XYZ-IBD

Introduction

The XYZ Industrial Bin Picking Dataset (XYZ-IBD) is a bin-picking dataset for 6D pose estimation and depth estimation, capturing real-world industrial-grade complexity through challenging geometries, materials, occlusions and clutter. The dataset represents authentic robotic grasping scenarios with millimeter-accurate annotations.

Objects: 15 highly-reflective industrial parts with various shapes and sizes (diameters ranging 5cm~30cm)
Scenes: 75 scenes, 22k RGB/grayscale/depth imagesand 91k milimeter-level 6D pose annotations
Instances: 24 instances per scene on average, with up to 60 instances in some scenes
Sensors: XYZ Robotics DLP structured light, Photoneo PhoXi laser scanner, RealSense D415 stereo camera
Synthetic Data: Realistic bin-picking scene simulation, providing 50k+ synthetic training images

Benchmarks & Challenges

BOP Industrial Track

The XYZ-IBD dataset is part of the Benchmark for 6D Object Pose Estimation (BOP) and serves as one of the official BOP-Industrial datasets for the BOP Challenge 2025. The XYZ-IBD dataset is evaluated on model-based 2D and 6D object detection tasks, following the BOP Challenge 2025 metrics and is hosted on the official BOP-Industrial Leaderboard. The challenge will also be featured at the ICCV 2025 R6D Workshop.

TRICKY Monocular Depth Track

The XYZ-IBD dataset is also associated with the Transparent & Reflective Objects In the Wild Challenges (TRICKY) for the Monocular Depth Estimation task. Evaluation on XYZ-IBD for this task follows the TRICKY 2025 metrics and is hosted on the official TRICKY 2025 Leaderboard, which will be featured as part of the TRICKY Workshop at ICCV 2025.

Download

Object 6D Pose Estimation

The pose estimation dataset is available for download at the following links. You can also find the dataset in both the BOP website or HuggingFace.

PBR-BlenderProc BOP format training images [Part 1] [Part 2]
Validation Data
Testing Data
Object Models

Depth Estimation

The depth estimation dataset is available for download at the following links. You can also find the dataset in the TRICKY website.

Real & Synthetic Training Data

Gallery

Scene Examples

Real industrial parts

CAD models of the industrial parts

Hardware Setup

Citation & License

XYZ-IBD is licensed under the CC BY-NC-SA 4.0 license. Custom license available upon request, if you are interested in using the dataset for commercial purposes, please contact us.

If you find this dataset helpful, please cite the following paper:

@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}, }

Contact

For any questions or issues regarding the dataset, please contact:
junwen.huang@tum.de
peter.yu@xyzrobotics.com