Colombian Land-Cover Semantic Segmentation Data
Description
CoL-Seg (Colombian Diverse Landscapes Semantic Segmentation Dataset) is a high-quality, large-scale semantic segmentation dataset specifically designed for high-resolution remote sensing (HRRS) images. Developed to address the critical scarcity of annotated datasets tailored to the unique geographical, topographical, and thematic diversity of tropical landscapes, CoL-Seg comprises images across 15 diverse Colombian municipalities. The dataset features distinct semantic classes essential for comprehensive land-cover analysis, urban growth monitoring, and environmental management. It was meticulously constructed using a robust semi-automated annotation pipeline that integrates iterative pseudo-labeling with strategic human-in-the-loop refinement, ensuring high pixel-level accuracy and consistency while optimizing annotation efficiency.CoL-Seg serves as a reliable benchmark and open training foundation for state-of-the-art deep learning segmentation models in remote sensing and geospatial data analysis.
Files
Institutions
- Universidad Nacional de ColombiaBogota D.C., Bogotá