利用物体共现关系完成场景的 Halluci-Net

Apr, 2020

利用物体共现关系完成场景的 Halluci-Net

Halluci-Net: Scene Completion by Exploiting Object Co-occurrence Relationships

Kuldeep Kulkarni, Tejas Gokhale, Rajhans Singh, Pavan Turaga, Aswin Sankaranarayanan

TL;DR本文提出了一种基于 `Halluci-Net` 的两阶段深度网络方法，它学习了场景中物体之间的共现关系，并利用这些关系产生了密集和完整的标签映射，该方法在 Cityscapes 数据集上获得了比两个基线方法更好的性能指标，包括 Fréchet Inception Distance（FID）、语义分割准确性以及物体共现相似性，并展示了 ADE20K 数据集子集（包含卧室图像）的定性结果。

Abstract

We address the new problem of complex scene completion from sparse label maps. We use a two-stage deep network based method, called `halluci-net', that uses object co-occurrence relationships to produce a dense and complete label map. The generated dense label map is fed into a state-o