Neural Non-Rigid Tracking

TL;DR

This paper presents a novel end-to-end learnable non-rigid tracking method that improves reconstruction accuracy and speed by predicting dense correspondences with a neural network and integrating them into a differentiable optimization framework.

Contribution

The authors introduce a differentiable, end-to-end trainable non-rigid tracker that predicts dense correspondences and confidences, enabling robust optimization for non-rigid reconstruction.

Findings

Achieves improved reconstruction performance over state-of-the-art methods.

Predicts correspondences 85 times faster than comparable deep-learning approaches.

Automatically down-weights outliers and wrong correspondences through learned confidence estimation.

Abstract

We introduce a novel, end-to-end learnable, differentiable non-rigid tracker that enables state-of-the-art non-rigid reconstruction by a learned robust optimization. Given two input RGB-D frames of a non-rigidly moving object, we employ a convolutional neural network to predict dense correspondences and their confidences. These correspondences are used as constraints in an as-rigid-as-possible (ARAP) optimization problem. By enabling gradient back-propagation through the weighted non-linear least squares solver, we are able to learn correspondences and confidences in an end-to-end manner such that they are optimal for the task of non-rigid tracking. Under this formulation, correspondence confidences can be learned via self-supervision, informing a learned robust optimization, where outliers and wrong correspondences are automatically down-weighted to enable effective tracking. Compared to state-of-the-art approaches, our algorithm shows improved reconstruction performance, while simultaneously achieving 85 times faster correspondence prediction than comparable deep-learning based methods. We make our code available.