Structure Learning for 3D Point Cloud Generation from Single RGB Images

dc.contributor.authorCharrada, Tarek Benen_US
dc.contributor.authorLaga, Hamiden_US
dc.contributor.authorTabia, Hedien_US
dc.contributor.editorChaine, Raphaëlleen_US
dc.contributor.editorDeng, Zhigangen_US
dc.contributor.editorKim, Min H.en_US
dc.date.accessioned2023-10-09T07:34:14Z
dc.date.available2023-10-09T07:34:14Z
dc.date.issued2023
dc.description.abstract3D point clouds can represent complex 3D objects of arbitrary topologies and with fine-grained details. They are, however, hard to regress from images using convolutional neural networks, making tasks such as 3D reconstruction from monocular RGB images challenging. In fact, unlike images and volumetric grids, point clouds are unstructured and thus lack proper parameterization, which makes them difficult to process using convolutional operations. Existing point-based 3D reconstruction methods that tried to address this problem rely on complex end-to-end architectures with high computational costs. Instead, we propose in this paper a novel mechanism that decouples the 3D reconstruction problem from the structure (or parameterization) learning task, making the 3D reconstruction of objects of arbitrary topologies tractable and thus easier to learn. We achieve this using a novel Teacher-Student network where the Teacher learns to structure the point clouds. The Student then harnesses the knowledge learned by the Teacher to efficiently regress accurate 3D point clouds. We train the Teacher network using 3D ground-truth supervision and the Student network using the Teacher’'s annotations. Finally, we employ a novel refinement network to overcome the upper-bound performance that is set by the Teacher network. Our extensive experiments on ShapeNet and Pix3D benchmarks, and on in-the-wild images demonstrate that the proposed approach outperforms previous methods in terms of reconstruction accuracy and visual quality.en_US
dc.description.number7
dc.description.sectionheadersModeling by Learning
dc.description.seriesinformationComputer Graphics Forum
dc.description.volume42
dc.identifier.doi10.1111/cgf.14942
dc.identifier.issn1467-8659
dc.identifier.pages11 pages
dc.identifier.urihttps://doi.org/10.1111/cgf.14942
dc.identifier.urihttps://diglib.eg.org:443/handle/10.1111/cgf14942
dc.publisherThe Eurographics Association and John Wiley & Sons Ltd.en_US
dc.subjectKeywords: CNN, Deep learning, 3D parameterization CCS Concepts: Computing methodologies -> Point-based models
dc.subjectCNN
dc.subjectDeep learning
dc.subject3D parameterization CCS Concepts
dc.subjectComputing methodologies
dc.subjectPoint
dc.subjectbased models
dc.titleStructure Learning for 3D Point Cloud Generation from Single RGB Imagesen_US
Files
Original bundle
Now showing 1 - 2 of 2
No Thumbnail Available
Name:
v42i7_14_14942.pdf
Size:
6.88 MB
Format:
Adobe Portable Document Format
No Thumbnail Available
Name:
paper1189_mm.pdf
Size:
10.35 MB
Format:
Adobe Portable Document Format
Collections