UVStyle-Net: Unsupervised Few-shot Learning of 3D Style Similarity Measure for B-Reps

Published in

IEEE International Conference on Computer Vision(2021)

External Links:

Generate Graph DownloadPDF

TL;DR

UVStyle-Net is proposed, a style similarity measure for B-Reps that leverages the style signals in the second order statistics of the activations in a pre-trained (unsupervised) 3D encoder, and learns their relative importance to a subjective end-user through few-shot learning.

Abstract

Boundary Representations (B-Reps) are the industry standard in 3D Computer Aided Design/Manufacturing (CAD/CAM) and industrial design due to their fidelity in representing stylistic details. However, they have been ignored in the 3D style research. Existing 3D style metrics typically operate on meshes or point clouds, and fail to account for end-user subjectivity by adopting fixed definitions of style, either through crowd-sourcing for style labels or hand-crafted features. We propose UVStyle-Net, a style similarity measure for B-Reps that leverages the style signals in the second order statistics of the activations in a pre-trained (unsupervised) 3D encoder, and learns their relative importance to a subjective end-user through few-shot learning. Our approach differs from all existing data-driven 3D style methods since it may be used in completely unsupervised settings, which is desirable given the lack of publicly available labeled B-Rep datasets. More importantly, the few-shot learning accounts for the inherent subjectivity associated with style. We show quantitatively that our proposed method with B-Reps is able to capture stronger style signals than alternative methods on meshes and point clouds despite its significantly greater computational efficiency. We also show it is able to generate meaningful style gradients with respect to the input shape, and that few-shot learning with as few as two positive examples selected by an end-user is sufficient to significantly improve the style measure. Finally, we demonstrate its efficacy on a large unlabeled public dataset of CAD models. Source code and data are available at github.com/AutodeskAILab/UVStyle-Net.

Authors

UVStyle-Net: Unsupervised Few-shot Learning of 3D Style Similarity Measure for B-Reps

Published in

IEEE International Conference on Computer Vision(2021)

External Links:

Generate Graph DownloadPDF

TL;DR

Abstract

Authors

J. Lambourne

3 papers

P. Jayaraman

2 papers

Peter Meltzer

1 papers

Hooman Shayani

1 papers

A. Khasahmadi

1 papers

Aditya Sanghi

2 papers

References42 items

3DSNet: Unsupervised Shape-to-Shape 3D Style Transfer

TSIT: A Simple and Versatile Framework for Image-to-Image Translation

Swapping Autoencoder for Deep Image Manipulation

UV-Net: Learning from Curve-Networks and Solids

Neural subdivision

Learning Furniture Compatibility with Graph Neural Networks

TL;DR

Abstract

Authors

TL;DR

Abstract

Authors

References42 items

3DSNet: Unsupervised Shape-to-Shape 3D Style Transfer

TSIT: A Simple and Versatile Framework for Image-to-Image Translation

Swapping Autoencoder for Deep Image Manipulation

UV-Net: Learning from Curve-Networks and Solids

Neural subdivision

Learning Furniture Compatibility with Graph Neural Networks

Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs

PSNet: A Style Transfer Network for Point Cloud Stylization on Geometry and Color

Deep Learning for 3D Point Clouds: A Survey

Analyzing and Improving the Image Quality of StyleGAN

SciPy 1.0: fundamental algorithms for scientific computing in Python

A review on deep learning techniques for 3D sensed data classification

Multimodal Style Transfer via Graph Cuts

ABC: A Big CAD Model Dataset for Geometric Deep Learning

A Style-Based Generator Architecture for Generative Adversarial Networks

Adjustable Real-time Style Transfer

How Powerful are Graph Neural Networks?

A Survey on Deep Learning

A survey on Deep Learning Advances on Different 3D Data Representations

Dynamic Graph CNN for Learning on Point Clouds

Multi-content GAN for Few-Shot Font Style Transfer

3D Object Classification via Spherical Projections

Deep model style: Cross-class style compatibility for 3D furniture within a scene

PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space

Arbitrary Style Transfer in Real-Time with Adaptive Instance Normalization

EMNIST: an extension of MNIST to handwritten letters

Improved Texture Networks: Maximizing Quality and Diversity in Feed-Forward Stylization and Texture Synthesis

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

Geometric Deep Learning: Going beyond Euclidean data

Understanding intermediate layers using linear classifier probes

Image Style Transfer Using Convolutional Neural Networks

Identifying Style of 3D Shapes using Deep Metric Learning

ShapeNet: An Information-Rich 3D Model Repository

Style compatibility for 3D furniture models

Multi-view Convolutional Neural Networks for 3D Shape Recognition

3D ShapeNets: A deep representation for volumetric shapes

Swapping

Partial Entity Structure: A Compact Boundary Representation for Non-Manifold Geometric Modeling

3D Patch-Based Sparse Learning for Style Feature Extraction

MeshCNN: A Network with an Edge

Topological Structures for Geometric Modeling

ements of style : Learning perceptual shape style similarity

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names