LoopDB: A Loop Closure Dataset for SLAM

Overview

LoopDB is a challenging loop closure dataset comprising over 1000 high-resolution images captured across diverse environments, including parks, indoor scenes, parking spaces, and individual objects. Each scene is represented by a sequence of five consecutive images, making it ideal for benchmarking loop closure algorithms used in Simultaneous Localization and Mapping (SLAM) systems.

Sample images from the LoopDB dataset showing various environments

Key Features

• Diverse Environments: Indoor (gyms, dorms, hotels) and outdoor (parks, streets, bus stations) scenes
• High Resolution: 48MP images captured with Huawei Nova 7i camera
• Ground Truth: Computed rotations and translations between consecutive images
• Comprehensive Metadata: Detailed information about each image and transformation
• Benchmarking Ready: Structured for easy evaluation of loop closure algorithms

Dataset Structure

LoopDB is organized into two main datastreams:

1. Datastream 1: Contains the captured scene images and basic metadata
2. Datastream 2: Contains transformation data between images

The metadata provides transformations (rotation and translation) between consecutive images.

Directory Structure

dataset/
├── datastream_1/
│   ├── samples/
│   │   └── left/        # Image files
│   └── data_descriptor.csv
└── datastream_2/
    └── data_descriptor.csv

Data Format

Each scene in LoopDB consists of 5 consecutive images with the following information:

• Root Image: First image in the sequence (reference frame)
• Transformations: Rotation and translation between consecutive images
• Metadata: Timestamp, sampling time, and other information

Getting Started

Installation

1. Clone this repository:

   git clone https://github.com/RovisLab/LoopDB loopdb
   cd loopdb

2. Install required packages:

   pip install -r requirements.txt

3. Download the dataset:

   # Create dataset directory
   mkdir -p data/datastream_1/samples/left
   mkdir -p data/datastream_2

   Download the dataset from Google Drive:

   • LoopDB Dataset (Google Drive)
   • Extract the downloaded archive into the left directory
   • Ensure the structure matches the expected dataset format (see Dataset Structure section)

4. Set up the dataset path:

   • Edit the DATASET_PATH variable in Api/server.py
   • Alternatively, specify the path as a command-line argument

Basic Usage

Exploring the Dataset via Api

python Api/server.py

Then open your browser to http://localhost:5000 to browse the dataset.

Command-Line Exploration

# Validate dataset integrity
python tools/visualization/cli_explorer.py validate

# View metadata for a specific image
python tools/visualization/cli_explorer.py view --metadata <timestamp>

# View an image
python tools/visualization/cli_explorer.py view --image <timestamp>

# Display image sequences grouped by root image
python tools/visualization/cli_explorer.py sequences

Calculate Transformations

python tools/processing/calculate_metadata/loopdb_calculation.py

Upload New scene

python tools/collection/upload_scene.py

Tools

This repository includes several tools for working with the LoopDB dataset:

Data Collection

• upload_images.py: GUI tool for uploading new scence to the dataset.

Data Processing

• update_metadata.py: Calculates transformations between images using SIFT/ORB
• find_matches.py: Finds keypoint matches between frames for single scence using

Visualization

• cli_explorer.py: Command-line interface for dataset exploration .
• server.py: API access for browsing the dataset .

Research and Citation

Contributing

We welcome contributions to improve the dataset and associated tools. Please feel free to submit pull requests or open issues for discussion.

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0) - see the LICENSE file for details.

Acknowledgements

The authors are with the Robotics, Vision and Control Laboratory (RovisLab, https://www.rovislab.com), Transilvania University of Brasov, Romania.