Interpolation-Aware Padding for 3D Sparse Convolutional Neural Networks

TL;DR

This paper introduces an interpolation-aware padding scheme for 3D sparse CNNs that improves the accuracy of point-wise features in fine-grained 3D vision tasks by ensuring neighboring voxels exist during convolution.

Contribution

The paper proposes a novel padding method that enhances 3D sparse CNN performance by involving adjacent empty voxels in computations, improving point-wise feature accuracy.

Findings

Higher prediction accuracy in semantic segmentation and 3D detection tasks.

Superiority over existing padding schemes like zero-padding and octree-padding.

Effective in conjunction with feature interpolation methods.

Abstract

Sparse voxel-based 3D convolutional neural networks (CNNs) are widely used for various 3D vision tasks. Sparse voxel-based 3D CNNs create sparse non-empty voxels from the 3D input and perform 3D convolution operations on them only. We propose a simple yet effective padding scheme --- interpolation-aware padding to pad a few empty voxels adjacent to the non-empty voxels and involve them in the 3D CNN computation so that all neighboring voxels exist when computing point-wise features via the trilinear interpolation. For fine-grained 3D vision tasks where point-wise features are essential, like semantic segmentation and 3D detection, our network achieves higher prediction accuracy than the existing networks using the nearest neighbor interpolation or the normalized trilinear interpolation with the zero-padding or the octree-padding scheme. Through extensive comparisons on various 3D segmentation and detection tasks, we demonstrate the superiority of 3D sparse CNNs with our padding scheme in conjunction with feature interpolation.