, achieving more lawmaking values per EV across the lath AND representing a fairly wide range of dynamic range. Not throwing out ten% of your code values helps here.

You tin also see the “bereft lawmaking values” manifesting when loading up images in RT by the fact that the histogram starts looking really wonky for South-Log3 nearly immediately.

The disadvantage of HLG being that if you happen to have older Sonys that don’t practise HLG in the mix, matching HLG to South-Log2 might exist more of a PITA than South-Log2 to S-Log2, but that’southward a niche corner instance.

Non sure where 5-Log falls within this.

— hide signature —

Context is primal. If I have quoted someone else’south mail when replying, delight do not reply to something I say without reading text that I have quoted, and agreement the reason the quote function exists.



Sony a6000



Pentax K-5



Pentax K-01



Sony a6300



Canon EF 85mm F1.8 USM

+5 more

Source: https://www.dpreview.com/forums/post/64976691