Residual Attention Graph Convolutional Network for Geometric 3D Scene Classification

TL;DR

This paper introduces a Residual Attention Graph Convolutional Network that leverages intrinsic 3D geometric context for scene classification, outperforming existing methods without relying on point features.

Contribution

The work presents a novel graph convolutional network that exploits 3D geometric context directly, avoiding the limitations of depth-only features in scene classification.

Findings

Outperforms state-of-the-art methods on NYU Depth v1 and SUN-RGBD datasets.

Effectively exploits intrinsic 3D geometric information.

Works with both organized and unorganized 3D data.

Abstract

Geometric 3D scene classification is a very challenging task. Current methodologies extract the geometric information using only a depth channel provided by an RGB-D sensor. These kinds of methodologies introduce possible errors due to missing local geometric context in the depth channel. This work proposes a novel Residual Attention Graph Convolutional Network that exploits the intrinsic geometric context inside a 3D space without using any kind of point features, allowing the use of organized or unorganized 3D data. Experiments are done in NYU Depth v1 and SUN-RGBD datasets to study the different configurations and to demonstrate the effectiveness of the proposed method. Experimental results show that the proposed method outperforms current state-of-the-art in geometric 3D scene classification tasks.