Abstract:In the past decade, deep neural networks have achieved significant progress in point cloud learning. However, collecting large-scale precisely-annotated training data is extremely laborious and expensive, which hinders the scalability of existing point cloud datasets and poses a bottleneck for efficient exploration of point cloud data in various tasks and applications. Label-efficient learning offers a promising solution by enabling effective deep network training with much-reduced annotation efforts. This paper presents the first comprehensive survey of label-efficient learning of point clouds. We address three critical questions in this emerging research field: i) the importance and urgency of label-efficient learning in point cloud processing, ii) the subfields it encompasses, and iii) the progress achieved in this area. To achieve this, we propose a taxonomy that organizes label-efficient learning methods based on the data prerequisites provided by different types of labels. We categorize four typical label-efficient learning approaches that significantly reduce point cloud annotation efforts: data augmentation, domain transfer learning, weakly-supervised learning, and pretrained foundation models. For each approach, we outline the problem setup and provide an extensive literature review that showcases relevant progress and challenges. Finally, we share insights into current research challenges and potential future directions. A project associated with this survey has been built at <a class="link-external link-https" href="https://github.com/xiaoaoran/3D_label_efficient_learning" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? The paper "A Survey of Label-Efficient Deep Learning for 3D Point Clouds" aims to address the difficulty and expense of annotating 3D point cloud data. Specifically, the paper focuses on the following key issues: 1. **Importance of Label Efficiency**: Collecting large-scale, accurately annotated 3D point cloud data is very labor-intensive and costly, which limits the scalability of existing 3D point cloud datasets and becomes a bottleneck for efficiently exploring point cloud data in various tasks and applications. 2. **Subfields of Label-Efficient Learning**: The paper explores different subfields of label-efficient learning in point cloud processing, including data augmentation, domain transfer learning, weakly supervised learning, and pre-trained foundational models. 3. **Progress and Challenges**: The paper systematically reviews the latest progress in these subfields and discusses the challenges and future directions in current research. ### Main Contributions of the Paper 1. **Proposed Classification System**: The paper proposes a classification system for organizing label-efficient learning methods based on different types of label data prerequisites. 2. **Detailed Literature Review**: A detailed literature review of each label-efficient learning method is provided, showcasing related progress and challenges. 3. **Future Research Directions**: The paper discusses the limitations and challenges of current research and proposes several promising directions for future research. ### Background and Motivation - **Challenges of 3D Point Cloud Annotation**: 3D point cloud data is often incomplete, sparse, and lacks color information, leading to semantic and geometric ambiguities. Additionally, variations in 3D views increase the complexity of annotation. - **Limitations of Existing Datasets**: Existing public datasets are limited in scale and diversity because the annotation process requires significant time and human effort. - **Necessity of Label-Efficient Learning**: To overcome the limitations of existing 3D point cloud data, label-efficient learning becomes crucial and urgent. ### Main Content 1. **Data Augmentation**: Increasing the amount and diversity of data by generating new training data, thereby improving the model's generalization ability. 2. **Domain Transfer Learning**: Using annotated source domain data to train models that perform well on unannotated target domain data. 3. **Weakly Supervised Learning**: Training robust models using weakly annotated point cloud data. 4. **Pre-trained Foundational Models**: Reducing annotation requirements by leveraging unsupervised or multi-modal pre-training to facilitate 3D modeling. ### Conclusion Through a systematic review and classification, the paper provides researchers and practitioners with a comprehensive overview of label-efficient learning for 3D point clouds, filling gaps in the existing literature and pointing out directions for future research.

A Survey of Label-Efficient Deep Learning for 3D Point Clouds

From Virtuality To Reality: A Learning-based Point Cloud Labeling Method With Synthesis Scene

Towards Open-Set Semantic Labeling in 3D Point Clouds : Analysis on the Unknown Class.

Unsupervised Point Cloud Representation Learning with Deep Neural Networks: A Survey

Deep Learning for 3D Point Clouds: A Survey

Deep Learning for 3D Point Cloud Enhancement: A Survey

Deep Learning for 3D Point Cloud Understanding: A Survey

Self-supervised Learning for Pre-Training 3D Point Clouds: A Survey

A Deep Active Learning Framework for Semantic Labeling of Point Clouds

labelCloud: A Lightweight Domain-Independent Labeling Tool for 3D Object Detection in Point Clouds

Self-Supervised Learning of 3D Point Clouds: A Survey

Self-Supervised Learning for Point Clouds Data: A Survey

Deep Learning on Point Clouds and Its Application: A Survey

Deep Learning-based 3D Point Cloud Classification: A Systematic Survey and Outlook

Label-Efficient Learning on Point Clouds using Approximate Convex Decompositions

Superpoint-guided Semi-supervised Semantic Segmentation of 3D Point Clouds

Label Propagation from ImageNet to 3D Point Clouds

A Deep Learning Network for Semantic Labeling of Large-Scale Urban Point Clouds

Self-Supervised Deep Learning on Point Clouds by Reconstructing Space