Unsupervised cycle‐consistent deformation for shape matching

Groueix, Thibault; Fisher, Matthew; Kim, Vladimir G.; Russell, Bryan; Aubry, Mathieu

doi:10.1111/cgf.13794

Cited by 39 publications

(62 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The key idea to use Multi-Layer Perceptron (MLP) on point clouds was initially proposed for shape classification and segmentation by Qi et al [42] with an architecture called PointNet, and for 3D point cloud generation in [15]. To answer the difficulty of annotating 3D data, new approaches are able to perform self-supervised and unsupervised feature learning [60,44,27] and low-shot segmentation [24,55,21]. Especially related to ours is the recent 3D capsule approach [61] that explicitly tries to design a shape representation invariant to 3D transformations and can be applied to many tasks.…”

Section: Related Workmentioning

confidence: 99%

“…With K = 10 prototypes and D = 5, our model has 4.6M parameters. For comparison, the reconstruction models proposed by Wang et al [55] and Groueix et al [24] have respectively 2.6M and 10.0M parameters. See our supplementary material for additional details.…”

Section: Parameterization and Training Detailsmentioning

confidence: 99%

“…With K = 10 prototypes and D = 5, our model has 4.6M parameters. For comparison, the reconstruction models proposed by Wang et al[55] and Groueix et al[24] have respectively 2.6M and 10.0M parameters. Our model can be trained on a single NVIDIA GeForce RTX 2080Ti within a few hours on the 3 991 samples of ModelNet10, and in less than a day on ShapeNet-Core.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Representing Shape Collections with Alignment-Aware Linear Models

Loiseau,

Monnier,

Aubry

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

In this paper, we revisit the classical representation of 3D point clouds as linear shape models. Our key insight is to leverage deep learning to represent a collection of shapes as affine transformations of low-dimensional linear shape models. Each linear model is characterized by a shape prototype, a low-dimensional shape basis and two neural networks. The networks take as input a point cloud and predict the coordinates of a shape in the linear basis and the affine transformation which best approximate the input. Both linear models and neural networks are learned end-to-end using a single reconstruction loss. The main advantage of our approach is that, in contrast to many recent deep approaches which learn feature-based complex shape representations, our model is explicit and every operation occurs in 3D space. As a result, our linear shape models can be easily visualized and annotated, and failure cases can be visually understood. While our main goal is to introduce a compact and interpretable representation of shape collections, we show it leads to state of the art results for few-shot segmentation. Code and data are available at: https:// romainloiseau.github.io/deep-linear-shapes

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Parameterization and Training Detailsmentioning

confidence: 99%

See 1 more Smart Citation

Representing Shape Collections with Alignment-Aware Linear Models

Loiseau,

Monnier,

Aubry

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…To achieve the goals, the key is to choose a suitable parameterization of deformations. One option is to follow the recent target-driven deformation network [38,9,45,34], which parameterizes the deformation as new positions of all the mesh vertices. However, such a large degree of freedom often results in the loss of fine-grained geometric details and tends to cause undesirable distortions.…”

Section: Introductionmentioning

confidence: 99%

DeepMetaHandles: Learning Deformation Meta-Handles of 3D Meshes with Biharmonic Coordinates

Liu¹,

Sung²,

Mĕch³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…These features are then input to an offset decoder which predicts per-vertex offsets to deform the source to produce a deformed shape similar to the target. Groueix et al[277] also perform pervertex deformation, leveraging not only reconstruction loss, but also cycle-consistency loss.…”

mentioning

confidence: 99%

A Revisit of Shape Editing Techniques: from the Geometric to the Neural Viewpoint

Yuan

Lai

et al. 2021

Preprint

View full text Add to dashboard Cite

3D shape editing is widely used in a range of applications such as movie production, computer games and computer aided design. It is also a popular research topic in computer graphics and computer vision. In past decades, researchers have developed a series of editing methods to make the editing process faster, more robust, and more reliable. Traditionally, the deformed shape is determined by the optimal transformation and weights for an energy term. With increasing availability of 3D shapes on the Internet, data-driven methods were proposed to improve the editing results. More recently as the deep neural networks became popular, many deep learning based editing methods have been developed in this field, which is naturally data-driven. We mainly survey recent research works from the geometric viewpoint to those emerging neural deformation techniques and categorize them into organic shape editing methods and man-made model editing methods. Both traditional methods and recent neural network based methods are reviewed.

show abstract

Unsupervised cycle‐consistent deformation for shape matching

Cited by 39 publications

References 38 publications

Representing Shape Collections with Alignment-Aware Linear Models

Representing Shape Collections with Alignment-Aware Linear Models

DeepMetaHandles: Learning Deformation Meta-Handles of 3D Meshes with Biharmonic Coordinates

A Revisit of Shape Editing Techniques: from the Geometric to the Neural Viewpoint

Contact Info

Product

Resources

About