metadata

pretty_name: HO-Tracker Challenge
language:
  - en
license: mit
task_categories:
  - other
tags:
  - hand-object
  - 3d
  - python
size_categories:
  - n<10K

HO-Tracker Challenge — HANDS Workshop @ ICCV 2025

Dataset

Sample training data is provided in data/train_sample.

To browse the dataset locally:

# Step 1: Install dependencies
pip install open3d==0.18.0
pip install git+https://github.com/lixiny/manotorch.git

# Step 2: Download the MANO model from https://mano.is.tue.nl/downloads/
#         Place the extracted MANO assets under the `data/` directory
#         (e.g., `data/mano_v1_2`).

# Step 3: Launch the viewer
python vis_demo.py

Note (OakInk V2)

In OakInk V2, MANO parameters are obtained by fitting to SMPL-X meshes. As a result, hand keypoints computed from MANO may differ from the original OakInk V2 keypoints by a few millimeters. Please choose the set that best suits your use case.