Spatial feature mapping for 6DoF object pose estimation

This work aims to estimate 6Dof (6D) object pose in background clutter. Considering the strong occlusion and noise, we propose utilize spatial structure for better tackling this challenging task. Observing that 3D mesh can be naturally abstracted by a graph, build graph using points as vertices connections edges. We construct corresponding mapping from 2D image features filling fusion of features. Afterward, Graph Convolutional Network (GCN) is applied help feature exchange among objects’ space. To address problem rotation symmetry ambiguity objects, spherical convolution utilized are combined with convolutional mapped graph. Predefined keypoints voted 6DoF obtained via fitting optimization. Two scenarios inference, one depth information other without it discussed. Tested on datasets YCB-Video LINEMOD, experiments demonstrate effectiveness our proposed method.