tpoellabauer/YCB-V-DS

We provide our stereo recordings for the YCB-V object dataset, as well as some of our physically-based stereo renderings.

[image]

Contents

• —stereo+depth recordings/ — raw stereo video recordings (.mkv)
• —train_pbr_left*.z*, train_pbr_right*.z* — physically-based rendered training data (split zip archives), left/right stereo
• —test_pbr_left.zip — physically-based rendered test data
• —test_bboxes/ — BOP-format annotation JSONs:
• —test_targets_pbr_left.json — flat list of {im_id, inst_count, obj_id, scene_id} targets (shown in the Dataset Viewer above)
• —test_pbr_left.json, test_pbr_right.json, test_pbr_stereo.json — per-scene/image estimated bounding boxes (bbox_est, plus bbox_est_left/bbox_est_right for the stereo file), keyed by "<scene_id>/<im_id>". Nested/heterogeneous structure — not rendered by the Viewer, download directly.
• —camera.json — camera intrinsics

Related datasets

• —tpoellabauer/YCB-V-LUMA — chroma/luma-key recordings of the same objects
• —tpoellabauer/YCB-V-RAND — randomized-texture/lighting renderings
• —tpoellabauer/YCB-LUMA — luminance-keyed localization follow-up

Paper: arxiv.org/abs/2402.05610

If you find this dataset useful, please cite

@article{pollabauer2024extending,
  title={Extending 6D Object Pose Estimators for Stereo Vision},
  author={Pöllabauer, Thomas and Emrich, Jan and Knauthe, Volker and Kuijper, Arjan},
  journal={arXiv preprint arXiv:2402.05610},
  year={2024}
}