Generating natural images with direct Patch Distributions Matching

TL;DR

This paper introduces a simple, training-free method for generating high-quality images by directly matching patch distributions using Sliced Wasserstein Distance, outperforming some GANs in speed and quality.

Contribution

The authors propose a novel, efficient algorithm that explicitly minimizes patch distribution differences without training, improving image generation quality and speed.

Findings

Outperforms single-image-GANs in quality

Requires no training and is easy to implement

Generates high-quality images in seconds

Abstract

Many traditional computer vision algorithms generate realistic images by requiring that each patch in the generated image be similar to a patch in a training image and vice versa. Recently, this classical approach has been replaced by adversarial training with a patch discriminator. The adversarial approach avoids the computational burden of finding nearest neighbors of patches but often requires very long training times and may fail to match the distribution of patches. In this paper we leverage the recently developed Sliced Wasserstein Distance and develop an algorithm that explicitly and efficiently minimizes the distance between patch distributions in two images. Our method is conceptually simple, requires no training and can be implemented in a few lines of codes. On a number of image generation tasks we show that our results are often superior to single-image-GANs, require no training, and can generate high quality images in a few seconds. Our implementation is available at https://github.com/ariel415el/GPDM