HouseCat6D is a multi-modal category level 6D object pose and grasping dataset with highly diverse household object categories of different photometric complexity and a high number of varying scenes covering large viewpoint distributions. It comprises room-scale high-quality camera trajectories and object poses without markers in realistic scenarios including occlusions as well as dense grasping pose annotation. Data includes synchronized RGB, depth from active stereo, and polarimetric RGB+P images in scenes comprising objects without texture, strong reflections, or translucency.

ARTTRACK2 tracking system and sets of infrared marker bodies we used for our setup. Once at least four infrared spheres are detected from at least two cameras, the tracking system provides the pose of the marker body as transformation from tracker system base to marker body base.

(a): Pre-scanning 3D models. (b): Pivot calibration to calibrate measurement tip from the tracking body. (c): Pose annotation of objects using measurement tip. (d): Hand-Eye-Calibration to calibrate camera center of tracking body. (e): Camera trajectory recording (f): Post-processing step to reduce synchronization-induced trajectory error.

@inproceedings{jung2024housecat6d,

  title={HouseCat6D-A Large-Scale Multi-Modal Category Level 6D Object Perception Dataset with Household Objects in Realistic Scenarios},

  author={Jung, HyunJun and Wu, Shun-Cheng and Ruhkamp, Patrick and Zhai, Guangyao and Schieber, Hannah and Rizzoli, Giulia and Wang, Pengyuan and Zhao, Hongcheng and Garattoni, Lorenzo and Meier, Sven and Roth, Daniel and Navab, Nassir and Busam, Benjamin},

  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},

  pages={22498--22508},

  year={2024}

}

}

journal={arX preprint arXiv:2309.09