Abstract:Point cloud data plays an essential role in robotics and self-driving applications. Yet, annotating point cloud data is time-consuming and nontrivial while they enable learning discriminative 3D representations that empower downstream tasks, such as classification and segmentation. Recently, contrastive learning-based frameworks have shown promising results for learning 3D representations in a self-supervised manner. However, existing contrastive learning methods cannot precisely encode and associate structural features and search the higher dimensional augmentation space efficiently. In this paper, we present CLR-GAM, a novel contrastive learning-based framework with Guided Augmentation (GA) for efficient dynamic exploration strategy and Guided Feature Mapping (GFM) for similar structural feature association between augmented point clouds. We empirically demonstrate that the proposed approach achieves state-of-the-art performance on both simulated and real-world 3D point cloud datasets for three different downstream tasks, i.e., 3D point cloud classification, few-shot learning, and object part segmentation.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are two limitations of existing contrastive learning methods when dealing with 3D point cloud data: 1. **Contrast Ambiguity**: - **Global Contrast Ambiguity (GCA)**: Through augmentation operations such as cropping and non - rigid transformation, the shape of the augmented object may be completely different from the original object, leading to confusion in contrastive learning. For example, if the back part is removed from a "chair" point cloud, the result may be similar to a sample in the "table" category. - **Local Contrast Ambiguity (LCA)**: Local feature contrast techniques regard the features of other points in the same point cloud as negative samples, but due to symmetry and similar shapes within the object, these features may actually be similar. 2. **Curse of Dimensionality**: - Contrastive learning requires multiple augmentation operations to learn discriminative 3D point cloud representations, but exploring the high - dimensional augmentation space is time - consuming and it is difficult to ensure coverage. Existing methods usually adopt random sampling, which leads to ineffective augmentation and high computational cost. To solve these problems, the authors propose a new contrastive learning framework **CLR - GAM**, which contains two main modules: 1. **Guided Feature Mapping (GFM)**: - It is used to correlate the features of the same structure between two augmented point clouds, and can effectively correlate features even in the case of severely deformed shapes. - By tightly coupling local feature correlation and global feature contrast, it improves the accuracy of feature representation. 2. **Guided Augmentation (GA)**: - It efficiently explores the high - dimensional augmentation space through a dynamic exploration strategy, ensuring that a diverse range of augmentation operations can be covered with a limited number of samples. - It uses a memory bank to save the explored augmented samples and selects novel augmentation operations by calculating the distance between new samples and existing samples. Through these two modules, CLR - GAM can more effectively learn discriminative 3D point cloud representations and has achieved state - of - the - art performance in multiple downstream tasks, including 3D point cloud classification, few - shot learning, and object part segmentation.

CLR-GAM: Contrastive Point Cloud Learning with Guided Augmentation and Feature Mapping

Point Contrastive Prediction with Semantic Clustering for Self-Supervised Learning on Point Cloud Videos

Point Cloud Understanding via Attention-Driven Contrastive Learning

SegContrast: 3D Point Cloud Feature Representation Learning Through Self-Supervised Segment Discrimination

GAM : Gradient Attention Module of Optimization for Point Clouds Analysis

Synergizing Contrastive Learning and Optimal Transport for 3D Point Cloud Domain Adaptation

Dcpoint: Global-Local Dual Contrast for Self-Supervised Representation Learning of 3d Point Clouds

Generalized 3D Self-supervised Learning Framework via Prompted Foreground-Aware Feature Contrast

Masked Scene Contrast: A Scalable Framework for Unsupervised 3D Representation Learning

ToThePoint: Efficient Contrastive Learning of 3D Point Clouds Via Recycling

Bringing Masked Autoencoders Explicit Contrastive Properties for Point Cloud Self-Supervised Learning

PointALCR: adversarial latent GAN and contrastive regularization for point cloud completion

EPContrast: Effective Point-level Contrastive Learning for Large-scale Point Cloud Understanding

GroupContrast: Semantic-aware Self-supervised Representation Learning for 3D Understanding

TriCI: Triple Cross-Intra Branch Contrastive Learning for Point Cloud Analysis

Self-Supervised Intra-Modal and Cross-Modal Contrastive Learning for Point Cloud Understanding

A Category-Contrastive Guided-Graph Convolutional Network Approach for the Semantic Segmentation of Point Clouds

Point‐AGM : Attention Guided Masked Auto‐Encoder for Joint Self‐supervised Learning on Point Clouds

PointCG: Self-supervised Point Cloud Learning via Joint Completion and Generation

3-D Object Detection With Balanced Prediction Based on Contrastive Point Loss

Unsupervised contrastive learning with simple transformation for 3D point cloud data