CaSPR: Learning Canonical Spatiotemporal Point Cloud Representations

TL;DR

CaSPR introduces a novel method for learning object-centric, canonical spatiotemporal point cloud representations that enable robust analysis and reconstruction of dynamic objects over time, even with irregular sampling.

Contribution

The paper presents CaSPR, a new approach that supports spacetime continuity, robustness to irregular sampling, and generalization to unseen objects through a two-step encoding and generative process.

Findings

Effective shape reconstruction from sparse data

Accurate camera pose estimation in dynamic scenes

Robust correspondence estimation with irregular samples

Abstract

We propose CaSPR, a method to learn object-centric Canonical Spatiotemporal Point Cloud Representations of dynamically moving or evolving objects. Our goal is to enable information aggregation over time and the interrogation of object state at any spatiotemporal neighborhood in the past, observed or not. Different from previous work, CaSPR learns representations that support spacetime continuity, are robust to variable and irregularly spacetime-sampled point clouds, and generalize to unseen object instances. Our approach divides the problem into two subtasks. First, we explicitly encode time by mapping an input point cloud sequence to a spatiotemporally-canonicalized object space. We then leverage this canonicalization to learn a spatiotemporal latent representation using neural ordinary differential equations and a generative model of dynamically evolving shapes using continuous normalizing flows. We demonstrate the effectiveness of our method on several applications including shape reconstruction, camera pose estimation, continuous spatiotemporal sequence reconstruction, and correspondence estimation from irregularly or intermittently sampled observations.