XYZ-IBD
收藏资源简介:
XYZ-IBD是一个针对工业6D姿态估计的高精度拣选数据集,捕捉真实世界工业复杂性,包括具有挑战性的物体几何形状、反光材料、严重的遮挡和密集的杂乱。数据集反映了真实的机器人操作场景,具有毫米级精确的注释。该数据集包含15种无纹理、金属和大多对称的物体,这些物体形状和大小各异。这些物体被严重遮挡,并在高密度的情况下随机排列在容器中,复制了真实世界拣选的挑战。XYZ-IBD使用两个高精度工业相机和一个商用相机收集,提供RGB、灰度和深度图像。它包含75个多视图真实世界场景,以及一个大规模的合成数据集,在模拟拣选条件下渲染。我们采用细致的注释流程,包括防反光喷剂、多视图深度融合和半自动注释,实现了工业操作所需的毫米级姿态标注精度。
XYZ-IBD is a high-precision picking dataset for industrial 6D pose estimation, which captures the complexity of real-world industrial environments, including challenging object geometries, reflective materials, severe occlusions and dense clutter. The dataset reflects realistic robotic manipulation scenarios with millimeter-precise annotations. It contains 15 textureless, metallic and mostly symmetric objects with varying shapes and sizes. These objects are severely occluded and randomly arranged in containers at high density, replicating the challenges of real-world bin picking. XYZ-IBD was collected using two high-precision industrial cameras and a commercial camera, providing RGB, grayscale and depth images. It includes 75 multi-view real-world scenes, as well as a large-scale synthetic dataset rendered under simulated picking conditions. We adopted a meticulous annotation pipeline, including anti-reflective spray, multi-view depth fusion and semi-automatic annotation, to achieve the millimeter-level pose annotation accuracy required for industrial operations.
XYZ-IBD 数据集概述
基本信息
- 全称: XYZ Industrial Bin Picking Dataset (XYZ-IBD)
- 主要用途: 用于物体6D姿态估计和单目深度估计的高精度工业数据集
- 特点: 捕捉真实工业级复杂场景,包含毫米级精确标注
数据集内容
- 物体:
- 15个高反光工业零件
- 尺寸范围: 直径5cm~30cm
- 多种形状和尺寸
- 场景:
- 75个场景
- 22,000张RGB/灰度/深度图像
- 91,000个毫米级6D姿态标注
- 实例:
- 平均每场景24个实例
- 部分场景最多60个实例
- 传感器:
- XYZ Robotics DLP结构光
- Photoneo PhoXi激光扫描仪
- RealSense D415立体相机
- 合成数据:
- 真实感箱拣场景模拟
- 50,000+张合成训练图像
基准测试与挑战
BOP工业赛道
- 用途: 6D物体姿态估计基准测试
- 关联活动:
- BOP Challenge 2025 Industrial Track
- ICCV 2025 R6D Workshop
TRICKY单目深度赛道
- 用途: 单目深度估计挑战
- 关联活动:
- TRICKY Challenge 2025
- ICCV 2025 TRICKY Workshop
下载资源
6D姿态估计数据
- PBR-BlenderProc BOP格式训练图像 [Part 1] [Part 2]
- 验证数据
- 测试数据
- 物体模型
深度估计数据
- 真实与合成训练数据
许可与引用
- 许可证: CC BY-NC-SA 4.0 (可申请商业用途定制许可)
- 引用文献: bibtex @misc{huang2025xyzibdhighprecisionbinpickingdataset, title={XYZ-IBD: High-precision Bin-picking Dataset for Object 6D Pose Estimation Capturing Real-world Industrial Complexity}, author={Junwen Huang and Jizhong Liang and Jiaqi Hu and Martin Sundermeyer and Peter KT Yu and Nassir Navab and Benjamin Busam}, year={2025}, eprint={2506.00599}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2506.00599}, }
联系方式
- junwen.huang@tum.de
- peter.yu@xyzrobotics.com




