3DIAS: 3D Shape Reconstruction with Implicit Algebraic Surfaces

Abstract

3D Shape representation has substantial effects on 3D shape reconstruction. Primitive-based representations approximate 3D shape mainly by a set of simple implicit primitives, but the low flexibility of the primitives limit their outputs resolutions. Moreover, setting a sufficient number of primitives for an arbitrary shape is challenging. To overcome these issues, we propose ...

Authors

Mohsen Yavartanoo*

Jaeyoung Chung*

Reyhaneh Neshatavar

Kyoung Mu Lee

Acknowledgement

This work was supported in part by an IITP grant funded by the Korean government [No. 2021-0-01343, Artificial Intelligence Graduate School Program (Seoul National University)].

You can freely acces to the PDF, code, poster, and check the citations via the following links.

Presentation

Watch the video of our presentation in ICCV2021.

We propose a constrained implicit algebraic surface as the primitive with few learnable coefficients and higher geometrical complexities and a deep neural network to produce these primitives.


Whatch more videos on our YouTube channel.

3DIAS poseter presentation.

The remaining time to our presentation in ICCV2021.

00
Days
00
Hours
00
Minutes
00
Seconds

Exprimental Results

Qualitative Comaprision

Qualitative comparison on single RGB image 3D shape reconstruction. SIF, AtlasNet, OccNet, CvxNet, and our 3DIAS output reconstructed 3D shape from the given RGB image. Comparison with other methods for the samples shown in CvxNet.

Complexity of primitives

Qualitative results on unsupervised semantic segmentation. We visualize the results of 3DIAS for some samples in the category of airplane.

Unsupervised Semantics Segmentation

The complexity of our primitives. The first and the second rows show the reconstructed shapes and their corresponding primitives for two samples. The proposed primitive can effectively present curved and torus shapes.

Authors