Journal
2021 INTERNATIONAL CONFERENCE ON 3D VISION (3DV 2021)
Volume -, Issue -, Pages 484-493Publisher
IEEE COMPUTER SOC
DOI: 10.1109/3DV53792.2021.00058
Keywords
-
Categories
Funding
- ANR project EnHerit [ANR-17-CE23-0008, 2020-AD011011756]
- Agence Nationale de la Recherche (ANR) [ANR-17-CE23-0008] Funding Source: Agence Nationale de la Recherche (ANR)
Ask authors/readers for more resources
This study examines the effectiveness of deep multi-view stereo methods when utilizing internet photo collections. Results show issues with unsupervised training on wild data, but supervised deep depthmap-based MVS methods perform exceptionally well in reconstructing a small number of internet images.
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.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available