MAG-Comp: Memory-Augmented Geometry-Driven Amodal Point Cloud Completion for Occluded Industrial Objects

Point cloud completion is crucial for robotic tasks, especially with occluded and noisy industrial data. While two-dimensional image guidance has been traditional, pure point cloud methods increasingly achieve state-of-the-art results, making amodal completeness—recovering both visible and occluded parts—critical for robust interaction. Inspired by these insights, we propose MAG-Comp, a novel framework maximizing geometric information for amodal point cloud completion. MAG-Comp utilizes a Hierarchical Geometric Feature Encoder, a Class-Agnostic Geometric Memory Bank for shape priors, and a Dynamic Amodal Region Inference Module for explicit occluded geometry reconstruction. Experiments on ShapeNet-Amodal and an Industrial Bin-Picking Dataset confirm MAG-Comp's superior performance, achieving a Chamfer Distance of 1.58x10-3 and an Amodal IoU of 54.20% on ShapeNet-Amodal, consistently outperforming state-of-the-art methods. The framework demonstrates robustness to varying occlusion, strong generalization, and competitive inference efficiency, making it suitable for real-time industrial applications requiring precise amodal three-dimensional representations.