WAVESTYLEGAN: WAVELET-GENERATIVE ADVERSARIAL NETWORK
Article Sidebar
Main Article Content
Abstract
In this paper a novel generative adversarial network for images WaveStyleGAN that is based on StyleGAN-like architectures, was developed. Key features of the model suggested are processing of wavelet features of images, usage of self-modulated convolutions and modified blocks of Fast Fourier Convolutions in the discriminator. The changes implemented helped to reduce model complexity and its size when compared to the base models’ versions. The model was trained on a dataset of human faces FFHQ in 1024×1024 resolution. It was able to keep a high quality of generated images with considerable decrease in a number of training iterations. Additionally, inference time on CPU was reduced by up to 3 times when compared to the original model, which significantly expands its capabilities for deployments to environments which don’t provide access to computations on GPU.
Article Details
This work is licensed under a Creative Commons Attribution 4.0 International License.
References
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis / P. Esser, S. Kulal, A. Blattmann et al. // arXiv.org. – 2024. – DOI: 10.48550/arXiv.2403.03206.
DEsignBench: Exploring and Benchmarking DALL-E 3 for Imagining Visual Design / K. Lin, Z. Yang, L. Li et al. // arXiv.org. – 2023. – DOI: 10.48550/arXiv.2310.15144.
Analyzing and Improving the Image Quality of StyleGAN / T. Karras, S. Laine, M. Aittala et al. // arXiv.org. – 2019. – DOI: 10.48550/arXiv.1912.04958.
Alias-Free Generative Adversarial Networks / T. Karras, M. Aittala, S. Laine et al. // arXiv.org. – 2021. – DOI: 10.48550/arXiv.2106.12423.
Sauer A., Schwarz K., Geiger A. StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasets // arXiv.org. – 2022. – DOI: 10.48550/arXiv.2202.00273.
SWAGAN: A Style-based Wavelet-driven Generative Model / R. Gal, D. Cohen, A. Bermano, D. Cohen-Or // arXiv.org. – 2021. – DOI: 10.48550/arXiv.2102.06108.
LiteVAE: Lightweight and Efficient Variational Autoencoders for Latent Diffusion Models / S. Sadat, J. Buhmann, D. Bradley et al. // arXiv.org. – 2024. – DOI: 10.48550/arXiv.2405.14477.
Chi L., Jiang B., Mu Y. Fast Fourier Convolution // Advances in Neural Information Processing Systems 33 (NeurIPS 2020). – 2020. – URL: https://papers.nips.cc/paper_files/paper/2020/hash/2fd5d41ec6cfab47e32164d5624269b1-Abstract.html (date of access: 07.09.2025).
Differentiable Augmentation for Data-Efficient GAN Training / S. Zhao, Z. Liu, J. Lin et al. // arXiv.org. – 2020. – DOI: 10.48550/arXiv.2006.10738.
Most read articles by the same author(s)
- U. VARABEI, A. MALEVICH, WAVELET-BASED EFFECTIVE IMAGES DOWNSCALING IN NEURAL NETWORKS, Vestnik of Polotsk State University. Part C. Fundamental Sciences: No. 2 (2024)