Augmentation

250 public types in this namespace, organized by kind.

Models & Types (192)

Type Summary
AdaIN<T> Adaptive Instance Normalization (AdaIN) for style transfer augmentation.
AdasynAugmenter<T> Implements ADASYN (Adaptive Synthetic Sampling) for imbalanced datasets.
Affine<T> Applies random affine transformations (rotation, scale, shear, translation) to an image.
AudioNoise<T> Adds background noise to audio data.
AugMax<T> AugMax (Wang et al., 2021) - adversarial composition of random augmentations.
AugmentationAppliedEventArgs<T> Event arguments raised when an augmentation is applied.
AugmentationContext<T> Provides runtime context for augmentation operations including random state, training mode, and spatial targets.
AugmentationPipeline<T, TData> Represents a pipeline of augmentations that are applied in sequence or composition.
AugmentationPolicyEntry<T> An entry in an augmentation policy.
AugmentationPolicy<T> Defines a serializable augmentation policy with named transforms and parameters.
AugmentationRecommendation Represents a recommendation for an augmentation with its configuration.
AugmentationScheduler<T> Schedules augmentation strength changes during training.
AugmentationSearchSpace Represents the search space for augmentation hyperparameters.
AugmentationValidationResult Result of validating augmentations for a task.
AugmentedSample<T, TData> Represents a sample with its data and associated spatial targets.
AutoAugment<T> AutoAugment (Cubuk et al., 2018) - applies learned augmentation policies.
AutoContrast<T> Maximizes image contrast by stretching the intensity range to fill [0, max].
BBoxClipToImage<T> Clips bounding boxes to image boundaries.
BBoxSafeRandomCrop<T> Random crop that ensures bounding boxes remain valid after cropping.
BiasFieldCorrection<T> Simulates or corrects MRI bias field (intensity non-uniformity).
BorderlineSmoteAugmenter<T> Implements Borderline-SMOTE for imbalanced datasets, focusing on samples near the decision boundary.
BoundingBox<T> Represents a bounding box annotation for object detection.
BoxBlur<T> Applies simple box (mean) blur using a uniform kernel.
Brightness<T> Adjusts the brightness of an image by adding a random offset to all pixel values.
CLAHE<T> Applies Contrast Limited Adaptive Histogram Equalization (CLAHE).
CenterCropOrPad<T> Crops or pads an image to reach the target size, centering the content.
CenterCrop<T> Crops the center region of an image to a specified size.
ChannelDropout<T> Randomly zeros out one or more color channels.
ChannelShuffle<T> Randomly shuffles the order of color channels.
Closing<T> Morphological closing (dilation followed by erosion) - removes small dark spots.
Clouds<T> Simulates cloud overlay on the image.
CoarseDropout<T> Drops rectangular regions from the image (similar to Cutout/Random Erasing).
ColorConstancy<T> Applies color constancy correction using the Gray World assumption or Max-RGB method.
ColorJitter<T> Applies random combinations of brightness, contrast, saturation, and hue adjustments.
Compose<T, TData> Applies multiple augmentations sequentially.
Contrast<T> Adjusts the contrast of an image by scaling pixel values around the mean.
CopyPaste<T> Copy-Paste augmentation (Ghiasi et al., 2020) - copies random regions from a source image and pastes them onto the target image.
CropNonEmptyMaskIfExists<T> Crops the image to a region that contains non-empty mask content.
CutMix<T> Cuts a rectangular region from one image and pastes it onto another (CutMix augmentation).
Cutout<T> Randomly masks out (cuts out) rectangular regions of an image.
DADA<T> DADA - Differentiable Automatic Data Augmentation (Li et al., 2020).
DatasetCharacteristics Represents metadata about a dataset for augmentation recommendations.
Defocus<T> Simulates defocus blur using a circular (disc) kernel.
Denormalize<T> Reverses normalization of an image tensor, restoring original pixel value ranges.
Dilate<T> Morphological dilation - expands bright regions using a structuring element.
Downscale<T> Downscales then upscales the image to simulate resolution degradation.
ElasticTransform<T> Applies elastic deformation to an image (Simard et al., 2003).
Emboss<T> Applies an emboss filter to create a 3D shadow effect.
Equalize<T> Equalizes the image histogram per channel (same as HistogramEqualization with per-channel mode).
Erode<T> Morphological erosion - shrinks bright regions using a structuring element.
FDA<T> Fourier Domain Adaptation (Yang & Soatto, 2020) - swaps low-frequency spectral components between source and target images for domain adaptation.
FMix<T> FMix augmentation (Harris et al., 2020) - mixes images using Fourier-space masks.
FancyPCA<T> Fancy PCA color augmentation as described in AlexNet (Krizhevsky et al.
FastAutoAugment<T> Fast AutoAugment (Lim et al., 2019) - efficient augmentation policy search using density matching.
FeatureDropout<T> Randomly masks (sets to zero) features in tabular data.
FeatureNoise<T> Adds Gaussian noise to numerical features in tabular data.
FiveCrop<T> Extracts five crops from an image: four corners and center.
Fog<T> Simulates fog effect by blending the image with a fog layer.
FrameDropout<T> Randomly drops frames from video.
Frost<T> Simulates frost/ice crystal patterns on the image.
GammaCorrection<T> Applies gamma correction to adjust image brightness non-linearly.
GaussianBlur<T> Applies Gaussian blur to an image.
GaussianNoise<T> Adds Gaussian noise to an image.
GhostingArtifact<T> Simulates MRI ghosting artifacts caused by motion during acquisition.
GlassBlur<T> Applies glass-like blur by randomly displacing pixels then smoothing.
GrayscaleToRgb<T> Converts a grayscale image to RGB by replicating the single channel across all three channels.
GridDistortion<T> Applies grid-based distortion to an image.
GridDropout<T> Drops cells from a regular grid overlay on the image.
GridMask<T> GridMask augmentation (Chen et al.
HideAndSeek<T> Hide-and-Seek augmentation (Singh & Lee 2017) that randomly hides grid patches.
HistogramColorTransfer<T> Transfers color distribution from a reference image using histogram matching.
HistogramEqualization<T> Applies standard histogram equalization to improve image contrast.
HistogramMatching<T> Matches the histogram of an image to a reference histogram or target distribution.
HorizontalFlip<T> Flips an image horizontally (left-right mirror).
HueSaturationValue<T> Randomly adjusts hue, saturation, and value in HSV space.
HyperparameterDefinition Represents a hyperparameter definition for AutoML search.
ISONoise<T> Simulates camera sensor noise (ISO noise) with separate color and intensity components.
ImageCompression<T> General compression artifact simulation that randomly selects between JPEG and WebP styles.
ImagePreprocessor<T> Unified preprocessing pipeline builder for chaining image transformations.
ImageTensor<T> Represents an image as a tensor with image-specific metadata and operations.
IntensityNormalization<T> Medical image intensity normalization using z-score or min-max normalization.
Invert<T> Inverts all pixel values in the image (creates a negative).
JpegCompression<T> Simulates JPEG compression artifacts by quantizing DCT coefficients.
KSpaceMotion<T> Simulates K-space motion artifacts in MRI by introducing phase shifts.
Keypoint<T> Represents a keypoint annotation for pose estimation and landmark detection.
LabelMixingEventArgs<T> Event arguments for label mixing operations in augmentations like Mixup and CutMix.
LongestMaxSize<T> Resizes the image so that the longest edge equals the specified max size, preserving aspect ratio.
MaskDropout<T> Drops random object masks from the segmentation mask.
MedianBlur<T> Applies median filter blur, effective at removing salt-and-pepper noise while preserving edges.
MinIoURandomCrop<T> Random crop with minimum IoU guarantee for bounding boxes (SSD-style).
MixUp<T> Blends two images together by weighted averaging (MixUp augmentation).
MorphologicalGradient<T> Morphological gradient - the difference between dilation and erosion, highlighting edges.
Mosaic9<T> 9-image mosaic variant that arranges images in a 3x3 grid.
Mosaic<T> YOLO-style 4-image mosaic augmentation (Bochkovskiy et al., 2020).
MotionBlur<T> Applies directional motion blur simulating camera or object motion.
MultiplicativeNoise<T> Applies multiplicative Gaussian noise: output = input * (1 + noise).
NearMissUnderSampler<T> Implements NearMiss undersampling for imbalanced datasets.
Normalize<T> Normalizes an image tensor with per-channel mean and standard deviation.
OneOf<T, TData> Randomly selects and applies exactly one augmentation from a set.
Opening<T> Morphological opening (erosion followed by dilation) - removes small bright spots.
OpticalDistortion<T> Simulates barrel and pincushion lens distortion.
PadToSquare<T> Pads an image to make it square while preserving the original content centered.
Pad<T> Pads an image with configurable padding amounts and fill modes.
Perspective<T> Applies a random perspective transformation to an image.
PiecewiseAffine<T> Applies piecewise affine transformation by dividing the image into triangular regions.
PitchShift<T> Shifts the pitch of audio without changing tempo using WSOLA (Waveform Similarity Overlap-Add).
PixelDropout<T> Randomly sets individual pixels to a fill value.
PlasmaTransform<T> Plasma fractal transformation - applies a procedural plasma-like color perturbation.
PoissonNoise<T> Adds Poisson (shot) noise that scales with pixel intensity, simulating photon counting noise.
PolicySearchSpace Represents a complete search space for augmentation policies.
Posterize<T> Reduces the number of bits per color channel, creating a poster-like effect.
PuzzleMix<T> PuzzleMix (Kim et al., 2020) - optimal mixing using saliency and local statistics.
Rain<T> Simulates rain streaks on the image.
RandAugment<T> RandAugment (Cubuk et al., 2019) - applies N random augmentations at magnitude M.
RandomBrightnessContrast<T> Randomly adjusts brightness and contrast simultaneously.
RandomCrop<T> Randomly crops a region from the image.
RandomDeletion<T> Randomly deletes words from text.
RandomErasing<T> Random Erasing augmentation (Zhong et al.
RandomGamma<T> Applies random gamma correction with a random gamma value within a range.
RandomInsertion<T> Randomly inserts synonyms of existing words into the text.
RandomResizedCrop<T> Randomly crops and resizes a region of the image (PyTorch-style RandomResizedCrop).
RandomSizedBBoxSafeCrop<T> Random-sized crop with bounding box safety, followed by resize to target size.
RandomSwap<T> Randomly swaps the positions of words in text.
RandomToneCurve<T> Applies a random tone curve transformation to adjust image tonality.
RandomUnderSampler<T> Implements random undersampling to balance imbalanced datasets.
ResizeMix<T> ResizeMix augmentation - resizes one image and pastes it onto another.
ResizeWithAspectRatio<T> Resizes an image to fit within target dimensions while preserving the aspect ratio, then pads to reach the exact target size.
Resize<T> Resizes an image to a target size using configurable interpolation.
RgbShift<T> Independently shifts each RGB channel by a random amount.
RgbToBgr<T> Converts between RGB and BGR color channel orderings.
RgbToGrayscale<T> Converts an RGB image to grayscale using configurable channel weights.
RgbToHls<T> Converts an image between RGB and HLS (Hue, Lightness, Saturation) color spaces.
RgbToHsv<T> Converts an image between RGB and HSV color spaces.
RgbToLab<T> Converts an image between RGB and CIE Lab* color space.
RgbToXyz<T> Converts an image between RGB and CIE XYZ color space.
RgbToYuv<T> Converts an image between RGB and YUV color spaces.
Rotation<T> Rotates an image by a random angle within a specified range.
RowShuffle<T> Shuffles rows within a batch of tabular data.
SaliencyMix<T> SaliencyMix - saliency-guided image mixing that pastes salient regions.
SaltAndPepperNoise<T> Adds salt-and-pepper (impulse) noise by randomly setting pixels to black or white.
SamplePairing<T> Sample Pairing augmentation (Inoue, 2018) - averages two images together.
Saturation<T> Adjusts the saturation (color intensity) of an image.
Scale<T> Scales an image by a random factor within a specified range.
SegmentationMask<T> Represents a segmentation mask for pixel-level annotations.
Shadow<T> Adds random shadow regions to the image.
Sharpen<T> Applies a sharpening filter using a 3x3 kernel.
SkeletonDefinition Represents a skeleton definition for pose estimation.
SliceSelection<T> Randomly selects a slice from a multi-channel volume (simulates 3D volume slice selection).
SmallestMaxSize<T> Resizes the image so that the shortest edge equals the specified max size, preserving aspect ratio.
SmoteAugmenter<T> Implements SMOTE (Synthetic Minority Over-sampling Technique) for imbalanced datasets.
SmoteEnnAugmenter<T> Implements SMOTE-ENN combination for imbalanced datasets.
SmoteTomekAugmenter<T> Implements SMOTE-Tomek combination for imbalanced datasets.
SnapMix<T> SnapMix (Huang et al., 2021) - semantically proportional mixing for fine-grained recognition.
Snow<T> Simulates snow effect on the image.
Solarize<T> Inverts all pixel values above a threshold, creating a solarization effect.
SomeOf<T, TData> Randomly selects and applies N augmentations from a set.
SpatialTransform<T> Applies spatial transformations (flips, rotations) consistently to all frames.
Spatter<T> Simulates spatter effects (mud, rain drops) on the image.
SpeckleNoise<T> Adds multiplicative speckle noise: output = input + input * noise.
SpeedChange<T> Changes the playback speed of video by resampling frames.
SpikeArtifact<T> Simulates MRI spike (Herringbone) artifacts caused by electrical interference in k-space.
StackSlices<T> Stacks adjacent 2D slices to create multi-channel input (useful for 3D volume processing).
StyleMix<T> Style mixing augmentation - transfers statistical style (mean/variance) from a reference image.
SunFlare<T> Simulates sun flare / lens flare effect on the image.
Superpixels<T> Replaces random regions with their superpixel (average color) representation.
SvmSmoteAugmenter<T> Implements SVM-SMOTE for imbalanced datasets, using SVM decision boundary to identify borderline samples.
SynonymReplacement<T> Replaces random words with their synonyms.
TabularMixUp<T> Applies MixUp augmentation to tabular data by linearly interpolating between samples.
TemporalCrop<T> Randomly crops a temporal segment from video.
TemporalFlip<T> Reverses the order of frames in a video.
TenCrop<T> Extracts ten crops from an image: five crops plus their horizontal flips.
TestTimeAugmentationResult<TOutput> Contains the result of a Test-Time Augmentation prediction, including individual and aggregated predictions.
TextureOverlay<T> Overlays a procedurally generated random texture on the image.
ThinPlateSpline<T> Applies thin plate spline (TPS) transformation to an image.
TimeShift<T> Shifts audio forward or backward in time.
TimeStretch<T> Stretches or compresses audio in time without changing pitch.
ToFloat<T> Converts an image tensor to floating-point representation with configurable scaling.
ToGray<T> Converts to grayscale with random channel weights, outputting 3 channels.
ToSepia<T> Applies a sepia tone filter to the image.
ToTensor<T> Converts an image to a normalized tensor with values in [0, 1].
TokenMix<T> TokenMix - token-level mixing for transformer architectures.
TomekLinksAugmenter<T> Implements Tomek Links removal for cleaning decision boundaries in imbalanced datasets.
TransMix<T> TransMix (Chen et al., 2022) - attention-guided mixing for Vision Transformers.
TrivialAugment<T> TrivialAugmentWide (Muller & Hutter, 2021) - applies a single random augmentation with uniformly sampled magnitude.
UniformAugment<T> UniformAugment - applies a random number of randomly selected augmentations with uniformly sampled magnitudes.
UnsharpMask<T> Applies unsharp masking to sharpen the image.
VerticalFlip<T> Flips an image vertically (top-bottom mirror).
VideoColorJitter<T> Applies color jitter (brightness, contrast, saturation) to video frames.
VolumeChange<T> Randomly changes the volume (gain) of audio.
WebPCompression<T> Simulates WebP compression artifacts (similar to JPEG but with different characteristics).
WindowLevel<T> CT/MRI window/level adjustment for medical image augmentation.
ZoomBlur<T> Applies radial/zoom blur emanating from the image center.

Base Classes (13)

Type Summary
AudioAugmenterBase<T> Base class for audio data augmentations.
AugmentationBase<T, TData> Abstract base class for all augmentations providing common functionality.
ImageAugmenterBase<T> Base class for image data augmentations.
ImageMixingAugmenterBase<T> Base class for image augmentations that mix multiple images together.
LabelMixingAugmentationBase<T, TData> Base class for label-mixing augmentations like Mixup and CutMix.
SpatialAugmentationBase<T, TData> Base class for augmentations that transform spatial targets (bounding boxes, keypoints, masks).
SpatialImageAugmenterBase<T> Base class for image augmentations that transform spatial targets.
SpatialVideoAugmenterBase<T> Base class for spatial video augmentations that apply image transforms to all frames.
TabularAugmenterBase<T> Base class for tabular data augmentations.
TabularMixingAugmenterBase<T> Base class for tabular augmentations that mix multiple samples together.
TemporalAugmenterBase<T> Base class for temporal video augmentations that modify the frame sequence.
TextAugmenterBase<T> Base class for text data augmentations.
VideoAugmenterBase<T> Base class for video data augmentations.

Interfaces (11)

Type Summary
IAugmentationFactory<T, TData> Factory for creating augmentations from configuration.
IAugmentationPolicy Non-generic base interface for augmentation policies.
IAugmentationPolicy<T, TData> Interface for composable augmentation policies.
IAugmentationRecommender Interface for recommending augmentations based on task and data characteristics.
IAugmentationSearcher<T, TData> Interface for AutoML search algorithms over augmentation spaces.
IAugmentation<T, TData> Base interface for all data augmentations across domains (image, audio, tabular).
IAutoMLAugmentation<T, TData> Interface for augmentations that expose their hyperparameter search space.
IAutoMLPolicy<T, TData> Interface for policies that expose their complete search space.
ILabelMixingAugmentation<T, TData> Interface for augmentations that modify labels (e.g., Mixup, CutMix).
ISpatialAugmentation<T, TData> Interface for augmentations that can transform spatial targets (bounding boxes, keypoints, segmentation masks).
IUnderSampler<T> Interface for undersampling techniques that reduce the majority class.

Enums (23)

Type Summary
AugmentOp Augmentation operation type for AutoAugment policies.
AugmentationOrder Specifies how multiple augmentations are applied in a pipeline.
AugmentationTaskType Specifies the type of machine learning task for augmentation recommendations.
AutoAugmentPolicy AutoAugment policy preset.
BorderMode Specifies how to handle pixels that fall outside the image bounds during transformation.
BoundingBoxFormat Specifies the format of bounding box coordinates.
ChannelOrder Specifies the channel ordering of an image tensor.
ColorConstancyMethod Color constancy correction method.
ColorSpace Specifies the color space of an image.
DataModality Enum representing detected data modality types.
HyperparameterType Specifies the type of hyperparameter.
IntensityNormMethod Intensity normalization method.
InterpolationMode Specifies the interpolation method for image resizing.
MaskEncoding Specifies the encoding format for segmentation masks.
MaskType Specifies the type of segmentation mask.
MixingStrategy Specifies the mixing strategy used for label mixing.
NearMissVersion<T> NearMiss variant versions.
NoiseType Types of audio noise.
PaddingMode Specifies the padding mode for filling new pixels.
PredictionAggregationMethod Specifies how to combine predictions from multiple augmented versions of the same input.
RemovalStrategy<T> Strategy for removing samples in Tomek links.
ScheduleType Schedule type for augmentation strength.
SpatterMode Spatter effect type.

Options & Configuration (8)

Type Summary
AudioAugmentationSettings Audio-specific augmentation settings with industry-standard defaults.
AugmentationConfig Unified configuration for data augmentation with industry-standard defaults.
AugmentationConfig<T, TInput> Strongly-typed augmentation configuration parameterised by the model’s numeric type and input type.
ImageAugmentationSettings Image-specific augmentation settings with industry-standard defaults.
SampledConfiguration Represents a sampled configuration from the search space.
TabularAugmentationSettings Tabular-specific augmentation settings with industry-standard defaults.
TextAugmentationSettings Text-specific augmentation settings with industry-standard defaults.
VideoAugmentationSettings Video-specific augmentation settings with industry-standard defaults.

Helpers & Utilities (3)

Type Summary
DataModalityDetector Utility for auto-detecting data modality from input types.
ImagePresets<T> Preset preprocessing configurations for common models.
PolicyRegistry<T> Registry for built-in and custom augmentation policies.