Learning generative texture models with extended Fields-of-Experts

摘要: We evaluate the ability of popular Field-of-Experts (FoE) to model structure in images. As a test case we focus on modeling synthetic and natural textures. find that even for single textures, FoE provides insufficient flexibility learn good generative models ‐ it does not perform any better than much simpler Gaussian FoE. propose an extended version (allowing bimodal potentials) demonstrate this novel formulation, when trained with approximation likelihood gradient, gives rise more powerful specific visual produces significantly results texture task.