NeRD++: Improved 3D-mirror symmetry learning from a single image

TL;DR

NeRD++ enhances 3D mirror symmetry learning from a single image by using explicit feature correlation and spherical convolutions, resulting in improved data efficiency and faster inference compared to previous methods.

Contribution

It introduces a novel approach that replaces 4D feature volumes with explicit pixel correlation and employs multi-stage spherical convolutions for better efficiency.

Findings

Achieves faster inference speed.

Requires less annotated data.

Demonstrates improved accuracy on synthetic and real datasets.

Abstract

Many objects are naturally symmetric, and this symmetry can be exploited to infer unseen 3D properties from a single 2D image. Recently, NeRD is proposed for accurate 3D mirror plane estimation from a single image. Despite the unprecedented accuracy, it relies on large annotated datasets for training and suffers from slow inference. Here we aim to improve its data and compute efficiency. We do away with the computationally expensive 4D feature volumes and instead explicitly compute the feature correlation of the pixel correspondences across depth, thus creating a compact 3D volume. We also design multi-stage spherical convolutions to identify the optimal mirror plane on the hemisphere, whose inductive bias offers gains in data-efficiency. Experiments on both synthetic and real-world datasets show the benefit of our proposed changes for improved data efficiency and inference speed.