README.md

---
license: cc-by-4.0
---

# FusionX Multimodal Sample Dataset

This repository contains a sample multimodal dataset including:

- RGB camera frames (30 Hz, JPG)
- Depth images (16-bit PNG in millimeter)
- Tactile glove data (aligned per-frame in Parquet)
- Per-recording metadata

## Structure

To reduce API rate limits and improve upload stability, each recording is stored as a ZIP archive:

Each ZIP file contains a single recording directory structured as follows:

```
recording_YYYYMMDD_HHMMSS/
├── frames.parquet          # Per-frame aligned metadata + tactile arrays
├── rgb/                    # 30 Hz RGB camera frames (JPEG)
│   ├── 000000.jpg
│   ├── 000001.jpg
│   └── ...
└── depth/                  # 16-bit depth images in millimeters (PNG)
    ├── 000000.png
    ├── 000001.png
    └── ...
```

## How to Use

1. Download the repository (or use `huggingface_hub.snapshot_download`)
2. Unzip all files inside `zips/`

## After Extraction

After unzipping all archives inside the `zips/` directory, you will obtain the original dataset structure:

```
dataset/
├── dataset_info.json
└── recording_YYYYMMDD_HHMMSS/
    ├── frames.parquet
    ├── rgb/
    │   ├── 000000.jpg
    │   └── ...
    └── depth/
        ├── 000000.png
        └── ...
```

Each `recording_*` directory represents a single synchronized capture session containing:
- Vision data (RGB + depth)
- Tactile glove data
- Frame-level alignment metadata

## License

Creative Commons Attribution 4.0 (CC BY 4.0)

You are free to:
- Use
- Share
- Modify
- Use commercially

As long as attribution is provided.