Learning 3D shape completion under weak supervision
From MaRDI portal
Abstract: We address the problem of 3D shape completion from sparse and noisy point clouds, a fundamental problem in computer vision and robotics. Recent approaches are either data-driven or learning-based: Data-driven approaches rely on a shape model whose parameters are optimized to fit the observations; Learning-based approaches, in contrast, avoid the expensive optimization step by learning to directly predict complete shapes from incomplete observations in a fully-supervised setting. However, full supervision is often not available in practice. In this work, we propose a weakly-supervised learning-based approach to 3D shape completion which neither requires slow optimization nor direct supervision. While we also learn a shape prior on synthetic data, we amortize, i.e., learn, maximum likelihood fitting using deep neural networks resulting in efficient shape completion without sacrificing accuracy. On synthetic benchmarks based on ShapeNet and ModelNet as well as on real robotics data from KITTI and Kinect, we demonstrate that the proposed amortized maximum likelihood approach is able to compete with recent fully supervised baselines and outperforms data-driven approaches, while requiring less supervision and being significantly faster.
Recommendations
- Multi-stage point completion network with critical set supervision
- Recovering the shape of objects in 3D point clouds with partial occlusions
- Learning single-image 3D reconstruction by generative modelling of shape, pose and shading
- Inferring 3D shapes from image collections using adversarial networks
- Pix2Shape: towards unsupervised learning of 3D scenes from images using a view-based representation
Cites work
Cited in
(8)- Unsupervised 3D shape segmentation and co-segmentation via deep learning
- Multi-stage point completion network with critical set supervision
- Pix2Shape: towards unsupervised learning of 3D scenes from images using a view-based representation
- Inferring 3D shapes from image collections using adversarial networks
- Temporally coherent general dynamic scene reconstruction
- Robust attentional aggregation of deep feature sets for multi-view 3D reconstruction
- 3D grasp saliency analysis via deep shape correspondence
- Recovering the shape of objects in 3D point clouds with partial occlusions
Describes a project that uses
Uses Software
This page was built for publication: Learning 3D shape completion under weak supervision
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2226180)
