SeRP: Self-Supervised Representation Learning Using Perturbed Point Clouds

TL;DR

SeRP introduces a self-supervised learning framework for 3D point clouds using an encoder-decoder architecture with perturbation reconstruction, improving downstream classification accuracy and addressing limitations of Transformer-based autoencoders.

Contribution

The paper proposes SeRP, a novel self-supervised learning method for 3D point clouds that enhances representation quality and introduces VASP for discrete representation learning.

Findings

Pretrained models outperform from-scratch models by 0.5-1% in classification accuracy.

SeRP effectively learns high-level features from perturbed point clouds.

VASP employs vector-quantization for improved Transformer autoencoder representations.

Abstract

We present SeRP, a framework for Self-Supervised Learning of 3D point clouds. SeRP consists of encoder-decoder architecture that takes perturbed or corrupted point clouds as inputs and aims to reconstruct the original point cloud without corruption. The encoder learns the high-level latent representations of the points clouds in a low-dimensional subspace and recovers the original structure. In this work, we have used Transformers and PointNet-based Autoencoders. The proposed framework also addresses some of the limitations of Transformers-based Masked Autoencoders which are prone to leakage of location information and uneven information density. We trained our models on the complete ShapeNet dataset and evaluated them on ModelNet40 as a downstream classification task. We have shown that the pretrained models achieved 0.5-1% higher classification accuracies than the networks trained from scratch. Furthermore, we also proposed VASP: Vector-Quantized Autoencoder for Self-supervised Representation Learning for Point Clouds that employs Vector-Quantization for discrete representation learning for Transformer-based autoencoders.