Drop The GAN: In Defense Of Patches Nearest Neighbors As Single Image Generative Models

Abstract

Single image generative models perform synthesis and manipulation tasks by capturing the distribution of patches within a single image. The classical (pre Deep Learning) prevailing approaches for these tasks are based on an optimization process that maximizes patch similarity between the input and generated output. Recently, however, Single Image GANs were introduced both as a superior solution for such manipulation tasks, but also for remarkable novel generative tasks. Despite their impressiveness, single image GANs require long training time (usually hours) for each image and each task. They often suffer from artifacts and are prone to optimization issues such as mode collapse. In this paper, we show that all of these tasks can be performed without any training, within several seconds, in a unified, surprisingly simple framework. We revisit and cast the "good-old" patch-based methods into a novel optimization-free framework. We start with an initial coarse guess, and then simply refi

Stats

• citations51
• S2 citations—
• github stars0
• HF likes0
• heat score12.87
• arxiv keygranot2021drop

Related papers

• Inverting The Generator Of A Generative Adversarial Network (2016)17.61
• Dark Side Augmentation: Generating Diverse Night Examples For Metric Learning (2023)8.34
• Semi-parametric Neural Image Synthesis (2022)0.00
• Regularizing Deep Hashing Networks Using GAN Generated Fake Images (2018)0.00
• Retrievegan: Image Synthesis Via Differentiable Patch Retrieval (2020)9.92
• Deep Semantic Hashing With Generative Adversarial Networks (2018)13.50
• Evaluating Post-training Compression In Gans Using Locality-sensitive Hashing (2021)0.00
• AR-RAG: Autoregressive Retrieval Augmentation For Image Generation (2025)0.00