README.md

---
AIGC:
    Label: "1"
    ContentProducer: 001191110102MACQD9K64018705
    ProduceID: 7625817874396938511-data_volume/files/所有对话/主对话/牛宿科技/tlabel-convert-dataset/README.md
    ReservedCode1: ""
    ContentPropagator: 001191110102MACQD9K64028705
    PropagateID: 3590332467129944#1786003455834
    ReservedCode2: ""
---
# TLabel Convert — One-Click Tactile Data Format Conversion

[中文](README_CN.md) | English

TLabel Convert is a CLI tool for converting tactile sensor data between formats. It provides a unified interface to transform data from 9 different tactile sensors into training-ready formats (LeRobot, FTP-1/MTTS).

## Quick Start

```bash
# Install
pip install tlabel>=0.19.0

# Convert a single file
tlabel convert --from gelsight --to lerobot --input data.pkl --output output_dir/

# Batch convert a directory
tlabel batch-convert --from univtac --to ftp1 --input-dir ./raw/ --output-dir ./converted/

# List all supported adapters
tlabel list-adapters

# Get detailed adapter info
tlabel adapter-info gelsight
```

## Supported Adapters

### Data Adapters (9 total)

| Adapter | Sensor Type | Input Format | Description |
|---------|-------------|--------------|-------------|
| `gelsight` | GelSight Mini / DIGIT | `.pkl`, `.pickle` | Vision-based tactile sensor |
| `paxini` | PaXini | `.h5`, `.hdf5` | High-resolution tactile array |
| `daimon` | Daimon DM-TacClaw | `.parquet` | LeRobot-compatible dataset |
| `tlabel` | Generic | `.json` | TLabel Format JSON |
| `touchd` | ToucHD-Force / AnyTouch 2 | `.npy` + directory | Multi-modal tactile data |
| `univtac` | UniVTAC | `.hdf5`, `.h5` | Cross-dataset format |
| `vtouch` | VTouch | `.h5`, `.hdf5` | Vision-based tactile |
| `ycb_slide` | YCB-Slide CMU DIGIT | `.npy` + directory | Sliding manipulation data |
| `tacquad` | TacQuad AnyTouch | `.csv` + directory | Multi-sensor array |

### Output Formats (2 total)

| Format | Description | Use Case |
|--------|-------------|----------|
| `lerobot` | [HuggingFace LeRobot](https://github.com/huggingface/lerobot) | VLA training, manipulation policies |
| `ftp1` | [FTP-1/MTTS Zarr](https://github.com/rail-berkeley/ftp1) | Foundation model pretraining |

## Example: Converting GelSight Data to LeRobot Format

```bash
# 1. Convert single file
tlabel convert \
  --from gelsight \
  --to lerobot \
  --input gelsight_data.pkl \
  --output lerobot_dataset/

# 2. Batch convert a directory
tlabel batch-convert \
  --from gelsight \
  --to lerobot \
  --input-dir ./raw_gelsight/ \
  --output-dir ./lerobot_datasets/
```

## What is TLabel?

[TLabel](https://github.com/liesliy/tlabel) is an open-source tactile data annotation standard. It provides:
- Unified schema for 14 semantic dimensions (contact, slip, force, etc.)
- 11 sensor/data adapters
- CLI tools for validation and conversion
- Visualization suite for tactile data

## Why Convert Formats?

Different robot learning frameworks expect different data formats:
- **LeRobot**: HuggingFace's robot learning framework (parquet + metadata)
- **FTP-1/MTTS**: Foundation model pretraining (Zarr format)

TLabel Convert bridges the gap between raw sensor data and training-ready formats, saving you hours of manual data wrangling.

## Documentation

- [TLabel Format Specification](https://github.com/liesliy/tlabel/blob/main/docs/tlabel-format.md)
- [Adapter Development Guide](https://github.com/liesliy/tlabel/blob/main/CONTRIBUTING.md)
- [Schema Design Document](https://github.com/liesliy/tlabel/blob/main/docs/TLabel_Design_Document.md)

## Citation

If you use TLabel in your research, please cite:

```bibtex
@software{tlabel2026,
  title={TLabel: A Unified Tactile Data Annotation Standard},
  author={Niusu Tech},
  year={2026},
  url={https://github.com/liesliy/tlabel}
}
```

## License

Apache-2.0

---