Pushing the Boundaries of View Extrapolation with Multiplane Images
It addresses the problem of limited view range and artifacts in computer vision for applications like VR/AR, representing an incremental improvement over existing multiplane image methods.
The paper tackles view synthesis from narrow baseline images by generating high-quality extrapolations with plausible disocclusions, achieving up to 4x the lateral viewpoint movement allowed by prior work.
We explore the problem of view synthesis from a narrow baseline pair of images, and focus on generating high-quality view extrapolations with plausible disocclusions. Our method builds upon prior work in predicting a multiplane image (MPI), which represents scene content as a set of RGB$α$ planes within a reference view frustum and renders novel views by projecting this content into the target viewpoints. We present a theoretical analysis showing how the range of views that can be rendered from an MPI increases linearly with the MPI disparity sampling frequency, as well as a novel MPI prediction procedure that theoretically enables view extrapolations of up to $4\times$ the lateral viewpoint movement allowed by prior work. Our method ameliorates two specific issues that limit the range of views renderable by prior methods: 1) We expand the range of novel views that can be rendered without depth discretization artifacts by using a 3D convolutional network architecture along with a randomized-resolution training procedure to allow our model to predict MPIs with increased disparity sampling frequency. 2) We reduce the repeated texture artifacts seen in disocclusions by enforcing a constraint that the appearance of hidden content at any depth must be drawn from visible content at or behind that depth. Please see our results video at: https://www.youtube.com/watch?v=aJqAaMNL2m4.