Weakly Supervised Semantic Point Cloud Segmentation: Towards 10× Fewer Labels

Xu, Xun; Lee, Gim Hee

doi:10.1109/cvpr42600.2020.01372

Cited by 189 publications

(180 citation statements)

References 12 publications

Supporting

Mentioning

178

Contrasting

Order By: Relevance

“…A variety of techniques were developed to reduce the requirement for large quantities of 3D supervision. The most typical strategy is to utilize synthetic data to quickly build large-scale 3D datasets [38]- [41], which has benefited a lot of tasks, such as 3D flow estimation [42] and 3D human reconstruction [43]. Some attempts [44], [45] were made to learn indoor 3D detection through semi-supervised learning techniques, i.e., combine labeled data with unlabeled data.…”

Section: Related Workmentioning

confidence: 99%

Towards A Weakly Supervised Framework for 3D Point Cloud Object Detection and Annotation

Meng¹,

Yang

Zhou

et al. 2021

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

It is quite laborious and costly to manually label LiDAR point cloud data for training high-quality 3D object detectors. This work proposes a weakly supervised framework which allows learning 3D detection from a few weakly annotated examples. This is achieved by a two-stage architecture design. Stage-1 learns to generate cylindrical object proposals under inaccurate and inexact supervision, obtained by our proposed BEV center-click annotation strategy, where only the horizontal object centers are click-annotated in bird's view scenes. Stage-2 learns to predict cuboids and confidence scores in a coarse-to-fine, cascade manner, under incomplete supervision, i.e., only a small portion of object cuboids are precisely annotated. With KITTI dataset, using only 500 weakly annotated scenes and 534 precisely labeled vehicle instances, our method achieves 86−97% the performance of current top-leading, fully supervised detectors (which require 3, 712 exhaustively annotated scenes with 15, 654 instances). More importantly, with our elaborately designed network architecture, our trained model can be applied as a 3D object annotator, supporting both automatic and active (human-in-the-loop) working modes. The annotations generated by our model can be used to train 3D object detectors, achieving over 95% of their original performance (with manually labeled training data). Our experiments also show our model's potential in boosting performance when given more training data. The above designs make our approach highly practical and open-up opportunities for learning 3D detection at reduced annotation cost.

show abstract

Section: Related Workmentioning

confidence: 99%

Towards A Weakly Supervised Framework for 3D Point Cloud Object Detection and Annotation

Meng¹,

Yang

Zhou

et al. 2021

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

show abstract

“…The constraints used in [31] are based on the prior knowledge of target size which is reasonable for medical image segmentation but not applicable for our case of baggage security screening where contraband materials can be of arbitrary sizes. Xu et al [32] addressed the incomplete supervision problems in point cloud segmentation. Several complementary components were combined in their framework including an incomplete supervision branch, an inexact supervision branch, Siamese self-supervision, spatial and color smoothness constraints.…”

Section: Weakly Supervised 3d Segmentationmentioning

confidence: 99%

On the Evaluation of Semi-Supervised 2D Segmentation for Volumetric 3D Computed Tomography Baggage Security Screening

Wang

Breckon

2021

2021 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

We address the automatic contraband material detection problem within volumetric 3D Computed Tomography (CT) data for baggage security screening. Distinct from the prohibited item detection using object detection techniques, contraband material detection is usually formulated as a segmentation problem due to the variations of their potential appearances and shapes. Previous studies have employed either morphological operation based traditional methods or 3D Convolutional Neural Networks (CNN) for 3D segmentation towards target material detection within volumetric 3D CT baggage security screening imagery. In this work, we investigate the effectiveness of 2D semantic segmentation techniques in this 3D CT segmentation problem. Specifically, we extract 2D slices from three planes of the 3D CT volumes and train a 2D segmentation model which is subsequently used to predict segmentation results for all the slices from a given test CT volume. Moreover, we also evaluate how the performance is affected when using a reduced number of annotated slices for training. As a result, it is demonstrated reasonable performance can be achieved with very limited annotated slices (1-2) per CT volume during training. Finally, we propose a semi-supervised learning framework for 3D CT segmentation. Using only 1/128 of the total number of annotated slices, our framework can achieve comparable performance with full supervision.

show abstract

“…Following the weakly supervised segmentation strategy in [7], we choose to define the total loss objective function L total (ȳ, p(y|P, g, e, Θ, Γ)) by composing the following four constraints based on the partially labelled ground truth ȳ of point cloud semantic category:…”

Section: Feature Fusionmentioning

confidence: 99%

“…To generate weakly supervised settings, we follow the 0.01 scheme as in the state-of-the-art weakly supervised point cloud segmentation (WSPCS) [7] method, in which only |M| N = 1% points with ground truth labels are randomly selected from each semantic part category of a shape model, thus there are 20 points Fig. 2 Comparison of the total loss as well as three accuracy criteria between WSPCS [7] and ours GECNN approach evaluated on the validation dataset in the training process. The comparison of train loss between WSPCS and GECNN on the train dataset is also illustrated.…”

Section: Validation On Shapenetmentioning

confidence: 99%

“…But the limited number of pooling layers will inevitably restrict the size of receptive field thus hard for complex scene abstraction. Another line of work is based on incomplete weakly supervised segmentation [7] through ensemble learning by introducing additional constraints such as augmented Siamese network, inexact multiinstance learning and spatial smoothness loss. This approach achieves significantly higher segmentation scores even at the very limited 1% and 10% weak-supervision level.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation