Grasp Affordance Dataset

The Grasp Affordance Dataset extends the GraspNet-1Billion dataset with object part segmentation labels for 30 selected objects that can be grasped by different parts to accomplish different tasks. It contains 82,944 segmentation images with 60 different affordance regions on 162 scenes of the GraspNet-1Billion dataset. These segmentation labels can be used for training and benchmarking vision-based grasp affordance estimation methods. For more information and the generation procedure of the dataset, please see the README.txt file and our publication [1]. 

Please read the LICENSE.txt file carefully before using this dataset. The commercial use of the Grasp Affordance Dataset is not permitted. If you use this dataset, please give proper attribution to all of the publications listed below: [1], [2], [3]. The publication [1] is for this dataset (Grasp Affordance Dataset) and the publications [2] and [3] are for the origin dataset (GraspNet-1Billion).

[1] Furkan Kaynar, Sudarshan Rajagopalan, Mahmoud Krichene, Eckehard Steinbach, 2026. Human-in-the-loop RGB semantic segmentation for intuitive teaching of novel grasping tasks via few-shot adaptation of lightweight neural networks. Computer Vision and Image Understanding, Volume 267, April 2026, 104620. DOI: https://doi.org/10.1016/j.cviu.2025.104620

[2] Hao-Shu Fang, Minghao Gou, Chenxi Wang, Cewu Lu, 2023. Robust grasping across diverse sensor qualities: The GraspNet-1Billion dataset. The International Journal of Robotics Research. 2023; Volume 42, Issue 12, Pages: 1094-1103. DOI: https://doi.org/10.1177/02783649231193710

[3] Hao-Shu Fang, Chenxi Wang, Minghao Gou, Cewu Lu, 2020. Graspnet-1billion: A large-scale benchmark for general object grasping. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 11441– 11450. DOI: https://doi.org/10.1109/CVPR42600.2020.01146