Parameter-efficient and resolution-robust network architectures for image-to-image translation

A computer program product for providing a convolutional neural network for image-to-image translation, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

• access an image file containing a first amount of data;
• input the image file into a convolutional neural network, the convolutional neural network including a plurality of Fourier layers, each of the Fourier layers including a Fourier transform, a linear feature transformation in a frequency domain, and an inverse Fourier transform, wherein each linear feature transformation in the frequency domain is shared by different frequency components to reduce a number of parameters;
• and
• output by the convolutional neural network, an output image file that includes contents that are translated from the input image file,
• wherein the convolutional neural network implements deep supervision comprising concatenation that combines features from the Fourier layers together and sends a result of the concatenation to a final convolutional layer.

A system comprising:

• a memory configured to store instructions;
• and
• a processor configured to execute the instructions to:
• access an image file containing a first amount of data;
• input the image file into a convolutional neural network, the convolutional neural network including a plurality of Fourier layers, each of the Fourier layers including a Fourier transform, a linear feature transformation in a frequency domain, and an inverse Fourier transform, wherein each linear feature transformation in the frequency domain is shared by different frequency components to reduce a number of parameters;
• and
• output by the convolutional neural network, an output image file that includes contents that are translated from the input image file,
• wherein the convolutional neural network implements deep supervision and the deep supervision comprises concatenation that combines features from the Fourier layers together and sends a result of the concatenation to a final convolutional layer.