allset_3d_objects — 3D object nearest neighbourhoods (AllSet, 2021)

Description

Bipartite networks of 3D object models and their nearest-neighbour groups in a feature space, as compiled by Chien et al. as benchmarks for node classification, following the construction of Feng et al. (2019). The objects come from two collections of 3D shapes: ModelNet40, a set of computer-aided design (CAD) models of common objects (such as airplanes, chairs and cars) collected from the web by Wu et al. and classified into 40 categories, and the NTU 3D model dataset of Chen et al. (in the version known as NTU2012, with 67 categories). According to Chien et al., the objects are described by visual features extracted from rendered views of the models with the Multi-View Convolutional Neural Network (MVCNN) and the Group-View Convolutional Neural Network (GVCNN). For each object, a group was formed consisting of the object and its four nearest neighbours in feature space (the source does not state exactly which features were used for this), so that every group has exactly five objects and every object is the centre of exactly one group. Nodes are objects and groups, and an edge connects each group to each of its five member objects (including its centre). Groups centred on different objects can have the same members; they are kept as separate nodes. The groups cannot be reproduced from the 100-dimensional features included in the data, which were presumably derived differently (how they were obtained is not documented).

modelnet40: 12,311 objects from ModelNet40, in 40 categories, and their 12,311 groups (of which 10,630 have distinct sets of members).
ntu2012: 2,012 objects from the NTU dataset, in 67 categories, and their 2,012 groups (of which 1,635 have distinct sets of members).

The networks have the following properties:

is_neighborhood: whether the node is a group (otherwise it is an object). Objects are the nodes 0 to n-1, in the order of the source data, and groups the nodes n to 2n-1.
Object nodes: label, the category of the object, as an integer index (1 to 40 for modelnet40 and 0 to 66 for ntu2012, as in the source); the category names are not included in the source data. features, the 100-dimensional feature vector of the object, as given in the source data.
Group nodes: center, the index of the object node on which the group is centred. This is inferred from the order of the groups in the source data: the i-th group always contains the i-th object, which is consistent with groups being listed in the order of their centres.
Object nodes have center set to -1, and group nodes have label set to -1 and empty features.

Tags

Informational Unweighted Metadata

Citation

E. Chien, C. Pan, J. Peng, O. Milenkovic, "You are AllSet: A Multiset Function Framework for Hypergraph Neural Networks." International Conference on Learning Representations (ICLR) (2022) openreview.net
Y. Feng, H. You, Z. Zhang, R. Ji, Y. Gao, "Hypergraph Neural Networks." Proceedings of the AAAI Conference on Artificial Intelligence 33(01), 3558-3565 (2019) 10.1609/aaai.v33i01.33013558 Sci-Hub
Z. Wu, S. Song, A. Khosla, F. Yu, L. Zhang, X. Tang, J. Xiao, "3D ShapeNets: A deep representation for volumetric shapes." 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 1912-1920 (2015) 10.1109/CVPR.2015.7298801 Sci-Hub
D.-Y. Chen, X.-P. Tian, Y.-T. Shen, M. Ouhyoung, "On Visual Similarity Based 3D Model Retrieval." Computer Graphics Forum 22(3), 223-232 (2003) 10.1111/1467-8659.00669 Sci-Hub

Upstream URL OK

https://github.com/jianhao2016/AllSet

Upstream license

MIT, https://raw.githubusercontent.com/jianhao2016/AllSet/6281a2f1a91f6f26040777bb0b2578fc035dc57a/LICENSE

Networks

Tip: click on a column header to sort the list, or hover over it for a description.

Column legend

$\left<k\right>$: Mean degree
$\sigma_k$: Standard deviation of the degrees
$\lambda_h$: Largest eigenvalue of the Hashimoto (non-backtracking) matrix
$\tau$: Random walk mixing time
$r$: Degree assortativity
$c$: Global clustering coefficient
$\oslash$: (Pseudo-) diameter
$S$: Relative size of the largest component
NPs: Node properties
EPs: Edge properties

Name	Nodes	Edges	$\left<k\right>$	$\sigma_k$	$\lambda_h$	$\tau$	$r$	$c$	$\oslash$	$S$	Kind	Mode	NPs	EPs	gt	GraphML	GML	csv
modelnet40	24,622	61,555	5.00	2.40	6.34	6191.88	-0.14	0.00	58	0.99	Undirected	Bipartite	is_neighborhood label features center		4.7 MiB	8.1 MiB	8.1 MiB	6.2 MiB
ntu2012	4,024	10,060	5.00	2.03	5.44	554.67	-0.13	0.00	26	1.00	Undirected	Bipartite	is_neighborhood label features center		1.3 MiB	956 KiB	954 KiB	1017 KiB