Learning Geometrically Consistent Mesh Corrections
This work addresses the high cost of building good 3D reconstructions for applications like mapping or scanning, though it is incremental as it builds on existing correction methods.
The paper tackles the problem of excessive smoothing and geometric inconsistencies in post-hoc correction of low-quality 3D meshes, resulting in a model that reduces gross errors by 45.3% to 77.5%, up to five times more than previous work.
Building good 3D maps is a challenging and expensive task, which requires high-quality sensors and careful, time-consuming scanning. We seek to reduce the cost of building good reconstructions by correcting views of existing low-quality ones in a post-hoc fashion using learnt priors over surfaces and appearance. We train a CNN model to predict the difference in inverse-depth from varying viewpoints of two meshes -- one of low quality that we wish to correct, and one of high-quality that we use as a reference. In contrast to previous work, we pay attention to the problem of excessive smoothing in corrected meshes. We address this with a suitable network architecture, and introduce a loss-weighting mechanism that emphasises edges in the prediction. Furthermore, smooth predictions result in geometrical inconsistencies. To deal with this issue, we present a loss function which penalises re-projection differences that are not due to occlusions. Our model reduces gross errors by 45.3%--77.5%, up to five times more than previous work.