Skip to content

FFT Shift Sync Task

The FFT shift task rearranges the quadrants of a batched 2D tensor so that the zero-frequency component moves between the image corners and the center. It mirrors the behavior of numpy.fft.fftshift and is typically used after a Fast Fourier Transform to center the spectrum for analysis or visualization.

Internally, each frame is split into four equal quadrants: top-left, top-right, bottom-left, and bottom-right, and opposite quadrants are swapped. The operation runs entirely on the GPU and supports real or complex data.

Note

For the zero-frequency component to land exactly at the center, both the width and height of the input frame should be even. When a dimension is odd, the middle row or column stays in place, matching the truncation behavior of fftshift.

Inputs

This task has a single input of shape (B, H, W), where:

  • B: batch size
  • H: frame height
  • W: frame width

Supported dtypes are uint8, uint16, float32, and complex32.
The input tensor must reside in device memory.

Outputs

This task produces one output of shape (B, H, W), matching the input dimensions.
The dtype of the output tensor is identical to the input tensor.
The output tensor is written in device memory.

Inplace

This task has an inplace relationship between its input and output.

Ownership

This task does not own any inputs or outputs.

Settings

Example:

1
{}