display-report measures displays and produces a standardized, open-source fidelity report. It sends hundreds of known code values to a display, captures the resulting light output with a spectroradiometer, and compares the two to quantify reproduction accuracy.
LED display calibration is a complex, per-pixel process handled by LED processors. display-report does not calibrate — it reports on the results of that vendor calibration. The more accurate a display's native calibration is, the easier it is for downstream tools to build on top of it — creative LUTs, colour space transforms, and camera-match workflows all benefit from a predictable starting point.
Traditional display spec sheets describe capability — peak brightness, gamut coverage, contrast ratio. They do not show whether a display tracks its target EOTF, holds a neutral grey scale, or reproduces colours accurately inside the gamut boundary. The numbers are best-case snapshots, not distributions, and they are not independently verifiable.
display-report's report format is open source. The metrics it presents, the visualizations it uses, the tolerance thresholds it applies, and the way it surfaces information to non-specialist readers are all visible in the codebase. Anyone can review those design choices, propose improvements, or adapt the format for their own context.
The report summarizes accuracy as distributions (mean + 95th percentile), not cherry-picked values. Results are comparable across vendors and display technologies because the report format and analysis are identical for every display measured.
The report is a single-page PDF. Each section uses a traffic-light tolerance scheme: green indicates performance within one JND (just noticeable difference), yellow indicates marginal performance, and red indicates clearly visible error.
The top of the report shows aggregate colour difference metrics:
- Mean dE 2000 and 95th percentile — perceptual colour difference weighted for typical viewing conditions. Good for judging how a human observer would perceive the display.
- Mean dE ITP and 95th percentile — perceptual colour difference in ICtCp space (ITU BT.2124). More sensitive than dE 2000, especially at low luminance. Good for understanding physical error across the full dynamic range.
- Reflectance and glossiness ratio (if supplied) — 45:0 and 45:45 reflectance factors that determine real-world black level and contrast ratio under ambient light.
Shows the measured colour error plotted on the CIE 1976 u'v' chromaticity diagram alongside MacAdam ellipses. The diagram clusters all test patches into 14 regions and draws an arrow from each cluster centre showing the mean error direction and magnitude, magnified 10x (same scale as the ellipses). If an arrow is roughly the same size as the nearby ellipse, the error in that region is approximately 1 standard deviation of colour matching (SDCM).
Three gamut outlines are overlaid for reference: P3-D65 (red dashed), BT.2020 (green dashed), and the display's estimated native gamut (black solid).
Plots the ideal PQ transfer function (red curve) against measured grey-ramp luminance values (blue dots) on log-log axes. The x-axis shows 10-bit code values; the y-axis shows luminance in cd/m² (nits). Two reference lines mark 1000 cd/m² (nits) (teal) and the display's measured maximum luminance (purple). A display tracking PQ correctly will have its dots fall directly on the red curve up to the display maximum, then clip above it.
Two vertically stacked subplots show how the display's white point drifts across luminance levels:
- CCT (top) — correlated colour temperature in Kelvin. The target is D65 (6504 K). Drift above the line means cooler/bluer; below means warmer/yellower.
- Duv (bottom) — green/magenta offset from the Planckian locus (CIE 1960). Positive values shift green; negative values shift magenta.
Both subplots use tolerance bands derived from ANSI C78.377 SDCM values: green is within 1 SDCM of D65, yellow within 4 SDCM, and red beyond 6 SDCM.
Plots per-patch brightness error derived from the intensity channel of ICtCp. Positive values mean the display is brighter than expected; negative values mean darker. Each dot is coloured by its test patch RGB value, making it easy to spot whether specific colours or luminance ranges are affected.
Tolerance bands: green for less than 1 JND, yellow for 1-8 JND, red for greater than 8 JND. The y-axis uses a symmetric log scale.
Plots the combined hue and saturation error from ICtCp, with brightness removed. This isolates colour reproduction error from brightness error. Only the magnitude is shown (no sign), so all values are zero or positive.
Same tolerance scheme as brightness error: green below 1 JND, yellow 1-8, red above 8. Dots are coloured by test patch RGB value.
| Band | Threshold | Meaning |
|---|---|---|
| Green | < 1 JND | Not perceptible under normal viewing |
| Yellow | 1 - 8 JND | Marginal — may be visible in demanding content |
| Red | > 8 JND | Clearly visible distortion |
A JND (just noticeable difference) is the smallest colour or brightness change a typical observer can detect under controlled conditions. In practice, moving or complex imagery raises the detection threshold, so yellow-zone errors are often acceptable.
git clone https://github.com/OpenDisplayEval/display-report.git
cd display-report
uv sync
uv run display-report measure --max-luminance 1500 --bit-depth 10 --save-directory ./measurements
uv run display-report analyze ./measurements/<file>.csmfSee USAGE.md for hardware setup, full CLI reference, and development instructions.
See LICENSE for details.