LEVIR-CD

LEVIR-CD is a large-scale remote sensing dataset designed for building change detection (CD), serving as a benchmark for deep learning-based CD algorithms. It provides high-resolution bitemporal satellite images to help researchers and developers analyze and detect structural changes over time.

Key Features of LEVIR-CD

• 637 Image Pairs: Each image pair is captured at different times, showcasing land-use changes over a span of 5 to 14 years.
• Very High Resolution (VHR): Each image patch has a 0.5m/pixel resolution with a 1024 × 1024 pixel size, ensuring detailed building change analysis.
• Building-Focused Changes: The dataset highlights structural modifications such as new constructions, demolitions, and expansions.
• Diverse Building Types: Includes residential villas, tall apartments, garages, warehouses, and other urban structures.
• Expert-Annotated Labels: All images are manually annotated with binary labels (1 for change, 0 for unchanged) by remote sensing experts for high-quality data integrity.
• 31,333 Change Instances: A vast number of annotated changes provide a rich source for training machine learning and deep learning models.

This dataset is sourced from Kaggle.