Light-weight pixel context encoders for image inpainting

TL;DR

This paper introduces Pixel Content Encoders, a lightweight neural network for image inpainting that preserves detail and achieves state-of-the-art results, while also enabling image extrapolation without architectural changes.

Contribution

The paper presents a novel, lightweight inpainting model with dilated convolutions that outperforms existing methods and can be used for image extrapolation.

Findings

Achieves state-of-the-art inpainting performance on benchmark datasets.

Has an order of magnitude fewer parameters than previous models.

Can generate content beyond image boundaries without architectural modifications.

Abstract

In this work we propose Pixel Content Encoders (PCE), a light-weight image inpainting model, capable of generating novel con-tent for large missing regions in images. Unlike previously presented convolutional neural network based models, our PCE model has an order of magnitude fewer trainable parameters. Moreover, by incorporating dilated convolutions we are able to preserve fine grained spatial information, achieving state-of-the-art performance on benchmark datasets of natural images and paintings. Besides image inpainting, we show that without changing the architecture, PCE can be used for image extrapolation, generating novel content beyond existing image boundaries.