I've been delving Fourier Neural Operator recently, but one aspect continues to perplex me: the "zero-shot" super-resolution.

The confusion lies in the implementation. To my understanding, the NN architecture has fixed input-output size.

Let's say our NN is trained to process 2D images of dimensions (100x100) pixels with pixel sizes = 1m x 1m.

My question is: how to use this neural network to produce images of (10000x10000) pixels with pixel sizes = 1cm x 1cm during inferences?

Thanks a lot.



You must log in to answer this question.

Browse other questions tagged .