Abstract

The dataset contains the source code, model weights and a set of input points, camera poses and images utilized for the experimental validation on the labeling of deformable linear objects, associated to a novel algorithm called DLO-WSL. The proposed approach uses deep learning techniques aiming at the precise creation of instance masks of deformable linear objects starting from the input points provided by a user. The source code comprises a deep convolutional neural network employed for computing the correction offset to be applied at the input points. The dataset is associated with the related publication: A. Caporali, M. Pantano, L. Janisch, D. Regulin, G. Palli and D. Lee, "A Weakly Supervised Semi-Automatic Image Labeling Approach for Deformable Linear Objects," in IEEE Robotics and Automation Letters, vol. 8, no. 2, pp. 1013-1020, Feb. 2023, doi: 10.1109/LRA.2023.3234799.