10-Bit vs 8-Bit Signal Path & Encoding Test Chart

This folder contains make_bitdepth_chart.py and 16-bit RGB / 10-bit video test patterns designed to visually tell whether an encoding pipeline, video player, GPU compositor, or display signal path is true 10-bit or quantized to 8-bit.

Unlike traditional smooth gradient ramps—which require squinting for faint Mach bands and are easily fooled by spatial/temporal dithering—this chart uses three gamma-independent optical mechanisms.


How the Three Sections Work

Between any two consecutive 8-bit integers $K$ and $K+1$ (which correspond to $4K$ and $4K+4$ in 10-bit), there are four 1-LSB 10-bit quarter-steps:

\[q = 0 \;(K.00), \quad q = 1 \;(K.25), \quad q = 2 \;(K.50), \quad q = 3 \;(K.75), \quad q = 4 \;(K + 1.00)\]

1. Top Section — Vanishing 8-BIT Watermark

  • Background: A 50/50 horizontal-line spatial dither between $q = 0$ (exact 8-bit $K$) and $q = 4$ (exact 8-bit $K+1$).
    • Because $K$ and $K+1$ are already exact 8-bit integers, an 8-bit bottleneck passes the dither background through 100% untouched, and your eye averages it to the midpoint $K + 0.5$ ($q = 2$).
    • Over a tiny $1/255$ span, every monitor EOTF ($\gamma = 2.2$, piecewise sRGB, $\gamma = 2.4$) is locally linear to within $0.002$ of a step, so this match is completely independent of monitor gamma.
  • Foreground (8-BIT watermark & top bars): Drawn in solid 10-bit midpoint $q = 2$ ($K + 0.5$).
    • True 10-bit path: Solid $q = 2$ matches the optical average of the $(q=0, q=4)$ dither, so 8-BIT and the bars vanish into a flat field.
    • 8-bit path: Solid $q = 2$ gets quantized to $q = 4$ (nearest rounding) or $q = 0$ (truncation), while the dither background stays at $q = 2$ — so 8-BIT pops out clearly.

2. Middle Section — Four 10-Bit Sub-Step Phase Gratings (A, B, C, D)

Each column contains four stacked rows of 16-pixel-wide vertical bars (aligned to $16 \times 16$ codec macroblocks/CTUs):

  • Row A: alternates $q = 0$ and $q = 1$ ($\Delta = 1$ LSB in 10-bit)
  • Row B: alternates $q = 1$ and $q = 2$ ($\Delta = 1$ LSB in 10-bit; straddles the 8-bit nearest-rounding threshold)
  • Row C: alternates $q = 2$ and $q = 3$ ($\Delta = 1$ LSB in 10-bit)
  • Row D: alternates $q = 3$ and $q = 4$ ($\Delta = 1$ LSB in 10-bit; straddles the 8-bit truncation threshold)

What you see:

  • True 10-bit path: All four rows (A, B, C, D) show the exact same faint 1-LSB bar pattern.
  • 8-bit nearest-rounded path: Rows A, C, and D collapse to completely flat zero, while Row B jumps 4× in contrast (a full 8-bit step)!
  • 8-bit truncated path: Rows A, B, and C collapse to completely flat zero, while Row D jumps 4× in contrast!

3. Bottom Section — 10-Bit (16-Step) vs 8-Bit (4-Step) Split Staircase

  • Left half of each column: Steps by $1/1023$ every row (16 quarter-steps spanning 4 8-bit values).
  • Right half of each column: Steps by $1/255$ every 4 rows (4 coarse 8-bit steps).
  • True 10-bit path: The left side steps $4\times$ finer than the right side, with a clear sawtooth step difference along the vertical center seam.
  • 8-bit path (no dither): Every group of 4 steps on the left collapses into the exact same 8-bit value as the right side, erasing the center seam so both halves merge into identical 4-step staircases.
  • 8-bit + spatial dither: Zoomed in, the .25 / .50 / .75 steps on the left show spatial dither grain, whereas on a true 10-bit path every step is a noiseless uniform plateau.

Limited-Range (tv) vs Full-Range (pc) Columns

  • Columns 1–3 (TV Shadow, TV Low-Mid, TV Midtone) are calibrated for Limited-Range (tv: 10-bit 64..940 / 8-bit 16..235) Y’CbCr video encodes ($876 = 4 \times 219$).
  • Columns 4–6 (Full Shadow, Full Low-Mid, Full Midtone) are calibrated for Full-Range (pc: 10-bit 0..1023 / 8-bit 0..255) direct 16-bit PNG viewing and full-range RGB pipelines.
Column Mode 16-Bit RGB Codes [q=0, 1, 2, 3, 4] 10-Bit Codes 8-Bit Nearest Round 8-Bit Truncated
1. TV Shadow tv [2394, 2469, 2548, 2618, 2693] [96, 97, 98, 99, 100] [24, 24, 25, 25, 25] [24, 24, 24, 24, 25]
2. TV Low-Mid tv [4788, 4863, 4942, 5012, 5087] [128, 129, 130, 131, 132] [32, 32, 33, 33, 33] [32, 32, 32, 32, 33]
3. TV Midtone tv [9576, 9651, 9730, 9800, 9875] [192, 193, 194, 195, 196] [48, 48, 49, 49, 49] [48, 48, 48, 48, 49]
4. Full Shadow pc [5140, 5204, 5270, 5333, 5397] [80, 81, 82, 83, 84] [20, 20, 21, 21, 21] [20, 20, 20, 20, 21]
5. Full Low-Mid pc [7196, 7260, 7326, 7389, 7453] [112, 113, 114, 115, 116] [28, 28, 29, 29, 29] [28, 28, 28, 28, 29]
6. Full Midtone pc [10280, 10344, 10410, 10473, 10537] [160, 161, 162, 163, 164] [40, 40, 41, 41, 41] [40, 40, 40, 40, 41]

Usage

Generate the 16-bit PNGs and simulated views:

python tests/bitdepth-8vs10/make_bitdepth_chart.py

Also encode the 10-bit and 8-bit test videos (requires ffmpeg via rez-env):

rez-env ffmpeg-8.1 -- python tests/bitdepth-8vs10/make_bitdepth_chart.py --encode

Copyright © 2022 ORI Contributors. Distributed under a CC BY 4.0 license.

This site uses Just the Docs, a documentation theme for Jekyll.