One-Shot Object Affordance Detection in the Wild

Affordance detection refers to identifying the potential action possibilities of objects in an image, which is a crucial ability for robot perception and manipulation. To empower robots with this unseen scenarios, we first study challenging one-shot affordance problem paper, i.e., given support image that depicts purpose, all scene common should be detected. end, devise One-Shot Detection Network (OSAD-Net) firstly estimates human purpose then transfers it help detect from candidate images. Through collaboration learning, OSAD-Net can capture characteristics between having same underlying learn good adaptation capability perceiving affordances. Besides, build large-scale purpose-driven dataset v2 (PADv2) by collecting labeling 30k images 39 103 object categories. With complex scenes rich annotations, our PADv2 used as test bed benchmark methods may also facilitate downstream vision tasks, such understanding, recognition, Specifically, conducted comprehensive experiments on including 11 advanced models several related research fields. Experimental results demonstrate superiority model over previous representative ones terms both objective metrics visual quality. The suite available at https://github.com/lhc1224/OSAD_Net .