Measure image quality
Quality measurement is integrated into the pipeline — there is no separate
measurement step. When you run just pipeline, each encoded image is measured
automatically before moving to the next source image.
How it works
Section titled “How it works”For every encoded variant, the pipeline measures four metrics:
- SSIMULACRA2 — perceptual quality (higher is better)
- PSNR — peak signal-to-noise ratio in dB (higher is better)
- SSIM — structural similarity, 0–1 (higher is better)
- Butteraugli — perceptual distance (lower is better)
Results are saved to data/metrics/<study-id>/quality.json.
Running measurements
Section titled “Running measurements”just pipeline format-comparison 30mThis encodes and measures in a single pass. See Run the pipeline for time budgets and advanced options.
Verifying tools
Section titled “Verifying tools”The measurement tools must be available in the environment:
just verify-toolsAll tools (ssimulacra2, butteraugli_main, ffmpeg) are pre-installed in the dev container.
See also
Section titled “See also”- Run the pipeline — the main command for encoding + measurement
- Analyze results — generate statistics and plots from measurements
- Tools reference — metric interpretation tables and CLI usage