Fourier transform-based image synthesis using neural networks
AssigneeNVIDIA Corporation
InventorsMorteza Mardani Korani, Guilin Liu, Aysegul Dundar, Shiqiu Liu, Andrew J. Tao, Bryan Christopher Catanzaro
Life of this patent
Estimating the term…
Abstract
Apparatuses, systems, and techniques to scale textured images using a Fourier transform in conjunction with one or more neural networks. In at least one embodiment, a neural network generates an expanded image from an input image by applying a Fourier transform to one or more feature maps generated by said neural network and up-scaling one or more resulting frequency domain feature maps before generating an expanded output image based on up-scaled feature maps.
Claims36
One or more processors comprising:
- circuitry to use one or more neural networks to:
- transform one or more pixels of an original image to a frequency domain by calculating a Fourier transform of a set of features corresponding to the one or more pixels from the original image;
- upsample the one or more transformed pixels in the frequency domain;
- and,
- perform one or more inverse Fast Fourier Transform (FFT) operations to convert the one or more transformed pixels to an image domain.
A system comprising:
- one or more processors to use one or more neural networks to;
- transform one or more pixels of an original image to a frequency domain by calculating a Fourier transform of a set of features corresponding to the one or more pixels from the original image;
- upsample the one or more transformed pixels in the frequency domain;
- and,
- perform one or more inverse Fast Fourier Transform (FFT) operations to convert the one or more transformed pixels to an image domain.
A non-transitory machine-readable medium having stored thereon a set of instructions, which if performed by one or more processors, cause the one or more processors to at least:
- use one or more neural networks to;
- transform one or more pixels of an original image to a frequency domain by calculating a Fourier transform of a set of features corresponding to the one or more pixels from the original image;
- upsample the one or more transformed pixels in the frequency domain;
- and,
- perform one or more inverse Fast Fourier Transform (FFT) operations to convert the one or more transformed pixels to an image domain.
A method comprising:
- using one or more neural networks to:
- transform one or more pixels of an original image to a frequency domain by calculating a Fourier transform of a set of features corresponding to the one or more pixels from the original image;
- upsample the one or more transformed pixels in the frequency domain;
- and
- perform one or more inverse Fast Fourier Transform (FFT) operations to convert the one or more transformed pixels to an image domain.
Dependencies are inferred from the claim text.