Struct ZensimResult 

pub struct ZensimResult { /* private fields */ }

Result from a zensim comparison.

Contains the final score, the raw distance used to derive it, and the full per-scale feature vector (useful for diagnostics or weight training).

Implementations

impl ZensimResult

pub fn nan() -> Self

Create a NaN sentinel result (for error/placeholder paths).

pub fn score(&self) -> f64

Quality score on a 0–100 scale. 100 = identical, 0 = maximally different. Derived from raw_distance via a power-law mapping.

pub fn raw_distance(&self) -> f64

Raw weighted feature distance before nonlinear mapping. Lower = more similar. Not bounded to a fixed range; depends on image content and weights.

pub fn features(&self) -> &[f64]

Per-scale raw features as a slice.

Layout: 4 scales × 3 channels (X, Y, B) × 19 features per channel = 228. See FeatureView for named access.

pub fn into_features(self) -> Vec<f64>

Consume the result and return the owned feature vector.

pub fn profile(&self) -> ZensimProfile

Which profile produced this score.

pub fn mean_offset(&self) -> [f64; 3]

Per-channel XYB mean offset: mean(src_xyb[c]) - mean(dst_xyb[c]).

Captures global color/luminance shifts (CMS errors, white balance changes). Channels: [X, Y, B], signed. Positive = distorted is darker/less saturated.

pub fn dissimilarity(&self) -> f64

Convert the score to a dissimilarity value.

Dissimilarity is (100 - score) / 100: 0 = identical, higher = worse. This is the inverse of the 0–100 score scale, normalized to 0–1.

See also score_to_dissimilarity for the standalone conversion.

pub fn approx_ssim2(&self) -> f64

Approximate SSIMULACRA2 score from the raw distance.

Direct power-law fit: 100 - 19.04 × d^0.598, calibrated on 344k synthetic pairs. MAE: 4.4 SSIM2 points, Pearson r = 0.974.

More accurate than mapping::zensim_to_ssim2(score) (MAE 4.9, r = 0.932) because it skips the intermediate score mapping.

pub fn approx_dssim(&self) -> f64

Approximate DSSIM value from the raw distance.

Direct power-law fit: 0.000922 × d^1.224, calibrated on 344k synthetic pairs. MAE: 0.00129, Pearson r = 0.952.

Significantly more accurate than mapping::zensim_to_dssim(score) (MAE 0.00213, r = 0.719) because DSSIM’s natural exponent (1.22) differs from the score mapping exponent (0.70).

pub fn approx_butteraugli(&self) -> f64

Approximate butteraugli distance from the raw distance.

Direct power-law fit: 2.365 × d^0.613, calibrated on 344k synthetic pairs. MAE: 1.65 distance units, Pearson r = 0.713.

Butteraugli’s weak correlation with our features (r = 0.71) limits approximation accuracy regardless of mapping choice.

Trait Implementations

impl Clone for ZensimResult

fn clone(&self) -> ZensimResult

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ZensimResult

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

Formats the value using the given formatter.