Abstract:Image inpainting aims to restore missing or damaged regions of an image with plausible visual content. Most existing methods always face challenges when dealing with large hole images, such as structural distortion and prominent artifacts, which mainly stem from the absence of adequate semantic guidance. To overcome this problem, this paper proposes a novel progressive image inpainting network driven by dynamic context, where the guided restoration dynamic semantic prior not only includes information from the known region but also considers the semantic inference during the filling process. Specifically, a multi-view cooperative strategy is proposed firstly to predict the inherent semantic information by estimating the distributions of the image, grayscale and edge of the masked image. In addition, to cope with potential semantic changes, an auxiliary generative unit is proposed for learning semantic information of the intermediate inpainting results, where the learned intermediate semantics are fused with the intrinsic semantics in a weighted manner to form a real-time updating dynamic context. Moreover, the dynamic semantic prior is propagated to various stages of inpainting to assist in constructing refined feature maps with multiple levels of resolution. Experiments on CelebA-HQ and Paris street view datasets demonstrate that the proposed approach can recover reasonable structures and realistic textures on large-scale masked images, achieving advanced performance.

Generative Image Inpainting with Residual Texture Prior and Cross-Layer Contextual Attention

Generative Image Inpainting with Residual Attention Learning

Image Inpainting Via Conditional Texture and Structure Dual Generation

Image Inpainting Based on Interactive Separation Network and Progressive Reconstruction Algorithm

Dynamic Context-Driven Progressive Image Inpainting with Auxiliary Generative Units

Image Inpainting: A Contextual Consistent and Deep Generative Adversarial Training Approach

Image Inpainting Using Wasserstein Generative Adversarial Network

Progressive Inpainting Strategy with Partial Convolutions Generative Networks (PPCGN).

Image Fine-grained Inpainting.

Multi‐stage Image Inpainting Using Improved Partial Convolutions

Image Inpainting with Contextual Reconstruction Loss.

Dilated Convolution Layers U-Net Generator Conv Feature DeConv Feature Attention Feature DenseNet Discriminator VGG 16 Feature Extrator L 1 Loss Adversarial Loss Perceptual Loss Style Loss Skip Connections

MUSICAL: Multi-Scale Image Contextual Attention Learning for Inpainting.

Image Inpainting Based on Double Joint Predictive Filtering and Wasserstein Generative Adversarial Networks.

DGCA: high resolution image inpainting via DR-GAN and contextual attention

A Progressive and Multi-Prior-Guided Network for Image Inpainting

Generative Image Inpainting Based on Wavelet Transform Attention Model

Semantic Image Inpainting with Multi-Stage Feature Reasoning Generative Adversarial Network

Image Inpainting Based on Contextual Coherent Attention GAN

A Context-Based Multi-Scale Discriminant Model for Natural Image Inpainting

Two-stage structure aware image inpainting based on generative adversarial networks