Paper notes: Deep generative Image Models using a Laplacian Pyramid of adversarial Networks

Deep generative Image Models using a Laplacian Pyramid of adversarial Networks

NIPS 2015

　　Abstract : This paper presents a generative parametric model capable of producing high quality natural images. Our approach uses the framework of the Laplacian pyramid framework to generate images from a thick-to-thin approach using CNN cascade. At each level of the pyramid, a GAN is used, and our approach produces higher resolution images.

Introduction : In the field of computer vision, the construction of a good production model is a natural image of the comparison of grass-roots problems. However, high-resolution images are still difficult to produce. We propose a way to produce a scene that looks much like the resolution: 32*32 and 64*64. To achieve this, we explored the multiscale structure of natural image and constructed a series of production models, each of which captures the image structure of a particular layer of the pyramid. This strategy translates the original problem into: a sequence of more manageable stages. In each dimension, we use GAN's ideas to build a CNN-generated model. The sample is painted in coarse-to-fine fashion, commencing with a low-frequency residual image. The second stage samples the BAND-PASS structure at the next level, based on the sampled residual. The next level continues the process, always on the output of the previous scale, up to the end. Therefore, drawing samples is an effective, intuitive forward propagation process: the random vector as input, through deep convolutional networks forward propagation, and then output an image.

　　

Paper notes: Deep generative Image Models using a Laplacian Pyramid of adversarial Networks