TopNet: Structural Point Cloud Decoder

By Lyne P. Tchapmi, Vineet Kosaraju, Hamid Rezatofighi, Ian David Reid, Silvio Savarese. Computer Vision and Pattern Recognition, 2019

522 citations recorded by the source corpus.

3D point cloud generation is of great use for 3D scene modeling and understanding. Real-world 3D object point clouds can be properly described by a collection of low-level and high-level structures such as surfaces, geometric primitives, semantic parts,etc. In fact, there exist many different representations of a 3D object point cloud as a set of point groups. Existing frameworks for point cloud genera-ion either do not consider structure in their proposed solutions, or assume and enforce a specific structure/topology,e.g. a collection of manifolds or surfaces, for the generated point cloud of a 3D object. In this work, we pro-pose a novel decoder that generates a structured point cloud without assuming any specific structure or topology on the underlying point set. Our decoder is softly constrained to generate a point cloud following a hierarchical rooted tree structure. We show that given enough capacity and allowing for redundancies, the proposed decoder is very flexible and able to learn any arbitrary grouping of points including any topology on the point set. We evaluate our decoder on the task of point cloud generation for 3D point cloud shape completion. Combined with encoders from existing frameworks, we show that our proposed decoder significantly outperforms state-of-the-art 3D point cloud completion methods on the Shapenet dataset

Subfields: Computational Mechanics, Computer Graphics and Computer-Aided Design

View Paper: https://www.semanticscholar.org/paper/b6cf10713451191de8f0d30211f85a1080249a74

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 Graphics and Computer-Aided Design

Citations in the corpus

Unsupervised 3D Shape Completion through GAN Inversion (2021, methodology): “Owing to the coarse-to-fine strategy [28, 21, 15, 30, 34], they achieve impressive results on in-domain data, but may fail to sufficiently generalize to real-world scans or partial shapes in other forms.”
Unsupervised Point Cloud Pre-training via Occlusion Completion (2020, background): “The structure of the completion model c(·) is an “encoder-decoder” network [10, 48, 51, 60].”
Revealing Occlusions with 4D Neural Fields (2022, background): “Our goal is somewhat similar to that of point completion networks [21, 55, 60, 62, 68], although these works typically operate on a per-object basis and address only self-occlusions or amodal completion.”
AutoSDF: Shape Priors for 3D Completion, Reconstruction and Generation (2022, unknown): “Most recent approaches [1, 7, 30, 47–49] formulate it as performing completion on point clouds and can infer plausible global shapes but have difﬁculty in either capturing ﬁne-grained details, conditioning on sparse inputs, or generating diverse samples.”
GRNet: Gridding Residual Network for Dense Point Cloud Completion (2020, methodology): “Recent methods [9,11] adopt Multi-layer Perceptrons (MLPs) to directly operate on point clouds and aggregate information across points with max pooling.”
Learn to Predict Sets Using Feed-Forward Neural Networks (2020, background): “Other examples of such tasks include predicting a set of topics or concepts in documents [31], sets of points in point cloud data [32], [33], [34], segmentation of object instances [35] and a set of trajectories in multiple object tracking [36], [37], [38].”
Morphing and Sampling Network for Dense Point Cloud Completion (2019, background): “…learning for point cloud analysis and generation, some reasonable works on 3D point cloud completion have been presented (Yuan et al. 2018; Gurumurthy and Agrawal 2019; Sarmad, Lee, and Kim 2019; Tchapmi et al. 2019), which prevents high memory footprints and artifacts caused by discretization.”
ShapeFormer: Transformer-based Shape Completion via Sparse Representation (2022, background): “Point clouds, in contrast, do not have such a problem and are popularly used lately for generation [1,25] and completion [62,71,72,74].”
SnowflakeNet: Point Cloud Completion by Snowflake Point Deconvolution with Skip-Transformer (2021, methodology): “Therefore, point cloud completion [53, 43], which aims to predict a complete shape from its partial observation, is vital for various downstream tasks.”