Struct CutMixAugmenter 

pub struct CutMixAugmenter {
    pub alpha: f64,
}

CutMix augmentation (ICCV 2019).

Instead of mixing pixels uniformly like Mixup, CutMix cuts a rectangular region from one image and pastes it to another. Labels are mixed proportionally to the area of the patch.

Reference: Yun et al. “CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features” (ICCV 2019)

Fields

alpha: f64

Beta distribution parameter for sampling mixing ratio.

Implementations

impl CutMixAugmenter

pub fn new(alpha: f64) -> TrainResult<Self>

Create a new CutMix augmenter.

Arguments

• alpha - Beta distribution parameter (typically 1.0)

Returns

New CutMix augmenter

pub fn augment_batch( &self, data: &ArrayView2<'_, f64>, labels: &ArrayView2<'_, f64>, rng: &mut StdRng, ) -> TrainResult<(Array2<f64>, Array2<f64>)>

Apply CutMix augmentation to a batch of data.

For 2D feature arrays, we treat the second dimension as a “spatial” dimension and cut rectangular regions along it.

Arguments

• data - Input data batch [N, features]
• labels - Corresponding labels [N, classes]
• rng - Random number generator

Returns

Tuple of (augmented_data, augmented_labels)

Trait Implementations

impl Clone for CutMixAugmenter

fn clone(&self) -> CutMixAugmenter

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl DataAugmenter for CutMixAugmenter

fn augment( &self, data: &ArrayView2<'_, f64>, _rng: &mut StdRng, ) -> TrainResult<Array2<f64>>

Augment the given data.

impl Debug for CutMixAugmenter

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

Formats the value using the given formatter.

impl Default for CutMixAugmenter

fn default() -> Self

Returns the “default value” for a type.