Module metrics 

Quality metrics for image comparison.

This module provides perceptual quality metrics for comparing reference and test images. Supported metrics:

• DSSIM: Structural dissimilarity metric (lower is better, 0 = identical)
• SSIMULACRA2: Perceptual similarity metric (higher is better, 100 = identical)
• Butteraugli: Perceptual difference metric (lower is better, <1.0 = imperceptible)
• PSNR: Peak Signal-to-Noise Ratio (higher is better) - NOT RECOMMENDED

Recommended Metrics

Prefer SSIMULACRA2 or Butteraugli over PSNR. PSNR does not correlate well with human perception. SSIMULACRA2 and Butteraugli are designed to match human visual perception of image quality.

Perception Thresholds

Based on empirical data from imageflow:

Level	DSSIM	SSIMULACRA2	Butteraugli	Description
Imperceptible	< 0.0003	> 90	< 1.0	Visually identical
Marginal	< 0.0007	> 80	< 2.0	Only A/B comparison reveals
Subtle	< 0.0015	> 70	< 3.0	Barely noticeable
Noticeable	< 0.003	> 50	< 5.0	Visible on inspection
Degraded	>= 0.003	<= 50	>= 5.0	Clearly visible artifacts

Modules

butteraugli

Butteraugli metric calculation.

dssim

DSSIM (Structural Dissimilarity) metric calculation.

ssimulacra2

SSIMULACRA2 metric calculation.

Structs

MetricConfig

Configuration for which metrics to calculate.

MetricResult

Results from metric calculations.

Enums

PerceptionLevel

Perceptual quality level based on metric thresholds.

Functions

calculate_psnr

Calculate PSNR between two images.