We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: 3D-LatentMapper: View Agnostic Single-View Reconstruction of 3D Shapes

Abstract: Computer graphics, 3D computer vision and robotics communities have produced multiple approaches to represent and generate 3D shapes, as well as a vast number of use cases. However, single-view reconstruction remains a challenging topic that can unlock various interesting use cases such as interactive design. In this work, we propose a novel framework that leverages the intermediate latent spaces of Vision Transformer (ViT) and a joint image-text representational model, CLIP, for fast and efficient Single View Reconstruction (SVR). More specifically, we propose a novel mapping network architecture that learns a mapping between deep features extracted from ViT and CLIP, and the latent space of a base 3D generative model. Unlike previous work, our method enables view-agnostic reconstruction of 3D shapes, even in the presence of large occlusions. We use the ShapeNetV2 dataset and perform extensive experiments with comparisons to SOTA methods to demonstrate our method's effectiveness.
Comments: Accepted to NeurIPS - WiML workshop 2022
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as: arXiv:2212.02184 [cs.CV]
  (or arXiv:2212.02184v1 [cs.CV] for this version)

Submission history

From: Alara Dirik [view email]
[v1] Mon, 5 Dec 2022 11:45:26 GMT (1747kb,D)

Link back to: arXiv, form interface, contact.