A Cross Branch Fusion-Based Contrastive Learning Framework for Point Cloud Self-supervised Learning

TL;DR

This paper introduces PoCCA, a novel contrastive learning framework for point cloud data that enables information exchange between branches before the loss, leading to improved self-supervised 3D representations without extra data.

Contribution

PoCCA is the first framework to incorporate cross-branch information exchange in point cloud contrastive learning, enhancing representation quality in a self-supervised manner.

Findings

Achieves state-of-the-art performance on downstream point cloud tasks.

Effectively learns rich 3D representations without additional training data.

Outperforms existing contrastive learning methods for point clouds.

Abstract

Contrastive learning is an essential method in self-supervised learning. It primarily employs a multi-branch strategy to compare latent representations obtained from different branches and train the encoder. In the case of multi-modal input, diverse modalities of the same object are fed into distinct branches. When using single-modal data, the same input undergoes various augmentations before being fed into different branches. However, all existing contrastive learning frameworks have so far only performed contrastive operations on the learned features at the final loss end, with no information exchange between different branches prior to this stage. In this paper, for point cloud unsupervised learning without the use of extra training data, we propose a Contrastive Cross-branch Attention-based framework for Point cloud data (termed PoCCA), to learn rich 3D point cloud representations. By introducing sub-branches, PoCCA allows information exchange between different branches before the loss end. Experimental results demonstrate that in the case of using no extra training data, the representations learned with our self-supervised model achieve state-of-the-art performances when used for downstream tasks on point clouds.