Adaptive Graph Convolution for Point Cloud Analysis

TL;DR

This paper introduces Adaptive Graph Convolution (AdaptConv), a novel method for point cloud analysis that dynamically generates kernels based on learned features, enhancing the ability to distinguish features across diverse 3D points.

Contribution

The paper presents AdaptConv, which internally adapts convolution kernels for point clouds, improving over fixed kernels and outperforming existing methods in classification and segmentation tasks.

Findings

Outperforms state-of-the-art methods on benchmark datasets

Effectively captures diverse relations between points

Enhances feature learning in 3D point cloud analysis

Abstract

Convolution on 3D point clouds that generalized from 2D grid-like domains is widely researched yet far from perfect. The standard convolution characterises feature correspondences indistinguishably among 3D points, presenting an intrinsic limitation of poor distinctive feature learning. In this paper, we propose Adaptive Graph Convolution (AdaptConv) which generates adaptive kernels for points according to their dynamically learned features. Compared with using a fixed/isotropic kernel, AdaptConv improves the flexibility of point cloud convolutions, effectively and precisely capturing the diverse relations between points from different semantic parts. Unlike popular attentional weight schemes, the proposed AdaptConv implements the adaptiveness inside the convolution operation instead of simply assigning different weights to the neighboring points. Extensive qualitative and quantitative evaluations show that our method outperforms state-of-the-art point cloud classification and segmentation approaches on several benchmark datasets. Our code is available at https://github.com/hrzhou2/AdaptConv-master.