跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

hdr-test-pattern-generation

RenoDX workflow for generating HDR/SDR test patterns, synthetic charts, ramps, gradients, hue sweeps, color bars, checkerboards, banding panels, gamut stress images, BT.709/BT.2020/AP1 comparisons, and scalar/energy-map validation sources. Use when making test pattern images or datasets for tonemap, gamut, LUT, PsychoV, RenoDRT, HDR, or cICP validation.

clshortfuse · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add clshortfuse/renodx --skill hdr-test-pattern-generation

技能说明


name: hdr-test-pattern-generation description: "RenoDX workflow for generating HDR/SDR test patterns, synthetic charts, ramps, gradients, hue sweeps, color bars, checkerboards, banding panels, gamut stress images, BT.709/BT.2020/AP1 comparisons, and scalar/energy-map validation sources. Use when making test pattern images or datasets for tonemap, gamut, LUT, PsychoV, RenoDRT, HDR, or cICP validation." argument-hint: "pattern type, source gamut/transfer, nits or linear range, dimensions, output format, and validation target"

HDR Test Pattern Generation

Use this small skill for creating source patterns and synthetic datasets. Larger skills should reference it instead of embedding pattern design rules.

Boundaries

  • Focus on generating deterministic test inputs: ramps, sweeps, charts, masks, and diagnostic images.
  • Use analysis-graphing for plots of pattern statistics or transform curves.
  • Use bt2020-png-generation when the final deliverable is a BT.2020 PQ RGB16 PNG with cICP or HDR ICC metadata.
  • Keep one-off experiments in a scratch output path; promote repeated generators or durable fixtures to tools/analysis/ or the relevant test asset folder.
  • Do not silently reuse an existing image if the task needs a controlled source domain; document the source gamut, transfer, white point, and value units.

First classify the pattern goal

Before generating pixels, state what the pattern is intended to expose:

  • Tone-map shape, shoulder, toe, mid-gray, or diffuse-white behavior.
  • Gamut mapping, hue preservation, negative-channel clipping, or out-of-gamut handling.
  • LUT precision, tetrahedral/trilinear interpolation, or banding.
  • PQ/HLG/SDR transfer correctness and viewer metadata behavior.
  • Spatial artifacts: edge halos, checkerboard instability, bloom thresholds, sharpening, or temporal reprojection.
  • Scalar/energy maps for RenoDRT/PsychoV/N2 reapply experiments.

Source domain checklist

Every pattern needs explicit metadata in the script, filename, or sidecar stats:

FieldExamples
Gamut / primariesBT.709, BT.2020, AP1, AP0
White pointD65, D60, adapted D60→D65
Transferscene-linear, display-linear nits, sRGB, PQ
Value scale0..1, stops around 1.0, absolute nits, diffuse-white-relative
Bit depth / formatEXR float, RGB16 PNG, 8-bit preview PNG/WebP
Clipping policypreserve negatives for stats, clip only at output, hard clip to gamut, mask out-of-gamut

Do not normalize pattern maxima unless normalization is the experiment.

Useful pattern families

PatternUse forNotes
Neutral ramp / step wedgeTone curves, PQ quantization, bandingInclude fine ramp plus labeled stops or nits steps.
Hue sweepGamut compression and hue preservationSweep hue in a known gamut; keep saturation/value and luminance assumptions explicit.
Primary/secondary chipsBT.709/BT.2020/AP1 comparisonInclude white/gray references and expected out-of-gamut chips.
OOG stress chartNegative channels, over-1 values, matrix/adaptation errorsPreserve unclipped stats before output clipping.
Banding panel8-bit/10-bit/16-bit path checksUse smooth ramps plus small-amplitude perturbations when needed.
Checker/edge patternSpatial filters, bloom, sharpening, TAA artifactsInclude hard edges and subpixel-aligned variants if relevant.
Synthetic scene chartEnd-to-end tone/gamut validationKeep source EXR and rendered outputs together.
Scalar/energy mapN2/RenoDRT/PsychoV energy diagnosticsReplicate to RGB only at the final visualization/export step.

Output guidance

  • Prefer EXR float for source fixtures that must preserve scene-linear or out-of-range values.
  • Use ordinary SDR PNG only for quick human previews or SDR-specific tests.
  • Use BT.2020 PQ RGB16 PNG plus cICP only for HDR delivery or viewer-signaling tests.
  • Keep stats beside generated outputs: min/max, nonfinite count, negative channels, over-range channels, chosen nits mapping, and source domain.
  • Name files with the important domain choices, for example bt709_linear_hue_sweep, bt2020_hdr_pq_100nits, or syntheticChart_rec709_finite.

Existing references

  • HDR pattern template for deterministic ramps, hue sweeps, and sidecar stats.
  • tests/D3D12HDR-test/images/syntheticChart_rec709.01.exr for a recurring synthetic chart source.
  • tests/D3D12HDR-test/images/RGB_sweep_smooth_31x50.exr for RGB sweep validation.
  • tools/analysis/synthetic_neutwo_energy_reapply_bt2020.py for synthetic chart to BT.2020 HDR output workflow.
  • tools/analysis/validate_mb_compress.py for BT.2020 hue sweep validation patterns.
  • tools/analysis/syntheticChart_rec709_finite/ for generated comparisons and stats.

Common mistakes to avoid

  • Do not mix BT.709, BT.2020, AP1, and AP0 values without naming the conversion path.
  • Do not apply sRGB transfer to data that is already linear or PQ.
  • Do not PQ-encode twice.
  • Do not clip negative or out-of-gamut values before recording diagnostics unless clipping is the behavior being tested.
  • Do not rely on screenshots as validation assets when exact pixels or metadata matter.
  • Do not create a new pattern generator when an existing source fixture or script already covers the same controlled case.