One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentation

Liu, Zhengzhe; Qi, Xiaojuan; Fu, Chi‐Wing

doi:10.1109/cvpr46437.2021.00177

Cited by 120 publications

(136 citation statements)

References 30 publications

Supporting

Mentioning

135

Contrasting

Order By: Relevance

“…Weakly Supervised 3D Semantic Segmentation Existing works explore the 3D semantic segmentation with various types of weak supervision, including 2D image [44], subcloud-level [48], segment-level [38], and point-level supervision [11,26,34,52]. The first three types can be grouped into indirect annotations [11].…”

Section: Related Workmentioning

confidence: 99%

“…around 22.3 minutes to annotate an indoor scene on average [5]). Thus, weakly * Equal contribution † Corresponding author ) the performance of PointMatch on the ScanNet-v2 and S3DIS datasets over various weakly supervised semantic segmentation settings: annotating 0.01%, 0.1% of points [11], 20 points per-scene [10], and "1thing1click" [26]. (c), (d) a comparison between previous works and the proposed approach.…”

Section: Introductionmentioning

confidence: 99%

“…Recently, several approaches are proposed for weakly supervised point cloud semantic segmentation with different kinds of weak labels, including projected 2D image [44], subcloud-level [48], segment-level [38], and point-level [10,11,26,52] supervision. In this paper, we focus on addressing the setting of sparse point-level labels, which is one of the most convenient annotation schemes in the application.…”

Section: Introductionmentioning

confidence: 99%

“…The key challenge of this task is the difficulty of learning a robust model given very sparse supervision in the point cloud (e.g. 0.1%, 0.01% of points annotated in [11] and around 0.02% in [26]). Existing solutions are mainly committed to alleviating the label spar-sity by reusing limited supervision, i.e., first probing the highly-related points [11] or super-voxels [26] and allowing them to share the same training labels.…”

Section: Introductionmentioning

confidence: 99%

“…0.1%, 0.01% of points annotated in [11] and around 0.02% in [26]). Existing solutions are mainly committed to alleviating the label spar-sity by reusing limited supervision, i.e., first probing the highly-related points [11] or super-voxels [26] and allowing them to share the same training labels. However, this line of works are explicitly constructed on label propagation and employ point cloud data as the propagation guidance, which suffers from (i) the insufficient exploitation of data information limits the learning efficiency, and (ii) the propagated labels strongly rely on the original annotation scale thus the performance is easily suppressed when given much fewer labels.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

PointMatch: A Consistency Training Framework for Weakly Supervised Semantic Segmentation of 3D Point Clouds

Wu¹,

Yan²,

Cai³

et al. 2022

Preprint

View full text Add to dashboard Cite

Semantic segmentation of point cloud usually relies on dense annotation that is exhausting and costly, so it attracts wide attention to investigate solutions for the weakly supervised scheme with only sparse points annotated. Existing works start from the given labels and propagate them to highly-related but unlabeled points, with the guidance of data, e.g. intra-point relation. However, it suffers from (i) the inefficient exploitation of data information, and (ii) the strong reliance on labels thus is easily suppressed when given much fewer annotations. Therefore, we propose a novel framework, PointMatch, that stands on both data and label, by applying consistency regularization to sufficiently probe information from data itself and leveraging weak labels as assistance at the same time. By doing so, meaningful information can be learned from both data and label for better representation learning, which also enables the model more robust to the extent of label sparsity. Simple yet effective, the proposed PointMatch achieves the state-of-theart performance under various weakly-supervised schemes on both ScanNet-v2 and S3DIS datasets, especially on the settings with extremely sparse labels, e.g. surpassing SQN by 21.2% and 17.2% on the 0.01% and 0.1% setting of ScanNet-v2, respectively.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

PointMatch: A Consistency Training Framework for Weakly Supervised Semantic Segmentation of 3D Point Clouds

Wu¹,

Yan²,

Cai³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

LiDAL: Inter-frame Uncertainty Based Active Learning for 3D LiDAR Semantic Segmentation

Bai

Wang

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

We propose LiDAL, a novel active learning method for 3D Li-DAR semantic segmentation by exploiting inter-frame uncertainty among LiDAR frames. Our core idea is that a well-trained model should generate robust results irrespective of viewpoints for scene scanning and thus the inconsistencies in model predictions across frames provide a very reliable measure of uncertainty for active sample selection. To implement this uncertainty measure, we introduce new inter-frame divergence and entropy formulations, which serve as the metrics for active selection. Moreover, we demonstrate additional performance gains by predicting and incorporating pseudo-labels, which are also selected using the proposed inter-frame uncertainty measure. Experimental results validate the effectiveness of LiDAL: we achieve 95% of the performance of fully supervised learning with less than 5% of annotations on the SemanticKITTI and nuScenes datasets, outperforming state-of-the-art active learning methods. Code release: https://github.com/hzykent/LiDAL

show abstract

DODA: Data-Oriented Sim-to-Real Domain Adaptation for 3D Semantic Segmentation

Ding

Yang

Jiang

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentation

Cited by 120 publications

References 30 publications

PointMatch: A Consistency Training Framework for Weakly Supervised Semantic Segmentation of 3D Point Clouds

PointMatch: A Consistency Training Framework for Weakly Supervised Semantic Segmentation of 3D Point Clouds

LiDAL: Inter-frame Uncertainty Based Active Learning for 3D LiDAR Semantic Segmentation

DODA: Data-Oriented Sim-to-Real Domain Adaptation for 3D Semantic Segmentation

Contact Info

Product

Resources

About