Tohoku‑University‑6D‑Pose‑Estimation‑Dataset

Dataset Overview

This dataset is designed to evaluate pose‑estimation algorithms for industrial part picking tasks. The scenes are crafted to mimic real factory environments, aiming to assess the pose‑estimation performance of picking systems.

Scene Characteristics

• Each scene contains only one type of part, randomly stacked to simulate a realistic factory environment.
• More than ten parts are present in each scene.
• Parts are typically made of resin, lacking distinct colour or texture, and often share the same shape.

Dataset Structure

The dataset is split into simulated data and real‑world scene data. Currently, simulated data for three part types have been uploaded; additional parts and real‑world scene data are being prepared and are expected to be released on September 15.

Sensor Information

The dataset uses a 3D sensor named Ensenso n35 and a 2D camera named USB 3 uEye CP Rev. 2. The relative pose between the 3D sensor and the 2D camera has been calibrated, ensuring that the correspondence between each pixel in the image and each point in the point cloud can be obtained via a .csv file.

Authors

The dataset was created by Diyi Liu, Shogo Arai, Jiaqi Miao, Jun Kinugawa, and Kazuhiro Kosuge.

License

The project follows the MIT License; see the LICENSE.md file for details.