10-Bit vs 8-bit Grading -- The Musical

Musicvid wrote on 6/9/2018, 1:29 PM

Preliminary, and will be repeated door to door to trap errors

No questions yet, please, I've yet to validate my workflow.

Virtually all salient questions will be answered in the docs.

Source is 8 bit 4:4:4 TIFF and marked for re-use.

Full size composite is 4K 2160x4320 8 bit 4:4:4 I will make that available because it's a pretty good reference for screen matching, too.

From top clockwise RYGCBM 5% nonlinear grade. Center is Reference keystone. (identical).

Comments are invited on the rough cut, but I won't tell you which is which until I've run the SSIM comparisons, which keeps this a truly double-blind survey. Will use distance formula first, an RMS Regression "may" come later, but I would need to retrain myself.

Comments

Musicvid wrote on 6/10/2018, 12:11 PM

The first comparison of 8 bit grading, vs. 8-10-8 as described in the literature gave results that will seem surprising to some, yet familiar to others.

Source is 8 bit 4:4:4 TIFF. Grading, applied in both 8 and 32 bit environs, is a 5% nonlinear, single channel, gamma-weighted curve.

Results of capturing a PNG image from the Compositing Difference preview indicate a 1-bit shift in the blacks, which is reminiscent of a metadata line in the original AVI spec, and which remains to this day. The absence of this disparity in high-bit projects is understandable, since it is an illogical outcome of zero source data in a float environment.

Whether an unaided observer can see the bit shift has been a subject of little debate over the past quarter-century, as nearly as I can determine.

Other than the observable 1-bit levels shift, which is a mathematical dead heat between the two images, if for different reasons, neither my unpredictable eyes, nor Vegas' Scopes, are able to reveal any detectable deviation, meaning the 10-bit environment expanded 255 bits to 256 bits, undetectably. Like slowing a PAL Movie DVD down to native 24p as a rough example.

This, being the only measurable difference between the 8-bit and 10-bit graded versions, suggests an early fail in this inquiry, and thus a relaxation, but not abandonment of my agreement to test this. Tempest in a teacup.

This is all not about seeing all these colours.

This is mostly about preventing unnecessary degradation during video processing.

As far as bringing effects back into 8 bit range, I keep the levels filter handy to either do this on the fly, but only in a couple of cases where I found myself dancing on the edge. I don't recall having ever needing to recall effects occuring natively outside my delivery range since my effects/colors track always has a track Studio RGB output. Why invest so much time in switching project properties when it is almost always a brute levels thing, and nothing else, including routine grading? I've asked this a couple of times this week, and gotten zero response.

Wwjd is our resident expert on baking a video and stuffing it all back into the turkey. Perhaps a couple of you Picassos could put your heads together and teach us some refined techniques through demonstration rather than direction, understanding that not everyone's head is able to work in the same creative space, as some creative heads are unable to work in a structured space.