3D Surface Reconstruction by Pointillism

TL;DR

This paper presents a novel deep learning approach for 3D surface reconstruction from a single image, leveraging a new loss function and multi-view correspondences to accurately infer depth maps, demonstrated on sculptures with good generalization.

Contribution

Introduces a new loss function utilizing image correspondences and a pipeline for automatic multi-view data generation for training a deep network.

Findings

Accurately infers depth maps from single images of sculptures.

The method generalizes well to synthetic images and new domains.

Demonstrates effective 3D shape reconstruction from limited data.

Abstract

The objective of this work is to infer the 3D shape of an object from a single image. We use sculptures as our training and test bed, as these have great variety in shape and appearance. To achieve this we build on the success of multiple view geometry (MVG) which is able to accurately provide correspondences between images of 3D objects under varying viewpoint and illumination conditions, and make the following contributions: first, we introduce a new loss function that can harness image-to-image correspondences to provide a supervisory signal to train a deep network to infer a depth map. The network is trained end-to-end by differentiating through the camera. Second, we develop a processing pipeline to automatically generate a large scale multi-view set of correspondences for training the network. Finally, we demonstrate that we can indeed obtain a depth map of a novel object from a single image for a variety of sculptures with varying shape/texture, and that the network generalises at test time to new domains (e.g. synthetic images).