TERRA-CD: A Benchmark Dataset for Semantic Change Detection
Published:
Technologies: PyTorch, Computer Vision
Description
- Dataset Creation: Created TERRA-CD, a benchmark dataset of 5,221 paired Sentinel-2 images across 232 cities with multi-level annotations for Change Detection.
- Model Development: Applied Computer Vision and Deep Learning techniques to design and evaluate multiple models (Siamese networks, STANet variants, HRSCD strategies), achieving 93.2% accuracy in semantic change detection.
- Recognition: Won Best Technical Implementation at College of Engineering Pune; research currently under review at a Q1 journal.