Dilated Point Convolutions: On the Receptive Field Size of Point Convolutions on 3D Point Clouds

Engelmann, Francis; Kontogianni, Theodora; Leibe, Bastian

doi:10.1109/icra40945.2020.9197503

Cited by 86 publications

(52 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We need to increase the probability that the information of the points in P l will propagate to P l +1 , allowing us to reduce the impact of some important points being discarded on the network performance. Inspired by [ 19 , 34 ], we simply stacked multiple NFFUs to form an extended neighborhood feature fusion block (ENFFB), which enabled us to reserve more information of points at a low cost.…”

Section: Methodsmentioning

confidence: 99%

Graph Attention Feature Fusion Network for ALS Point Cloud Classification

Yang

Zhang

Huang

2021

Sensors

View full text Add to dashboard Cite

Classification is a fundamental task for airborne laser scanning (ALS) point cloud processing and applications. This task is challenging due to outdoor scenes with high complexity and point clouds with irregular distribution. Many existing methods based on deep learning techniques have drawbacks, such as complex pre/post-processing steps, an expensive sampling cost, and a limited receptive field size. In this paper, we propose a graph attention feature fusion network (GAFFNet) that can achieve a satisfactory classification performance by capturing wider contextual information of the ALS point cloud. Based on the graph attention mechanism, we first design a neighborhood feature fusion unit and an extended neighborhood feature fusion block, which effectively increases the receptive field for each point. On this basis, we further design a neural network based on encoder–decoder architecture to obtain the semantic features of point clouds at different levels, allowing us to achieve a more accurate classification. We evaluate the performance of our method on a publicly available ALS point cloud dataset provided by the International Society for Photogrammetry and Remote Sensing (ISPRS). The experimental results show that our method can effectively distinguish nine types of ground objects. We achieve more satisfactory results on different evaluation metrics when compared with the results obtained via other approaches.

show abstract

Section: Methodsmentioning

confidence: 99%

Graph Attention Feature Fusion Network for ALS Point Cloud Classification

Yang

Zhang

Huang

2021

Sensors

View full text Add to dashboard Cite

show abstract

“…In [61], a point-wise convolution operator has been proposed, where the nearest points are placed into kernel cells and then convolved with kernel weights. In [62], a dilated point convolution operation is proposed to accumulate dilated neighboring features, instead of the K nearest neighbors.…”

Section: Point-based Methodsmentioning

confidence: 99%

Exploiting Structured CNNs for Semantic Segmentation of Unstructured Point Clouds from LiDAR Sensor

et al. 2021

View full text Add to dashboard Cite

Accurate semantic segmentation of 3D point clouds is a long-standing problem in remote sensing and computer vision. Due to the unstructured nature of point clouds, designing deep neural architectures for point cloud semantic segmentation is often not straightforward. In this work, we circumvent this problem by devising a technique to exploit structured neural architectures for unstructured data. In particular, we employ the popular convolutional neural network (CNN) architectures to perform semantic segmentation of LiDAR data. We propose a projection-based scheme that performs an angle-wise slicing of large 3D point clouds and transforms those slices into 2D grids. Accounting for intensity and reflectivity of the LiDAR input, the 2D grid allows us to construct a pseudo image for the point cloud slice. We enhance this image with low-level image processing techniques of normalization, histogram equalization, and decorrelation stretch to suit our ultimate object of semantic segmentation. A large number of images thus generated are used to induce an encoder-decoder CNN model that learns to compute a segmented 2D projection of the scene, which we finally back project to the 3D point cloud. In addition to a novel method, this article also makes a second major contribution of introducing the enhanced version of our large-scale public PC-Urban outdoor dataset which is captured in a civic setup with an Ouster LiDAR sensor. The updated dataset (PC-Urban_V2) provides nearly 8 billion points including over 100 million points labeled for 25 classes of interest. We provide a thorough evaluation of our technique on PC-Urban_V2 and three other public datasets.

show abstract

“…Pointbased approaches such as [33,38,39,49,55,56] perform semantic segmentation by splitting scenes into smaller chunks, effectively restricting the model's ability to learn from global context. Subsequent works [15,16,29,59] report improvements by increasing the spatial context. By leveraging data efficient sparse convolutions, recent voxelbased methods e.g., MinkowskiNet [9] and SparseCon-vNet [22] are capable of processing full scenes at once, thus, capturing the global context of the scene.…”

Section: Related Workmentioning

confidence: 99%

Mix3D: Out-of-Context Data Augmentation for 3D Scenes

Nekrasov

Schult

Litany³

et al. 2021

2021 International Conference on 3D Vision (3DV)

Self Cite

View full text Add to dashboard Cite

We present Mix3D, a data augmentation technique for segmenting large-scale 3D scenes. Since scene context helps reasoning about object semantics, current works focus on models with large capacity and receptive fields that can fully capture the global context of an input 3D scene. However, strong contextual priors can have detrimental implications like mistaking a pedestrian crossing the street for a car. In this work, we focus on the importance of balancing global scene context and local geometry, with the goal of generalizing beyond the contextual priors in the training set. In particular, we propose a "mixing" technique which creates new training samples by combining two augmented scenes. By doing so, object instances are implicitly placed into novel out-of-context environments, therefore making it harder for models to rely on scene context alone, and instead infer semantics from local structure as well. We perform detailed analysis to understand the importance of global context, local structures and the effect of mixing scenes. In experiments, we show that models trained with Mix3D profit from a significant performance boost on indoor (ScanNet, S3DIS) and outdoor datasets (SemanticKITTI). Mix3D can be trivially used with any existing method, e.g., trained with Mix3D, MinkowskiNet outperforms all prior state-of-the-art methods by a significant margin on the ScanNet test benchmark (78.1% mIoU). Code

show abstract

Dilated Point Convolutions: On the Receptive Field Size of Point Convolutions on 3D Point Clouds

Cited by 86 publications

References 28 publications

Graph Attention Feature Fusion Network for ALS Point Cloud Classification

Graph Attention Feature Fusion Network for ALS Point Cloud Classification

Exploiting Structured CNNs for Semantic Segmentation of Unstructured Point Clouds from LiDAR Sensor

Mix3D: Out-of-Context Data Augmentation for 3D Scenes

Contact Info

Product

Resources

About