Evaluation Dataset for Tracking Branched Deformable Linear Objects from Point Clouds

This dataset contains the evaluation data used in the PhD thesis *"Model-based Tracking of Branched Deformable Linear Objects from Point Clouds for Robotic Manipulation"* by Markus Wnuk, 2025, Technical University of Stuttgart, (to be submitted). It includes annotated 3D point clouds, kinematic configuration data, and metadata files that were used to quantitatively and qualitatively assess the proposed tracking method for Branched Deformable Linear Objects (BDLO).The dataset is intended to support reproducibility, benchmarking, and further research in the field of deformable object perception and robotic manipulation. ### Contents Background: RGB image of the scene background used for segmentation via background subtraction. BDLO Descriptions: Metadata describing each BDLO sample, including its graph structure (`model.json`) and label configuration (label IDs used for ground-truth reference). Camera Parameters: Intrinsic and extrinsic calibration data for the RGB-D camera used in the experiments. Evaluation Datasets: Scene-specific folders containing: Metadata (e.g., BDLO type, camera setup, grasping positions) RGB-D data (`.png` files for color and disparity) Results: Processed evaluation results, including: - Compact result files (`.pkl`) - Visualization plots (`.jpg`) ### Use Cases The dataset can be used for: - Benchmarking BDLO tracking and shape estimation methods- Developing perception pipelines for robotic manipulation of flexible or branched objects- Training and evaluation of learning-based or model-based approaches for deformable object localization ### Citation Please cite the following work if you use this dataset: > Markus Wnuk. (2025). *Model-based Tracking of Branched Deformable Linear Objects from Point Clouds for Robotic Manipulation* (PhD thesis). University of Stuttgart. DOI: 10.5281/zenodo.15665158 ### License This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0). You are free to use, modify, and distribute the data, provided appropriate credit is given.本数据集包含Markus Wnuk于2025年提交至斯图加特理工大学的博士论文《基于点云的分支可变形线性物体模型跟踪与机器人操作》（待提交）中所使用的评测数据。 其包含带标注的三维点云、运动学配置数据与元数据文件，用于对所提出的分支可变形线性物体（Branched Deformable Linear Objects, BDLO）跟踪方法开展定量与定性评估。本数据集旨在为可变形物体感知与机器人操作领域的可复现性研究、基准测试及后续科研工作提供支撑。 ### 数据集内容 背景：用于通过背景减除实现图像分割的场景背景RGB图像。 BDLO描述：描述每个BDLO样本的元数据，包括其图结构（`model.json`）与标签配置（用于真值参考的标签ID）。 相机参数：实验所用RGB-D相机的内、外标定数据。 评测数据集：按场景划分的文件夹，包含： - 元数据（如BDLO类型、相机设置、抓取位姿） - RGB-D数据（彩色图与视差图均为`.png`格式文件） 评测结果：经处理的评测结果，包括： - 紧凑格式结果文件（`.pkl`） - 可视化绘图（`.jpg`格式） ### 应用场景 本数据集可用于： - 基准测试BDLO跟踪与形状估计方法 - 开发面向柔性或分支物体机器人操作的感知流水线 - 训练并评测用于可变形物体定位的基于学习或基于模型的方法 ### 引用规范 若使用本数据集，请引用以下文献： > Markus Wnuk. (2025). *Model-based Tracking of Branched Deformable Linear Objects from Point Clouds for Robotic Manipulation* (博士学位论文). 斯图加特大学. DOI: 10.5281/zenodo.15665158 ### 授权协议 本数据集采用知识共享署名4.0国际许可协议（Creative Commons Attribution 4.0 International, CC BY 4.0）进行授权。您可自由使用、修改与分发本数据集，但需标注适当的引用来源。