Neural Rendering of object-based audio-visual scenes

这项研究将研究使用神经绘制方法从基于对象的表示中渲染具有视频真实感的一般动态场景。我们将通过深层网络架构(GAN、VAE等)开发自然空间和时间外观的习得表示。用于场景的实时渲染。其目标是能够从视频捕获得到的基于对象的场景表示中为自然场景再现高度逼真的内容。

This research will investigate the use of neural rendering methods to render general dynamic scenes from object-based representations with the realism of video. Learnt representations of natural spatial and temporal appearance will be exploited through deep network architectures (GAN, VAE etc.) for real time rendering of scenes. The goal is to enable the reproduction of highly-realistic content for natural scenes from object based scene representations derived from video capture.