Refining object proposals using structured edge and superpixel contrast in robotic grasping

Lu Chen, Panfeng Huang, Zhou Zhao

Abstract: Highlights •Eliminate the common assumptions made in grasping scenario, such as the fixed object position and monotonous manipulation background.•A translation-to-alignment mechanism to align input proposals.•Introduce the distance bias to improve the connectivity between superpixels.•Experiments are performed on both benchmark dataset and robotic task.