Fig. 4: Experimental point cloud part segmentation on ShapeNet dataset.
From: Random memristor-based dynamic graph CNN for efficient point cloud learning at the edge
a Schematic of RDGCNN for the part segmentation task on the ShapeNet dataset. b Experimental feature evolution of points in an airplane model from the ShapeNet dataset. c Comparison of part segmentation results of software trainable DGCNN baseline and our co-design. d Parameter count comparisons of trainable software DGCNN baseline and our co-design. e Representative segmentation results of our co-design. f Comparison of training costs between the fully trainable DGCNN, DGCNN with splitted EdgeConv, and RDGCNN with splitted random EdgeConv (ours). g Comparison of inference energy between GPU, NPU, and our co-design.
