File size: 2,207 Bytes
d9eeb43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
# Processed Episodes Gestures Dataset

This dataset contains processed gesture episodes with multi-camera video data and metadata from the GigaHands dataset.

## Dataset Structure

The dataset contains 16 episodes (p005-gopro-034 through p005-gopro-049), each with:

- **Multi-camera video recordings**: Up to 51 cameras per episode in MP4 format
- **Camera calibration data**: Intrinsics and extrinsics matrices for each camera
- **Episode metadata**: Timing information, scene names, and technical parameters
- **Additional processed data**: Extras numpy arrays with supplementary information

### Directory Structure

```
processed-episodes-gestures/
├── p005-gopro-034/
│   ├── metadata.json          # Episode metadata and camera parameters
│   ├── extras.npy            # Additional processed data
│   ├── cam_0/
│   │   └── rgb/
│   │       └── video.mp4     # Camera 0 video recording
│   ├── cam_1/
│   │   └── rgb/
│   │       └── video.mp4     # Camera 1 video recording
│   └── ... (up to cam_50)
├── p005-gopro-035/
│   └── ... (same structure)
└── ... (through p005-gopro-049)
```

### Metadata Format

Each episode's `metadata.json` contains:
- `n_cams`: Number of cameras (typically 51)
- `n_steps`: Number of time steps in the episode
- `scene_name`: Identifier for the scene
- `video_format`: Format of video files (jpg/mp4)
- Camera-specific data for each camera:
  - `extrinsics`: 4x4 transformation matrix
  - `intrinsics`: 3x3 camera intrinsic matrix
  - `resolution`: [width, height] of the camera

## Usage

You can load this dataset using the Hugging Face datasets library:

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("LocalWorldModels/processed-episodes-gestures")

# Access episode data
for episode in dataset:
    print(f"Episode: {episode['episode_name']}")
    print(f"Cameras: {episode['n_cams']}")
    print(f"Steps: {episode['n_steps']}")
```

## Citation

If you use this dataset, please cite the original GigaHands work and this processed version.

## License

Please refer to the original GigaHands dataset license terms.