Detail Preserved Point Cloud Completion via Separated Feature Aggregation

Abstract

Point cloud shape completion is a challenging problem in 3D vision and robotics. Existing learning-based frameworks leverage encoderdecoder architectures to recover the complete shape from a highly encoded global feature vector. Though the global feature can approximately represent the overall shape of 3D objects, it would lead to the loss of shape details during the completion process. In this work, instead of using a global feature to recover the whole complete surface, we explore the functionality of multi-level features and aggregate different features to represent the known part and the missing part separately. We propose two different feature aggregation strategies, named global & local feature aggregation(GLFA) and residual feature aggregation(RFA), to express the two kinds of features and reconstruct coordinates from their combination. In addition, we also design a refinement component to prevent the generated point cloud from non-uniform distribution and outliers. Extensive experiments have been conducted on the ShapeNet dataset. Qualitative and quantitative evaluations demonstrate that our proposed network outperforms current state-of-the art methods especially on detail preservation. Our code is available at https://github.com/XLechter/ Detail-Preserved-Point-Cloud-Completion-via-SFA.

ASFM-Net: Asymmetrical Siamese Feature Matching Network for Point Completion

“…SoftPoolNet [54] improves the max-pooling layer to a soft pooling layer, which can keep more information in multiply features. Zhang et al [65] propose a feature aggregation strategy to preserve the primitive details. Wang et al [51] design a cascaded refinement and adds the partial input into the decoder directly for high fidelity.…”

“…We compare our method with several current point cloud completion networks including FoldingNet [60], PCN [63], PointSetVoting [64], AtlasNet [11], RFA [65], TopNet [48], SoftPoolNet [54], and SA-Net [55]. Following them, the Chamfer Distance (CD) is used to evaluate quantitatively.…”

“…However, they all rely on the global feature vector extracted from the partial inputs to generate complete point clouds. Recently, RFA [65] realizes this problem and employs a feature aggregation strategy to enhance the representation of global features. However, it still can not solve the fundamental problem: The global features only extracted from the partial inputs must be incomplete and lose the geometric details.…”

Multi-View Partial (MVP) Point Cloud Challenge 2021 on Completion and Registration: Methods and Results

“…al. [16] 4.30 0.740 ECG [10] 3.58 0.753 GRNet [25] 3.87 0.692 NSFA [32] 3.77 0.783 VRCNet [12] 3.02 0.796 CRNet 2.51 0.824 increases the weight of MLPs (Model F in Table 5). These methods also ignore the important category information which contains discriminative semantics.…”

“…• Multi-Scale SPD module: To reveal fine local geometric details on the complete shape, existing methods [16,30,32] usually adopt folding-based strategy [26] to obtain the variations for learning different displacements for the duplicated points. However, the foldingbased strategy ignores the local shape characteristics contained in the original point due to the same 2D grids for sampling.…”