SaiNet: Stereo aware inpainting behind objects with generative networks
SaiNet: Stereo aware inpainting behind objects with generative networks
复制标题
DOI：
10.48550/arxiv.2205.07014
复制
发表时间：
2022-05
期刊：
影响因子：
--
通讯作者：
Violeta Men'endez Gonz'alez;Andrew Gilbert;Graeme Phillipson;Stephen Jolly;Simon Hadfield
中科院分区：
文献类型：
--
作者：
Violeta Men'endez Gonz'alez;Andrew Gilbert;Graeme Phillipson;Stephen Jolly;Simon Hadfield
在这项工作中，我们提出了一个端到端的网络立体一致的图像修复与修复对象背后的大缺失区域的目标。该模型由一个边缘引导的UNet网络使用部分卷积。我们通过引入视差损失来增强多视图立体一致性。更重要的是，我们开发了一个训练方案，其中模型是从代表对象遮挡的真实立体遮罩中学习的，而不是更常见的随机遮罩。这项技术是在监督下训练的。我们的评估显示，与以前的最先进的技术相比，具有竞争力的结果。
In this work, we present an end-to-end network for stereo-consistent image inpainting with the objective of inpainting large missing regions behind objects. The proposed model consists of an edge-guided UNet-like network using Partial Convolutions. We enforce multi-view stereo consistency by introducing a disparity loss. More importantly, we develop a training scheme where the model is learned from realistic stereo masks representing object occlusions, instead of the more common random masks. The technique is trained in a supervised way. Our evaluation shows competitive results compared to previous state-of-the-art techniques.