skip to main content

This content will become publicly available on December 5, 2024

Title: GeoLatent: A Geometric Approach to Latent Space Design for Deformable Shape Generators

We study how to optimize the latent space of neural shape generators that map latent codes to 3D deformable shapes. The key focus is to look at a deformable shape generator from a differential geometry perspective. We define a Riemannian metric based on as-rigid-as-possible and as-conformal-as-possible deformation energies. Under this metric, we study two desired properties of the latent space: 1) straight-line interpolations in latent codes follow geodesic curves; 2) latent codes disentangle pose and shape variations at different scales. Strictly enforcing the geometric interpolation property, however, only applies if the metric matrix is a constant. We show how to achieve this property approximately by enforcing that geodesic interpolations are axis-aligned, i.e., interpolations along coordinate axis follow geodesic curves. In addition, we introduce a novel approach that decouples pose and shape variations via generalized eigendecomposition. We also study efficient regularization terms for learning deformable shape generators, e.g., that promote smooth interpolations. Experimental results on benchmark datasets show that our approach leads to interpretable latent codes, improves the generalizability of synthetic shapes, and enhances performance in geodesic interpolation and geodesic shooting.

more » « less
Award ID(s):
Author(s) / Creator(s):
; ; ; ;
Publisher / Repository:
Date Published:
Journal Name:
ACM Transactions on Graphics
Page Range / eLocation ID:
1 to 20
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. We present Basis Restricted Elastic Shape Analysis (BaRe-ESA), a novel Riemannian framework for human body scan representation, interpolation and extrapolation. BaRe-ESA operates directly on unregistered meshes, i.e., without the need to establish prior point to point correspondences or to assume a consistent mesh structure. Our method relies on a latent space representation, which is equipped with a Riemannian (non-Euclidean) metric associated to an invariant higher-order metric on the space of surfaces. Experimental results on the FAUST and DFAUST datasets show that BaRe-ESA brings significant improvements with respect to previous solutions in terms of shape registration, interpolation and extrapolation. The efficiency and strength of our model is further demonstrated in applications such as motion transfer and random generation of body shape and pose. 
    more » « less
  2. The problem of using covariates to predict shapes of objects in a regression setting is important in many fields. A formal statistical approach, termed Geodesic regression model, is commonly used for modeling and analyzing relationships between Euclidean predictors and shape responses. Despite its popularity, this model faces several key challenges, including (i) misalignment of shapes due to pre-processing steps, (ii) difficulties in shape alignment due to imaging heterogeneity, and (iii) lack of spatial correlation in shape structures. This paper proposes a comprehensive geodesic factor regression model that addresses all these challenges. Instead of using shapes as extracted from pre-registered data, it takes a more fundamental approach, incorporating alignment step within the proposed regression model and learns them using both pre-shape and covariate data. Additionally, it specifies spatial correlation structures using low-dimensional representations, including latent factors on the tangent space and isotropic error terms. The proposed framework results in substantial improvements in regression performance, as demonstrated through simulation studies and a real data analysis on Corpus Callosum contour data obtained from the ADNI study. 
    more » « less
  3. Abstract

    We investigate the geometry of the space of immersed closed curves equipped with reparametrization-invariant Riemannian metrics; the metrics we consider are Sobolev metrics of possible fractional-order$$q\in [0,\infty )$$q[0,). We establish the critical Sobolev index on the metric for several key geometric properties. Our first main result shows that the Riemannian metric induces a metric space structure if and only if$$q>1/2$$q>1/2. Our second main result shows that the metric is geodesically complete (i.e., the geodesic equation is globally well posed) if$$q>3/2$$q>3/2, whereas if$$q<3/2$$q<3/2then finite-time blowup may occur. The geodesic completeness for$$q>3/2$$q>3/2is obtained by proving metric completeness of the space of$$H^q$$Hq-immersed curves with the distance induced by the Riemannian metric.

    more » « less
  4. Abstract Many natural real-valued functions of closed curves are known to extend continuously to the larger space of geodesic currents. For instance, the extension of length with respect to a fixed hyperbolic metric was a motivating example for the development of geodesic currents. We give a simple criterion on a curve function that guarantees a continuous extension to geodesic currents. The main condition of our criterion is the smoothing property, which has played a role in the study of systoles of translation lengths for Anosov representations. It is easy to see that our criterion is satisfied for almost all known examples of continuous functions on geodesic currents, such as nonpositively curved lengths or stable lengths for surface groups, while also applying to new examples like extremal length. We use this extension to obtain a new curve counting result for extremal length. 
    more » « less
  5. Summary

    Different methods of measuring cavitation resistance in fern petioles lead to variable results, particularly with respect to the P50metric. We hypothesised that the fern dictyostele structure affects air entry into the xylem, and therefore impacts the shape of the vulnerability curve.

    Our study examined this variation by comparing vulnerability curves constructed on petioles collected from evergreen and deciduous ferns in the field, with curves generated using the standard centrifuge, air‐injection and bench‐top dehydration methods. Additional experiments complemented the vulnerability curves to better understand how anatomy shapes estimates of cavitation resistance.

    Centrifugation and radial air injection generated acceptable vulnerability curves for the deciduous species, but overestimated drought resistance in the two evergreen ferns. In these hardy plants, axial air injection and bench‐top dehydration produced results that most closely aligned with observations in nature. Additional experiments revealed that the dictyostele anatomy impedes air entry into the xylem during spinning and radial air injection.

    Each method produced acceptable vulnerability curves, depending on the species being tested. Therefore, we stress the importance of validating the curves within situmeasures of water potential and, if possible, hydraulic data to generate realistic results with any of the methods currently available.

    more » « less