EventHO3D / README.md
mickeykang's picture
update README
2632f6a verified
metadata
license: cc-by-nc-4.0
task_categories:
  - object-detection
tags:
  - 6d-pose
  - event-camera
  - ho3d
  - novel-object
size_categories:
  - 1G<n<10G

EventHO3D (Event-Augmented Add-On for HO3D-v2)

Event streams and XMem object masks that augment the HO3D-v2 evaluation split for Event6D (CVPR 2026).

Events are simulated from the original RGB frames (no real event camera). This dataset does NOT redistribute the HO3D images/depth — download those from the official source and merge with this release.

Sequences

13 video runs across 4 novel YCB objects: AP10–AP14 (pitcher), MPM10–MPM14 (potted meat can), SB11, SB13 (bleach), SM1 (mustard).

Layout (this release)

EventHO3D/
├── evaluation_events/<seq>/   # simulated event streams (.npz per frame, ≈3 GB)
└── masks_XMem/<seq>/          # XMem object segmentation masks (≈80 MB)

Setup with original HO3D-v2

  1. Download HO3D-v2 evaluation from the official project page: https://www.tugraz.at/index.php?id=40231

  2. Place this release under the same parent so that evaluation/ (from HO3D-v2) and evaluation_events/ / masks_XMem/ (from here) sit side-by-side:

    EventHO3D/
    ├── evaluation/             # ← from official HO3D-v2 (rgb, depth, meta, hand_mask, ...)
    ├── evaluation_events/      # ← from this release
    ├── masks_XMem/             # ← from this release
    └── ycb_models/             # ← YCB-Video models, see below
    
  3. YCB-Video models (ycb_models/{ob_name}/textured_simple.obj) are required by Ho3dReader.get_gt_mesh(). Download from https://rse-lab.cs.washington.edu/projects/posecnn/ (YCB_Video_Models.zip), or set YCB_MODELS_PATH=/path/to/ycb_models to a copy you already have.

Download

huggingface-cli download mickeykang/EventHO3D --repo-type dataset \
    --local-dir ./data/EventHO3D

Event file timing

Each evaluation_events/<seq>/XXXXXX.npz holds the events that occurred during the inter-frame interval ending at HO3D RGB frame XXXXXX (between rgb/(XXXXXX-1).jpg and rgb/XXXXXX.jpg, ≈ 1/30 s). Inside each .npz, key data is a structured array with fields (x, y, t, p) — event pixel coords, timestamp in seconds, polarity ±1.

Citation

@inproceedings{kang2026event6d,
  title     = {Event6D: Event-based Novel Object 6D Pose Tracking},
  author    = {Kang, Jae-Young and
               Cho, Hoonehee and
               Lee, Taeyeop and
               Kang, Minjun and
               Wen, Bowen and
               Kim, Youngho and
               Yoon, Kuk-Jin},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026}
}

@inproceedings{hampali2020honnotate,
  title     = {HOnnotate: A method for 3D Annotation of Hand and Object Poses},
  author    = {Hampali, Shreyas and Rad, Mahdi and Oberweger, Markus and Lepetit, Vincent},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2020}
}