Escape from Cells: Deep Kd-Networks for the Recognition of 3D Point Cloud Models

By Roman Klokov, Victor S. Lempitsky. 2017

1,032 citations recorded by the source corpus.

We present a new deep learning architecture (called Kdnetwork) that is designed for 3D model recognition tasks and works with unstructured point clouds. The new architecture performs multiplicative transformations and shares parameters of these transformations according to the subdivisions of the point clouds imposed onto them by kdtrees. Unlike the currently dominant convolutional architectures that usually require rasterization on uniform twodimensional or three-dimensional grids, Kd-networks do not rely on such grids in any way and therefore avoid poor scaling behavior. In a series of experiments with popular shape recognition benchmarks, Kd-networks demonstrate competitive performance in a number of shape recognition tasks such as shape classification, shape retrieval and shape part segmentation.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Geology

How researchers see this paper

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

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Geology

Citations in the corpus

Revisiting Point Cloud Classification: A New Benchmark Dataset and Classification Model on Real-World Data (2019, neutral): “This situation is often seen in sliding window-based object detec-Method - 90.0 Flex-Convolution [14] - 90.2 Kd-Net [23] 88.5 90.6 (91.8 *) PointNet++ [32] 87.8 90.7 (91.9 w/ normal) SO-Net [24] 87.3 90.9 (93.4 w/ normal) KCNet [33] - 91 3DmFV [2] 86.3 91.4 SpecGCN [41] - 91.5 (92.1 w/ normal)…”
Neural 3D Mesh Renderer (2017, methodology): “Point clouds have been used for both recognition [12, 19, 21] and reconstruction [5].”
PointCNN: Convolution On X-Transformed Points (2018, background): “Octree [37, 47], Kd-Tree [22] and Hash [41] based approaches have been proposed to save computation by skipping convolution in empty space.”
Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline (2021, background): “…for Point-Cloud Analysis: A broad class of DNNs have emerged to process 3D points directly (Simonovsky & Komodakis, 2017; Zaheer et al., 2017; Klokov & Lempitsky, 2017; Xu et al., 2018; Atzmon et al., 2018; Wang et al., 2018a; Li et al., 2018a; Groh et al., 2018; Ben-Shabat et al., 2018; Xie…”
FoldingNet: Interpretable Unsupervised Learning on 3D Point Clouds (2017, background): “Some other options include octree-based [42] and kd-tree-based [28] neural networks.”
SO-Net: Self-Organizing Network for Point Cloud Analysis (2018, background): “Similarly in the Kd-Net [18], experiments are conducted using the kd-tree split directions without point information, i.e. feeding directions of the splits into a MLP.”
PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition (2018, background): “Kd-network [19] also works for unordered point cloud inputs by transforming them into kd-trees.”
PointASNL: Robust Point Clouds Processing Using Nonlocal Neural Networks With Adaptive Sampling (2020, neutral): “9 Kd-Net [15] pnt 32k 94.”
DensePoint: Learning Densely Contextual Representation for Efficient Point Cloud Processing (2019, methodology): “As in [21], we augment the input with random anisotropic scaling in range [-0.”
TopNet: Structural Point Cloud Decoder (2019, methodology): “This issue has been addressed using sparse representations [30, 32, 33, 17, 12, 25] which makes processing voxels more efficient.”
SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters (2018, methodology): “4 Kd-Network [9] (depth 15) 1024 points 91.”
Dynamic Filters in Graph Convolutional Networks (2017, background): “[12] developed a deep network based on kd-trees over 3D point clouds, sharing parameters across the tree based on the depth of a split and the direction of split.”
3D Semantic Segmentation with Submanifold Sparse Convolutional Networks (2017, methodology): “Kd-Networks [10] build a Kd-tree by recursively partitioning the space along the axis of largest variation until each leaf of the tree contains one input point.”
Triplet-Center Loss for Multi-view 3D Object Retrieval (2018, methodology): “We choose the following methods for comparisons, including GIFT [1], MVCNN [33], Wang [27], Li [27], Kdnetwork [15].”
Large-Scale 3D Shape Reconstruction and Segmentation from ShapeNet Core55 (2017, methodology): “The architecture for semantic segmentation is obtained from the classiﬁcation architecture in the way that is analogous to [10].”
Recurrent Slice Networks for 3D Segmentation of Point Clouds (2018, methodology): “In KD-network, a KD-tree was first built on input point clouds.”