SnapNet: 3D point cloud semantic labeling with 2D deep segmentation networks

Boulch, Alexandre; Guerry, Joris; Saux, Bertrand Le; Audebert, Nicolas

doi:10.1016/j.cag.2017.11.010

Cited by 432 publications

(304 citation statements)

References 10 publications

Supporting

Mentioning

264

Contrasting

Unclassified

Order By: Relevance

“…Higher resolution classification requires a dense analysis of the rendered views. Similar to the approach taken by Boulch et al 32 , we solved the resulting image-to-surface mapping problem by rendering the cell views with spatially subdivided unique colors 33 , thereby creating an efficient reversible mapping between the 2D view space and the 3D surface ( Fig. 6a-c).…”

Section: High-resolution Semantic Segmentation Of Neurite Surfacesmentioning

confidence: 99%

Learning cellular morphology with neural networks

Schubert

Dorkenwald

Jain

et al. 2018

Preprint

View full text Add to dashboard Cite

Reconstruction and annotation of volume electron microscopy data sets of brain tissue is challenging, but can reveal invaluable information about neuronal circuits. Significant progress has recently been made in automated neuron reconstruction, as well as automated detection of synapses. However, methods for automating the morphological analysis of nanometer-resolution reconstructions are less established, despite their diverse application possibilities. Here, we introduce cellular morphology neural networks (CMNs), based on multi-view projections sampled from automatically reconstructed cellular fragments of arbitrary size and shape. Using unsupervised training we inferred morphology embeddings ("Neuron2vec") of neuron reconstructions and trained CMNs to identify glia cells in a supervised classification paradigm which was used to resolve neuron reconstruction errors. Finally, we demonstrate that CMNs can be used to identify subcellular compartments and the cell types of neuron reconstructions.

show abstract

Section: High-resolution Semantic Segmentation Of Neurite Surfacesmentioning

confidence: 99%

Learning cellular morphology with neural networks

Schubert

Dorkenwald

Jain

et al. 2018

Preprint

View full text Add to dashboard Cite

show abstract

“…Laga and Nakajima (2007) used AdaBoost to learn a classifier that relies on a small subset of the features with the mean of weak classifiers. The convolutional neural network (CNN) has been used for semantic labeling of point clouds by transforming the point dataset to a voxelization of the space (Wu et al, 2015) or many picked snapshots of the point cloud (Boulch et al, 2017). Although in point cloud data with noise, uneven density and occlusions machine learning techniques give better results for extracting specific objects, they are usually slow and rely on the result of feature extraction process.…”

Section: A Classification Of the Segmentation Methodsmentioning

confidence: 99%

“…It is also referred to as object retrieval which has received significant attention in recent years (Pang and Neumann, 2013;Yu et al, 2015;Boulch et al, 2017;Yang et al, 2017). The crucial part of these methods is how to extract salient geometric and topological characteristics from the object shapes.…”

Section: A Classification Of the Segmentation Methodsmentioning

confidence: 99%

A Super Voxel-Based Riemannian Graph for Multi Scale Segmentation of Lidar Point Clouds

2018

ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:Automatically segmenting LiDAR points into respective independent partitions has become a topic of great importance in photogrammetry, remote sensing and computer vision. In this paper, we cast the problem of point cloud segmentation as a graph optimization problem by constructing a Riemannian graph. The scale space of the observed scene is explored by an octree-based oversegmentation with different depths. The over-segmentation produces many super voxels which restrict the structure of the scene and will be used as nodes of the graph. The Kruskal coordinates are used to compute edge weights that are proportional to the geodesic distance between nodes. Then we compute the edge-weight matrix in which the elements reflect the sectional curvatures associated with the geodesic paths between super voxel nodes on the scene surface. The final segmentation results are generated by clustering similar super voxels and cutting off the weak edges in the graph. The performance of this method was evaluated on LiDAR point clouds for both indoor and outdoor scenes. Additionally, extensive comparisons to state of the art techniques show that our algorithm outperforms on many metrics.

show abstract

“…In this regard, the most promising approaches rely on classifying each 3D point of a point cloud based on a transformation of all points within its local neighborhood to a voxel-occupancy grid serving as input for a 3D-CNN [6,[45][46][47]. Alternatively, 2D image projections may be used as input for a 2D-CNN designed for semantic segmentation and a subsequent back-projection of predicted labels to 3D space delivers the semantically labeled 3D point cloud [48,49].…”

Section: Feature Extractionmentioning

confidence: 99%

Geospatial Computer Vision Based on Multi-Modal Data—How Valuable Is Shape Information for the Extraction of Semantic Information?

Weinmann

2017

Remote Sensing

View full text Add to dashboard Cite

Abstract:In this paper, we investigate the value of different modalities and their combination for the analysis of geospatial data of low spatial resolution. For this purpose, we present a framework that allows for the enrichment of geospatial data with additional semantics based on given color information, hyperspectral information, and shape information. While the different types of information are used to define a variety of features, classification based on these features is performed using a random forest classifier. To draw conclusions about the relevance of different modalities and their combination for scene analysis, we present and discuss results which have been achieved with our framework on the MUUFL Gulfport Hyperspectral and LiDAR Airborne Data Set.

show abstract

SnapNet: 3D point cloud semantic labeling with 2D deep segmentation networks

Cited by 432 publications

References 10 publications

Learning cellular morphology with neural networks

Learning cellular morphology with neural networks

A Super Voxel-Based Riemannian Graph for Multi Scale Segmentation of Lidar Point Clouds

Geospatial Computer Vision Based on Multi-Modal Data—How Valuable Is Shape Information for the Extraction of Semantic Information?

Contact Info

Product

Resources

About