Struct DiffmapOptions 

pub struct DiffmapOptions {
    pub weighting: DiffmapWeighting,
    pub masking_strength: Option<f32>,
    pub sqrt: bool,
    pub include_edge_mse: bool,
    pub include_hf: bool,
}

Post-processing options for the diffmap signal.

These are applied after multi-scale fusion to shape the diffmap for specific use cases. All options default to off (raw SSIM error signal).

Fields

weighting: DiffmapWeighting

Channel weighting scheme (default: Trained).

masking_strength: Option<f32>

Apply contrast masking to suppress errors in complex/textured regions.

When enabled, each diffmap value is divided by a local complexity mask derived from the source image’s luminance variance: mask = 1 + strength × local_variance(Y_src).

This matches the perceptual intuition that errors in smooth areas (sky, gradients) are more visible than in textured areas (foliage, hair).

Default: None (no masking). Typical values: 2.0–8.0.

sqrt: bool

Apply sqrt to diffmap values after masking.

Compresses the dynamic range so that the diffmap is more linearly proportional to perceived distortion intensity. Similar to butteraugli’s sqrt(dc_masked + ac_masked) combination.

Default: false.

include_edge_mse: bool

Include edge artifact, edge detail loss, and MSE features in the diffmap.

When false (default), the diffmap only includes SSIM error. When true, each pixel also accumulates:

• Edge artifact: max(0, residual_dst² − residual_src²) — structure added by distortion (ringing, blocking).
• Edge detail loss: max(0, residual_src² − residual_dst²) — structure removed by distortion (blurring, smoothing).
• MSE: (src − dst)² — raw per-pixel squared difference.

Feature contributions are weighted by the profile’s trained weights for each feature type and channel. This makes the diffmap a more complete representation of the metric’s per-pixel view.

include_hf: bool

Include high-frequency energy features in the diffmap.

When true, each pixel also accumulates HF texture loss/gain signals:

• HF energy loss: max(0, (src−μ)² − (dst−μ)²) — texture energy removed by distortion (smoothing, quantization).
• HF magnitude loss: max(0, |src−μ| − |dst−μ|) — texture magnitude removed (L1 variant, more robust to outliers).
• HF energy gain: max(0, (dst−μ)² − (src−μ)²) — texture energy added by distortion (ringing, noise).

These correspond to trained features 10–12 and help encoder quantization loops (especially JXL AC coefficient tuning) see where HF texture is being lost or gained. Requires mu1/mu2 storage (same as include_edge_mse).

Default: false.

Trait Implementations

impl Clone for DiffmapOptions

fn clone(&self) -> DiffmapOptions

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for DiffmapOptions

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for DiffmapOptions

fn default() -> DiffmapOptions

Returns the “default value” for a type.

impl From<DiffmapWeighting> for DiffmapOptions

fn from(w: DiffmapWeighting) -> Self

Converts to this type from the input type.

impl Copy for DiffmapOptions