Deep Multi-View Stereo gone wild

Abstract

Deep multi-view stereo (MVS) methods have been developed and extensively compared on simple datasets, where they now outperform classical approaches. In this paper, we ask whether the conclusions reached in controlled scenarios are still valid when working with Internet photo collections. We propose a methodology for evaluation and explore the influence of three aspects of deep MVS methods: network architecture, training data, and supervision. We make several key observations, which we extensively validate quantitatively and qualitatively, both for depth prediction and complete 3D reconstructions. First, complex unsupervised approaches cannot train on data in the wild. Our new approach makes it possible with three key elements: upsampling the output, softmin based aggregation and a single reconstruction loss. Second, supervised deep depthmap-based MVS methods are state-of-the art for reconstruction of few internet images. Finally, our evaluation provides very different results than usual ones. This shows that evaluation in uncontrolled scenarios is important for new architectures.

arXiv commentAccepted to 3DV2021

References in corpus (11)

• What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
• YFCC100M: The New Data in Multimedia Research
• Computing the Stereo Matching Cost with a Convolutional Neural Network
• SurfaceNet: An End-to-end 3D Neural Network for Multiview Stereopsis
• Image Matching across Wide Baselines: From Paper to Practice
• End-to-End Learning of Geometry and Context for Deep Stereo Regression
• Learning Unsupervised Multi-View Stereopsis via Robust Photometric Consistency
• Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
• MVS^2: Deep Unsupervised Multi-view Stereo with Multi-View Symmetry
• M^3VSNet: Unsupervised Multi-metric Multi-view Stereo Network
• Self-supervised Multi-view Stereo via Effective Co-Segmentation and Data-Augmentation