GRNet: Gridding Residual Network for Dense Point Cloud Completion

• DOI: 10.48550/arxiv.2006.03761
• Publication Date: 2020-01-01
• Authors: Haozhe Xie; Hongxun Yao; Shangchen Zhou; Jiageng Mao; Shengping Zhang; Wenxiu Sun
• Keywords: Feature (linguistics); Perceptron

Estimating the complete 3D point cloud from an incomplete one is a key problem in many vision and robotics applications. Mainstream methods (e.g., PCN TopNet) use Multi-layer Perceptrons (MLPs) to directly process clouds, which may cause loss of details because structural context clouds are not fully considered. To solve this problem, we introduce grids as intermediate representations regularize unordered clouds. We therefore propose novel Gridding Residual Network (GRNet) for completion. In particular, devise two differentiable layers, named Reverse, convert between without losing information. also present Cubic Feature Sampling layer extract features neighboring points, preserves addition, design new function, namely Loss, calculate L1 distance predicted ground truth helpful recover details. Experimental results indicate that proposed GRNet performs favorably against state-of-the-art on ShapeNet, Completion3D, KITTI benchmarks.