2022/01/19 by Sara Björk, Björk, Sara, Jonas Nordhaug Myhre +3 · 3 citations
Computer Science · Mathematics · Physics and Astronomy · #Algorithm #Artificial intelligence #Bilinear interpolation #Computational Physics and Python Applications #Computer Vision and Pattern Recognition (cs.CV) #Computer science #FOS: Computer and information sciences #Fourier transform #Generative Adversarial Networks and Image Synthesis #Generative grammar #Generator (circuit theory) #Machine Learning (cs.LG) #Mathematics #Model Reduction and Neural Networks #Pattern recognition (psychology) #Power (physics) #Regularization (linguistics) #Sampling (signal processing) #cs.CV #cs.LG
paper · pdf · doi:10.48550/arxiv.2201.07544
published in arXiv (Cornell University) (Cornell University) · Submitted to ICASSP 2022, 2022 IEEE International Conference on Acoustics, Speech and Signal Processing
arxiv created 2022/01/19 · openalex publication_date 2022/01/19 · arxiv updated 2022/01/20 · openalex created_date 2022/04/03 · openalex updated_date 2026/08/05
Full characterization of the spectral behavior of generative models based on neural networks remains an open issue. Recent research has focused heavily on generative adversarial networks and the high-frequency discrepancies between real and generated images. The current solution to avoid this is to either replace transposed convolutions with bilinear up-sampling or add a spectral regularization term in the generator. It is well known that Variational Autoencoders (VAEs) also suffer from these issues. In this work, we propose a simple 2D Fourier transform-based spectral regularization loss for the VAE and show that it can achieve results equal to, or better than, the current state-of-the-art in frequency-aware losses for generative models. In addition, we experiment with altering the up-sampling procedure in the generator network and investigate how it influences the spectral performance of the model. We include experiments on synthetic and real data sets to demonstrate our results.