The article by Qiuhao Li and Shenghai Yuan examines the refinement of the Jacquard Grasp Dataset. The researchers propose the use of a Human-In-The-Loop (HIL) method to improve the dataset’s quality. The HIL approach employs deep learning networks to predict object positions and orientations for robotic grasping. Images with low Intersection over Union (IOU) values undergo human evaluation, after which they are categorized and either enhanced with additional information or removed entirely. The refined dataset, named Jacquard V2 Grasping Dataset, significantly improves the training and prediction performance of neural networks.
Publication date: 9 Feb 2024
Project Page: https://github.com/lqh12345/JacquardV2
Paper: https://arxiv.org/pdf/2402.05747
