This paper studies a combination of generative Markov random field (MRF) models and discriminatively trained deep convolutional neural networks (dCNNs) for synthesizing 2D images.
The recent work of Gatys et al., who characterized the style of an image by the statistics of convolutional neural network filters, ignited a renewed interest in the texture generation and image stylization problems.
This paper proposes Markovian Generative Adversarial Networks (MGANs), a method for training generative neural networks for efficient texture synthesis.
Given an input dynamic texture, statistics of filter responses from the object recognition ConvNet encapsulate the per-frame appearance of the input texture, while statistics of filter responses from the optical flow ConvNet model its dynamics.
Second, we show that the image generation with PSGANs has properties of a texture manifold: we can smoothly interpolate between samples in the structured noise space and generate novel samples, which lie perceptually between the textures of the original dataset.
Single image super-resolution is the task of inferring a high-resolution image from a single low-resolution input.
#11 best model for Image Super-Resolution on BSD100 - 4x upscaling
Generative adversarial networks (GANs) are a recent approach to train generative models of data, which have been shown to work particularly well on image data.
Gatys et al. (2015) showed that pair-wise products of features in a convolutional network are a very effective representation of image textures.