HybridCR: Weakly-Supervised 3D Point Cloud Semantic Segmentation via Hybrid Contrastive Regularization

Li, Mengtian; Xie, Yuan; Shen, Yunhang; Ke, Bo; Qiao, Ruizhi; Ren, Bo; Lin, S. P.; Ma, Lizhuang

doi:10.1109/cvpr52688.2022.01451

Cited by 69 publications

(38 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To address this issue, recent studies have adopted a weakly supervised learning approach to train networks with partial annotations of point clouds. Previous studies [44,5,45,48,49,16,26,47,24] improved the semantic segmentation performance close to that of fully supervised one on small-scale datasets (e.g., ShapeNet [4] and PartNet [28]) as well as large-scale datasets (e.g., S3DIS [2] and ScanNet-v2 [8]).…”

Section: Introductionmentioning

confidence: 77%

“…Existing studies generated semantically transformed types of point clouds, such as 2D segmentation maps [42], subcloud-level annotation [44], and superpoint [5]. With sparse annotation, previous approaches have employed pre-training method [16,48], contrastive learning [16,26,24], and learning distribution consistency [45,49,24,47] to learn spatial information of point clouds. For learning the topology of a point cloud, graph-structure was utilized to represent features of points [5,26,49].…”

Section: Related Work 21 Weakly Supervised Semantic Segmentation On P...mentioning

confidence: 99%

“…Furthermore, it is challenging to generate a high-quality feature representations from partially annotated point clouds. Thus, existing studies focus on the feature representation of labeled points shared with unlabeled point clouds [45,48,49,26,24,47]. To obtain the feature embedding, previous studies minimize the difference between the ground truth and projected label [42,48].…”

Section: Sparse Annotation Embeddingmentioning

confidence: 99%

“…To obtain the feature embedding, previous studies minimize the difference between the ground truth and projected label [42,48]. In addition to the aforementioned studies, other approaches optimize the divergence between two probabilistic distributions [45,49,26,24,47]. In the training process, the above studies employed the softmax function.…”

Section: Sparse Annotation Embeddingmentioning

confidence: 99%

See 3 more Smart Citations

GaIA: Graphical Information Gain based Attention Network for Weakly Supervised Point Cloud Semantic Segmentation

Lee¹,

Yang²,

Han³

2022

Preprint

View full text Add to dashboard Cite

While point cloud semantic segmentation is a significant task in 3D scene understanding, this task demands a time-consuming process of fully annotating labels. To address this problem, recent studies adopt a weakly supervised learning approach under the sparse annotation. Different from the existing studies, this study aims to reduce the epistemic uncertainty measured by the entropy for a precise semantic segmentation. We propose the graphical information gain based attention network called GaIA, which alleviates the entropy of each point based on the reliable information. The graphical information gain discriminates the reliable point by employing relative entropy between target point and its neighborhoods. We further introduce anchor-based additive angular margin loss, ArcPoint. The ArcPoint optimizes the unlabeled points containing high entropy towards semantically similar classes of the labeled points on hypersphere space. Experimental results on S3DIS and ScanNet-v2 datasets demonstrate our framework outperforms the existing weakly supervised methods. We have released GaIA at https://github.com/Karel911/GaIA.

show abstract

Section: Introductionmentioning

confidence: 77%

Section: Related Work 21 Weakly Supervised Semantic Segmentation On P...mentioning

confidence: 99%

Section: Sparse Annotation Embeddingmentioning

confidence: 99%

Section: Sparse Annotation Embeddingmentioning

confidence: 99%

See 2 more Smart Citations

GaIA: Graphical Information Gain based Attention Network for Weakly Supervised Point Cloud Semantic Segmentation

Lee¹,

Yang²,

Han³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The methods can be categorized into point-based [14,29,37,53,63,64,65] and projection-based [1,15,46,74,75,83] approaches. To learn from fewer labels, various strategies were proposed with point-based methods: enforcing geometric prior [85], local neighborhood propagation [30,44], for smarter use of the rare labels some used active learning [77], pseudo labelling [38,59,73], graph propagation [44,59], self-training [38,89], temporal matching [59,67].…”

Section: Related Workmentioning

confidence: 99%

COARSE3D: Class-Prototypes for Contrastive Learning in Weakly-Supervised 3D Point Cloud Segmentation

Li,

Cao,

de Charette

2022

Preprint

View full text Add to dashboard Cite

Annotation of large-scale 3D data is notoriously cumbersome and costly. As an alternative, weakly-supervised learning alleviates such a need by reducing the annotation by several order of magnitudes. We propose COARSE3D, a novel architecture-agnostic contrastive learning strategy for 3D segmentation. Since contrastive learning requires rich and diverse examples as keys and anchors, we leverage a prototype memory bank capturing class-wise global dataset information efficiently into a small number of prototypes acting as keys. An entropy-driven sampling technique then allows us to select good pixels from predictions as anchors. Experiments on three projection-based backbones show we outperform baselines on three challenging real-world outdoor datasets, working with as low as 0.001% annotations.

show abstract

Learning Inter-superpoint Affinity for Weakly Supervised 3D Instance Segmentation

Tang

Hui

Xie

2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

HybridCR: Weakly-Supervised 3D Point Cloud Semantic Segmentation via Hybrid Contrastive Regularization

Cited by 69 publications

References 26 publications

GaIA: Graphical Information Gain based Attention Network for Weakly Supervised Point Cloud Semantic Segmentation

GaIA: Graphical Information Gain based Attention Network for Weakly Supervised Point Cloud Semantic Segmentation

COARSE3D: Class-Prototypes for Contrastive Learning in Weakly-Supervised 3D Point Cloud Segmentation

Learning Inter-superpoint Affinity for Weakly Supervised 3D Instance Segmentation

Contact Info

Product

Resources

About