Multi-Scale Point-Wise Convolutional Neural Networks for 3D Object Segmentation From LiDAR Point Clouds in Large-Scale Environments

Ma, Lingfei; Li, Ying; Li, Jonathan; Tan, Wei; Yu, Yongtao; Chapman, Michael A.

doi:10.1109/tits.2019.2961060

Cited by 91 publications

(20 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Many solutions from this area are based on machine learning algorithms (especially neural networks and Deep Learning), for which it is necessary to provide large amounts of data for effective training. Most works, which analyze information included in LiDAR point clouds are focused on object segmentation, classification, and simulation of vehicle behavior during different circumstances [ 54 , 55 , 56 , 57 ]. Therefore, also the simulators themselves are focused mainly on accurate object generation and not the accurate reflection of the nature of the point clouds (noise caused by measurement errors and phenomena characteristic for this type of data, e.g., the rolling shutter effect).…”

Section: Related Workmentioning

confidence: 99%

LiDAR Point Cloud Generation for SLAM Algorithm Evaluation

Sobczak

Filus

Domański

et al. 2021

Sensors

View full text Add to dashboard Cite

With the emerging interest in the autonomous driving level at 4 and 5 comes a necessity to provide accurate and versatile frameworks to evaluate the algorithms used in autonomous vehicles. There is a clear gap in the field of autonomous driving simulators. It covers testing and parameter tuning of a key component of autonomous driving systems, SLAM, frameworks targeting off-road and safety-critical environments. It also includes taking into consideration the non-idealistic nature of the real-life sensors, associated phenomena and measurement errors. We created a LiDAR simulator that delivers accurate 3D point clouds in real time. The point clouds are generated based on the sensor placement and the LiDAR type that can be set using configurable parameters. We evaluate our solution based on comparison of the results using an actual device, Velodyne VLP-16, on real-life tracks and the corresponding simulations. We measure the error values obtained using Google Cartographer SLAM algorithm and the distance between the simulated and real point clouds to verify their accuracy. The results show that our simulation (which incorporates measurement errors and the rolling shutter effect) produces data that can successfully imitate the real-life point clouds. Due to dedicated mechanisms, it is compatible with the Robotic Operating System (ROS) and can be used interchangeably with data from actual sensors, which enables easy testing, SLAM algorithm parameter tuning and deployment.

show abstract

Section: Related Workmentioning

confidence: 99%

LiDAR Point Cloud Generation for SLAM Algorithm Evaluation

Sobczak

Filus

Domański

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…Table 6 reports the quantitative results achieved by our RandLA-Net with several recently published baselines [1], [9], [23], [37], [87], [88], and Figure 9 shows the qualitative results. Note that, only the 3D coordinates of each point are feed into the network for fair comparison [37].…”

Section: Semantic Segmentation On Benchmarksmentioning

confidence: 99%

Learning Semantic Segmentation of Large-Scale Point Clouds with Random Sampling

Yang

Xie

et al. 2021

IEEE Trans. Pattern Anal. Mach. Intell.

101

View full text Add to dashboard Cite

We study the problem of efficient semantic segmentation of large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale point clouds. In this paper, we introduce RandLA-Net, an efficient and lightweight neural architecture to directly infer per-point semantics for large-scale point clouds. The key to our approach is to use random point sampling instead of more complex point selection approaches. Although remarkably computation and memory efficient, random sampling can discard key features by chance. To overcome this, we introduce a novel local feature aggregation module to progressively increase the receptive field for each 3D point, thereby effectively preserving geometric details. Comparative experiments show that our RandLA-Net can process 1 million points in a single pass up to 200× faster than existing approaches. Moreover, extensive experiments on five large-scale point cloud datasets, including Semantic3D, SemanticKITTI, Toronto3D, NPM3D and S3DIS, demonstrate the state-of-the-art semantic segmentation performance of our RandLA-Net.

show abstract

“…We used the L001, L003, and L004 as the training set and the L002 as the test set. We compared our method with PointNet++ [20], DGCNN [43], KPConv [25], MS-PCNN [44], TG-Net [45], and RandLA-Net [23]. DGCNN proposes a dubbed edge convolution acting on graphs.…”

Section: Experiments and Analysismentioning

confidence: 99%

Multi-Scale Attentive Aggregation for LiDAR Point Cloud Segmentation

Geng

Lü

et al. 2021

Remote Sensing

View full text Add to dashboard Cite

Semantic segmentation of LiDAR point clouds has implications in self-driving, robots, and augmented reality, among others. In this paper, we propose a Multi-Scale Attentive Aggregation Network (MSAAN) to achieve the global consistency of point cloud feature representation and super segmentation performance. First, upon a baseline encoder-decoder architecture for point cloud segmentation, namely, RandLA-Net, an attentive skip connection was proposed to replace the commonly used concatenation to balance the encoder and decoder features of the same scales. Second, a channel attentive enhancement module was introduced to the local attention enhancement module to boost the local feature discriminability and aggregate the local channel structure information. Third, we developed a multi-scale feature aggregation method to capture the global structure of a point cloud from both the encoder and the decoder. The experimental results reported that our MSAAN significantly outperformed state-of-the-art methods, i.e., at least 15.3% mIoU improvement for scene-2 of CSPC dataset, 5.2% for scene-5 of CSPC dataset, and 6.6% for Toronto3D dataset.

show abstract

Multi-Scale Point-Wise Convolutional Neural Networks for 3D Object Segmentation From LiDAR Point Clouds in Large-Scale Environments

Cited by 91 publications

References 53 publications

LiDAR Point Cloud Generation for SLAM Algorithm Evaluation

LiDAR Point Cloud Generation for SLAM Algorithm Evaluation

Learning Semantic Segmentation of Large-Scale Point Clouds with Random Sampling

Multi-Scale Attentive Aggregation for LiDAR Point Cloud Segmentation

Contact Info

Product

Resources

About