lerobot-surgery 0.1.0

Dataset surgery utils (e.g., merge, episode remove) for LeRobotDataset v3

🔬 lerobot-surgery

⚠️ IMPORTANT: LeRobotDataset v3.0 Only It may work with future v3.x versions, but the code has only been tested on v3.0 so far. For v2.1 datasets, migrate to v3.0 first.

Dataset surgery utils (e.g., merge, episode remove) for LeRobotDataset v3.

---

✨ Features

• 🔀 Merge multiple datasets seamlessly into a single dataset
• ✂️ Remove specific episodes with automatic re-indexing
• 🖥️ CLI & Python API for flexible usage

---

📋 Requirements

• Python: 3.8 or higher
• LeRobot: LeRobotDataset version 3.0
• Dependencies: Automatically installed with the package

---

🚀 Installation

From PyPI

pip install lerobot-surgery

From source

git clone https://github.com/spiglerg/lerobot-surgery.git
cd lerobot-surgery
pip install -e .

Development installation

git clone https://github.com/spiglerg/lerobot-surgery.git
cd lerobot-surgery
pip install -e ".[dev]"

Verify installation:

lerobot-surgery --version
# Output: lerobot-surgery, version 0.1.0

lerobot-surgery --help
# Shows CLI help

---

📖 Quick Start

Command Line Interface

# Merge multiple datasets
lerobot-surgery merge dataset_a/ dataset_b/ dataset_c/ -o merged_dataset/

# Remove specific episodes (by index)
lerobot-surgery remove my_dataset/ 0 5 10 -o filtered_dataset/

# Display dataset information
lerobot-surgery info my_dataset/

# Show help
lerobot-surgery --help

Python API

from lerobot_surgery import merge_datasets, remove_episodes

# Merge datasets
merged = merge_datasets(
    source_paths=["dataset_a/", "dataset_b/", "dataset_c/"],
    output_path="merged_dataset/",
)
print(f"Merged {merged.num_episodes} episodes, {merged.num_frames} frames")

# Remove episodes
filtered = remove_episodes(
    dataset_path="my_dataset/",
    episode_indices=[0, 5, 10],  # Episodes to remove
    output_path="filtered_dataset/",
)
print(f"Remaining: {filtered.num_episodes} episodes")

---

📚 Detailed Usage

Merging Datasets

Combine multiple LeRobot datasets with identical structures (features, fps) into a single consolidated dataset.

Requirements:

• All datasets must have the same FPS
• All datasets must have identical features
• Datasets must be in v3.0 format

Episode Re-indexing: Episodes are automatically re-indexed sequentially starting from 0 in the merged dataset.

Removing Episodes

Remove specific episodes from a dataset while maintaining data integrity and re-indexing remaining episodes.

Episode Re-indexing: Remaining episodes are automatically re-indexed sequentially starting from 0.

🧪 Testing

Run the test suite:

# Install development dependencies
pip install -e ".[dev]"

# Run all tests
pytest

# Run with coverage
pytest --cov=lerobot_surgery --cov-report=html

# Run specific test file
pytest tests/test_merge.py -v

🙏 Acknowledgments

Built for the LeRobot framework.