SHOT: Unique signatures of histograms for surface and texture description

By Samuele Salti, Federico Tombari, Luigi Di Stefano. Computer Vision and Image Understanding, 2014

986 citations recorded by the source corpus.

Abstract This paper presents a local 3D descriptor for surface matching dubbed SHOT. Our proposal stems from a taxonomy of existing methods which highlights two major approaches, referred to as Signatures and Histograms, inherently emphasizing descriptiveness and robustness respectively. We formulate a comprehensive proposal which encompasses a repeatable local reference frame as well as a 3D descriptor, the latter featuring an hybrid structure between Signatures and Histograms so as to aim at a more favorable balance between descriptive power and robustness. A quite peculiar trait of our method concerns seamless integration of multiple cues within the descriptor to improve distinctiveness, which is particularly relevant nowadays due to the increasing availability of affordable RGB-D sensors which can gather both depth and color information. A thorough experimental evaluation based on datasets acquired with different types of sensors, including a novel RGB-D dataset, vouches that SHOT outperforms state-of-the-art local descriptors in experiments addressing descriptor matching for object recognition, 3D reconstruction and shape retrieval.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/6345842ec06145ac397b8b46d5f64113162a9268

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Fully Convolutional Geometric Features (2019, background): “SHOT [26] creates a 3D histogram of normal vectors.”
Voxelized GICP for Fast and Accurate 3D Point Cloud Registration (2020, background): “Many features have been proposed for point cloud registration, such as basic plane and edge features [10], [4], Fast Point Feature Histogram (FPFH) [11], and Signature of Histograms of OrienTations (SHOT) [12].”
Self-supervised Learning of Dense Shape Correspondence (2018, methodology): “For each remeshed shape k ∼ 70−150 Laplace-Beltrami eigenfunctions were calculated as well as a 352-dimensional SHOT descriptor [39], using 10 bins and a SHOT radius that was roughly chosen to 5% of the maximal pairwise geodesic distance.”
D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (2020, neutral): “ably ascribed to the small voxel size (1.875cm) used in PerfectMatch, while D3Feat is performed using 3cm voxel downsampling. Origin Rotated FMR (%) STD FMR (%) STD FPFH [28] 35.9 13.4 36.4 13.6 SHOT [30] 23.8 10.9 23.4 9.5 3DMatch [36] 59.6 8.8 1.1 1.2 CGF [15] 58.2 14.2 58.5 14.0 PPFNet [5] 62.3 1”
Deep Global Registration (2020, background): “Traditional hand-crafted features commonly summarize pairwise or higher-order relationships in histograms [20, 37, 40, 36, 35].”
Learning to Detect Good 3D Keypoints (2017, methodology): “In particular, throughout our experimental evaluation, we will consider descriptors as diverse as SHOT (Salti et al. 2014), FPFH (Rusu et al. 2009) and Spin Images (Johnson and Hebert 1999) in order to show how their matching pipelines can be enhanced significantly by deploying a companion detector…”
3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration (2018, background): “…approaches of the local 3D feature detectors and descriptors are heavily inﬂuenced by the widely studied 2D local features methods [17,2], and can be broadly categorized into handcrafted [35,5,26,25,13,30,28] and learning approaches [34,8,15,27,6] - i.e. pre-and post-deep learning approaches.”
RGBD Datasets: Past, Present and Future (2016, methodology): “One solution has been to fix the target objects to a calibration board to allow for ground-truth tracking using the RGB channels [45], while [86] and [84] have the poses manually aligned.”
Pointwise Rotation-Invariant Network with Adaptive Sampling and 3D Spherical Voxel Convolution (2018, methodology): “es from geometries could be retrospected to manual designed features, including Point Feature Histograms (PFH)[23], Fast Point Feature Histograms (FPFH)[22], Signature of Histogram Orientations (SHOT)[24], and Unique Shape Contexts (USC)[26]. These descriptors rely on delicate hand-craft design, and”
A comprehensive survey on point cloud registration (2021, neutral): “2 m ) SHOT [88] 27 rmseP( 0 .”
Deep Projective 3D Semantic Segmentation (2017, neutral): “Popular alternatives include features based on the 3D structure tensor [7,26,11,1], histogram-based descriptors [7,16,11] such as Spin Images [10] and SHOT [21], and simple color features [26,16,11].”
BOP: Benchmark for 6D Object Pose Estimation (2018, methodology): “The method was evaluated with several descriptors: 153d SI [19], 352d SHOT [30], 30d ECSAD [20], and 1536d PPFH [5].”
Multi-directional geodesic neural networks via equivariant convolution (2018, methodology): “2017], we used different inputs, SHOT [Salti et al. 2014] and WKS [Aubry et al. 2011] descriptors as well as the 3D coordinates of the shape, we denote by SHOTk the 64-dimensional SHOT descriptor with window radius equal to k percent of the shape area.”
Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards (2016, methodology): “We therefore make each di rotation-invariant by orienting its axes along the eigenvectors of a weighted covariance matrix of the 3D surface points that generate di as described in [37].”
RPM-Net: Robust Point Matching Using Learned Features (2020, methodology): “Generally, these descriptors accumulate measurements (typically number of points) into histograms according to their spatial coordinates [9, 14, 32], or their geometric attributes such as curvature [4] or normals [28].”
A Comprehensive Performance Evaluation of 3D Local Feature Descriptors (2015, background): “Signature of Histogram of Orientations (SHOT) (Tombari et al. 2010b; Salti et al. 2014) The descriptor encodes the histograms of the surface normals in different spatial locations.”