KPConvX: Modernizing Kernel Point Convolution with Kernel Attention

TL;DR

KPConvX introduces kernel attention to enhance kernel point convolution, enabling deeper architectures and outperforming state-of-the-art methods in 3D point cloud classification and segmentation tasks.

Contribution

The paper proposes KPConvX, a novel kernel attention mechanism for kernel point convolution, improving scalability and performance over existing KPConv models.

Findings

Outperforms state-of-the-art on ScanObjectNN, Scannetv2, S3DIS datasets.

Enables deeper architectures with KPConvD and kernel attention.

Validated through extensive ablation studies.

Abstract

In the field of deep point cloud understanding, KPConv is a unique architecture that uses kernel points to locate convolutional weights in space, instead of relying on Multi-Layer Perceptron (MLP) encodings. While it initially achieved success, it has since been surpassed by recent MLP networks that employ updated designs and training strategies. Building upon the kernel point principle, we present two novel designs: KPConvD (depthwise KPConv), a lighter design that enables the use of deeper architectures, and KPConvX, an innovative design that scales the depthwise convolutional weights of KPConvD with kernel attention values. Using KPConvX with a modern architecture and training strategy, we are able to outperform current state-of-the-art approaches on the ScanObjectNN, Scannetv2, and S3DIS datasets. We validate our design choices through ablation studies and release our code and models.