Operators

Function	Supported Device	Description
Add	HPU, CPU	Performs element-wise addition and supports Broadcasting.
BasicCrop	CPU	Crops input tensor along width, height and depth.
BitwiseAnd	HPU	Computes the output by doing bitwise AND operation on input tensors.
BitwiseOr	HPU	Computes the output by doing bitwise OR operation on input tensors.
BitwiseXor	HPU	Computes the output by doing bitwise XOR operation on input tensors.
Brightness	HPU	Defines the brightness of the image.
Cast	HPU	Changes the data type of the input tensor.
Clamp	HPU	Clamps all elements in input into the range [lowerBound, upperBound].
CocoReader	CPU	Reader for SSD Coco dataset which produces outputs as images and metadata.
ColorSpaceConversion	HPU	Transforms the image from one color representation to another color representation.
Concat	HPU	Concatenates tensors along one dimension.
CoinFlip	HPU, CPU	Outputs a tensor with random boolean values using a Bernoulli distribution.
Constant	HPU, CPU	Constant operator generates scalar constant value tensor.
Contrast	HPU	Defines the contrast of the image.
Crop	HPU, CPU	Crops input tensor with specified window dimensions and window position.
CropMirrorNorm	HPU	Performs fused cropping, mirroring, normalization, and type casting.
ExtCpuOp	CPU	Allows user to add existing CPU operators.
ExtHpuOp	HPU	Allows user to add existing TPC operators to media pipeline.
Flip	HPU	This operator flips selected dimensions (horizontal, vertical, depthwise).
GatherND	HPU	Gathers slices of input tensor into an output tensor of a specific rank.
GaussianBlur	HPU	Applies 1D Gaussian Blur on input tensor by applying grouped 1D convolution with input Gaussian kernel.
Hue	HPU	Modifies the hue of an image by specified degrees.
ImageDecoder	HPU	Decodes and resizes batch of images.
MediaConst	CPU	Generates constant data as per given shape and data type on every iteration.
MediaExtReaderOp	CPU	Integrates external data reader with media pipe.
MediaFunc	CPU	Allows user to generate new tensor data on every iteration.
MemCpy	HPU	Copies data from the input tensor to the output tensor.
Mult	HPU, CPU	Performs element-wise multiplication and supports Broadcasting.
Neg	HPU	Resultant output tensor has the sign flipped for each element in input.
Normalize	HPU	Normalizes the input tensor by subtracting mean and dividing by standard deviation.
Pad	HPU	Pads a tensor with zeros or any numbers. .
RandomBiasedCrop	CPU	Performs random crop operation based on label data.
RandomFlip	HPU, CPU	Flips images in a selected direction (horizontal or vertical) with a random predicate.
RandomNormal	HPU, CPU	Generates random numbers from a normal distribution.
RandomUniform	HPU, CPU	Generates random numbers from an uniform distribution.
ReadImageDatasetFromDir	CPU	Reader designed for image classification tasks.
ReadNumpyDatasetFromDir	CPU	Reads numpy data files and numpy label files either from given directory or file list and returns the batch of numpy images and labels.
ReduceMax	HPU	Computes the maximum of the input tensor’s elements along the provided dimension.
ReduceMin	HPU	Computes the minimum of the input tensor’s elements along the provided dimension.
Reshape	HPU	Returns a new tensor that has the same values as input tensor in the same order, except with a new shape given by size and tensorDim arguments.
Resize	HPU	Resizes an image by a scale factor using either bilinear, cubic, nearest neighbor or gaussian sampling.
Saturation	HPU	Modifies the saturation level of an image by a specified change factor.
Slice	HPU	Produces a slice of the input tensor along multiple axes.
Split	HPU	Splits a tensor into a list of tensors along the specified axis.
SSDMetadata	CPU	Takes the metadata output of Coco reader and performs operations such as crop, flip and encoding of bounding boxes.
Sub	HPU	Performs element-wise subtraction and supports Broadcasting.
Transpose	HPU	Produces a transposed tensor of the input tensor along multiple axes.
Where	HPU, CPU	Outputs a tensor of elements selected from either input0 or input1.
Zoom	CPU	Zoom operator supports zoom-in operation on the given images and labels.