Gram-GAN bridges pixel-based and GAN SR methods, generating sharp and realistic textures by explicitly matching generated patches to real HR textures via a Gram-based patch database. It ensures stable training, better generalization, and texture-aware outputs—ideal for satellite, medical, art, and fashion super-resolution tasks.
contentful pytorch generative-adversarial-network realistic vgg19 texture-analysis perceptual-quality gram-matrix faiss-vector-database rrdbnet
-
Updated
Nov 5, 2025 - Python