HoyerChou/EgocentricVideos

Egocentric Bimanual Hand Videos

1) Raw video of three bimanual dexterous manipulations tasks

• 2026-01-05_19-35-47-pick-place.mp4 (3 m 56 s)
• 2026-01-05_19-52-29-reorient-place.mp4 (3 m 15 s)
• 2026-01-05_19-57-25-bimanual-place.mp4 (3 m 15 s)

2) Segmented short clips for each task (unlabeled demonstrations)

• 2026-01-05_19-35-47-pick-place.zip (39 demos)
• 2026-01-05_19-52-29-reorient-place.zip (40 demos)
• 2026-01-05_19-57-25-bimanual-place.zip (40 demos)

3) Visualization, snapshots and teasers for final demonstrations

4) Extracted action labels for hand mesh / trajectory / keypoints

• 2026-01-05_19-35-47-pick-place_resLabels.zip (111 MB)
• 2026-01-05_19-52-29-reorient-place_resLabels.zip (102 MB)
• 2026-01-05_19-57-25-bimanual-place_resLabels.zip (105 MB)

The json label formats

saved_labels_dict = {
  "fps": 30,  # video FPS
  "height": height,  # video height, 1080
  "width": width,  # video width, 1920
  "frame_num": vis_end - vis_start + 1,  # video frames T (which is about 50 ~ 70)
  "camera_K": camK.tolist(),  # a 3x3 matrix for camera frame projection
  "left_hand_vertices": left_dict['vertices'].numpy()[0].tolist(),  # (T, N, 3), N = 778 for MANO 3D mesh
  "right_hand_vertices": right_dict['vertices'].numpy()[0].tolist(),  # (T, N, 3), N = 778 for MANO 3D mesh
  "left_hand_keypoints": np.array(kpts_trans_L).tolist(),  # (T, N, 3), N = 21 for MANO keypoints
  "right_hand_keypoints": np.array(kpts_trans_R).tolist(),  # (T, N, 3), N = 21 for MANO keypoints
  "rotation_w2c": np.array(R_w2c).tolist(),  # (T, 3, 3), rotation matrix for world frame to camera frame
  "translation_w2c": np.array(t_w2c).tolist(),  # (T, 3), translation vector for world frame to camera frame
  "rotation_c2w": np.array(R_c2w).tolist(),  # (T, 3, 3), rotation matrix for camera frame to world frame
  "translation_c2w": np.array(t_c2w).tolist(),  # (T, 3), translation vector for camera frame to world frame
  "global_orient_L": np.array(wrist_L_Rot_lst).tolist(),  # (T, 3, 3), for hand retargeting [2026.01.13]
  "global_orient_R": np.array(wrist_R_Rot_lst).tolist(),  # (T, 3, 3), for hand retargeting [2026.01.13]
}

Citation

If you use our data in your research, please cite with:

@article{zhou2025you,
  title={You Only Teach Once: Learn One-Shot Bimanual Robotic Manipulation from Video Demonstrations},
  author={Zhou, Huayi and Wang, Ruixiang and Tai, Yunxin and Deng, Yueci and Liu, Guiliang and Jia, Kui},
  journal={arXiv preprint arXiv:2501.14208},
  year={2025}
}

@article{zhou2026yoto,
  title={YOTO++: Learning Long-Horizon Closed-Loop Bimanual Manipulation from One-Shot Human Video Demonstrations},
  author={Zhou, Huayi and Wang, Ruixiang and Tai, Yunxin and Deng, Yueci and Liu, Guiliang and Jia, Kui},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2026},
  publisher={IEEE}
}